Fact-Check Dashboard

Start here

The documentary mixes well-supported history with compressed, overstated, or unproved conclusions. The entries below give the current verdict and link directly to the full sourced analysis. The classifications apply to the documentary’s wording and inference, not to whether the underlying history is worth investigating.

False or materially inaccurate

• 12:48 — “Standard deviation, correlation, the logit model, and logistic regression emerged specifically out of eugenics” — False as a bundled origin claim. Galtonian biometry and eugenics were tightly connected, but the logistic function predates Galtonian eugenics, while the modern logit and logistic-regression lineages are later.
• 18:44 — A universal Turing machine can “calculate everything within the mathematical universe” — False. Universality covers every computable function, not every mathematical problem; some problems are undecidable.
• 30:48–31:21 — Computers would match human “narrow reasoning” by 1990 and become a new intelligent race — The 1982 prediction was undefined and failed under its natural broad reading. Robert Jastrow’s hardware-growth intuition resembled Moore’s law, but transistor growth is not a measure of intelligence; a computer did not defeat the reigning world chess champion until 1997. “Race” here means a projected machine successor, not a demonstrated reference to human racial hierarchy.
• 1:10:49 — “AI is just a repackaging of our data” — Misleading if read as a technical description. Generative models are trained on human-produced data, but they learn statistical parameters and generate new sequences; they are not ordinarily databases that simply retrieve and rewrap stored records. Verbatim memorization and extraction nevertheless occur, so the underlying concern about appropriation, provenance, and privacy is real.

Overreaching or not established

• “AI has roots in eugenics,” therefore AI is eugenic in character — Does not follow from ancestry alone. The claim needs a surviving aim, category, institution, mechanism, or measurable effect.
• Is the G in AGI traceable to psychometric g? — Conceptually related, but not a demonstrated direct etymology or causal lineage. Foundational AGI writers explicitly called the relationship “loose”; the modern term was chosen mainly to distinguish broad-scope, transferable intelligence from narrow AI.
• 53:47–54:15 — Legg cited Gottfredson, therefore AGI inherited her racial theory of intelligence — The citation is real; the stronger inheritance claim is not established. Legg and Hutter cited her broad definition of intelligence while constructing a machine-performance measure based on goal achievement across environments. They did not cite her race-and-IQ claims as a premise of that measure.
• 19:52–21:34 — Pearson → the Ryle family → Gilbert Ryle → Turing → AI → “white flight from the body” — Historical associations are real, but the proposed transmission is not established. This is the full claim-by-claim fact-check and logical analysis.
• 20:59 — Full contextual quote and explanation of “white flight from the body” — A metaphor for racial power migrating from explicit biological classifications into apparently neutral behavioral, institutional, and computational procedures. Includes the complete 19:52–21:34 exchange and shows, step by step, where Zehner thinks “whiteness” can enter a computerized system. His other writing clarifies the mechanism, but not the documentary’s Ryle–Turing genealogy.
• 20:02 — John Ryle was vice-president of the Eugenics Society — Exact title not verified. Reliable sources document him as a council member, sterilization-committee member, and Galton lecturer.
• 29:45–30:51 — Male technological “birth” created a patrilineal Silicon Valley that does not need women — Recurring reproductive rhetoric and gendered exclusion are documented; the collective psychological motive is not. The passage supports a cultural interpretation, not proof that male founders built technology to appropriate women’s reproductive power.
• 1:08:04 — AI firms intentionally “prey on” single mothers, lower-caste people, and refugees — The labor pattern is documented; the universal intentional-exploitation claim is not. Outsourced data work in Argentina and India, recruitment among poor or marginalized communities, and annotation programs in refugee settings are real. Some programs explicitly present this as social inclusion. The film does not name the relevant firms or programs, establish their clients, or show that harmful targeting—rather than low-cost labor, language, inclusion policy, or several motives together—was the purpose in every case.
• 1:13:25 — “Data centers just use fresh water” and a new Kenyan facility will compete with residents — Blanket claim false; local risk unresolved. Cooling designs differ. The likely facility shown, iXAfrica NBOX1, does use utility water backed by a treated borehole and indirect adiabatic cooling, but publishes no water-use total. Another Nairobi operator says it uses no water for cooling. No evidence located establishes that a particular Kenyan center has already reduced household supply.

Supported with important qualification

• Was quantitative intelligence invented by eugenics? — Deeply entangled, not exclusively invented by it. Galton joined measurement to eugenic selection; Binet developed testing for a different educational purpose.
• Did American sterilization policy influence Nazi Germany? — Yes, materially—but “Hitler copied America” is too simple. American precedents mattered alongside Germany’s own racial-hygiene movement and the Nazis’ radicalization of it.
• 31:45–32:09 and 53:47–54:15 — Pioneer Fund and Linda Gottfredson — Substantially supported, with a needed distinction between fact and interpretation. The fund’s eugenic charter, founders, and long support for hereditarian race-and-IQ work are documented; Gottfredson received its money and later defended a partly genetic account of the U.S. Black–White test-score gap. “Kept race science alive” and “explicitly racist” are interpretive labels, not neutral descriptions of each publication.
• The documentary's broader five-part argument — Historical and methodological overlap is strong; conceptual, institutional, and identity claims require increasingly specific evidence.
• 27:25–27:57 — Silicon Valley's “special genius class of men” mythology — Substantially supported as an influential founder myth, not as the single essence of a diverse industry. Historians document lone-inventor and heroic-founder stories that hide collective labor; Lewis’s claims about unlimited wealth, power, and immunity from criticism are interpretive generalizations.
• 1:03:39–1:07:54 — OpenAI outsourced toxic-content labeling to Kenyan workers who encountered sexual material and reported psychological harm — Substantially supported. Contracts, payslips, OpenAI’s confirmation, and multiple worker interviews support the core account. The work helped build a toxicity detector used in ChatGPT; it was one safety component, not “a lot of the development of ChatGPT.” The best-documented take-home range was roughly $1.32–$2 per hour, not the film’s “less than a dollar.” The harms are credible worker reports and consistent with broader moderator research, but the film does not present clinical assessments of these particular workers.
• Kenyan data-center construction and water stress — Yes, expansion is real; exact water consumption is not public. NBOX1 is operating, Nxtra broke ground on a 44 MW GPU-ready facility in 2025, and other facilities are expanding. The much-publicized Microsoft–G42 Olkaria project stalled in 2026. Kenya is water-scarce and Nairobi has a documented supply deficit and neighborhood inequality, but “Nairobi has no fresh water” would be false: 99% had at least basic drinking-water service in the 2022 survey, although that measure does not mean continuous household piping.

Historically well supported

• Galton, Pearson, Fisher, eugenics, and statistics — Their eugenic commitments and their major roles in the history of statistics are well documented.
• The Pioneer Fund's eugenic origin and postwar funding network — Its 1937 charter expressly named heredity, eugenics, and “race betterment”; archival histories trace its founders’ racial-hygiene commitments and later financing of prominent hereditarian researchers.
• Statistical methods later migrated into medicine and other non-eugenic fields — This is why shared ancestry cannot settle the moral identity of every later application.
• What evidence would actually demonstrate eugenic continuity in an AI system? — A practical test based on aims, inherited categories, data, institutions, deployments, and measurable harms.

Context and counterarguments

• Reconstructed Cofnas–Murray response to Saini — A labeled reconstruction of the strongest population-structure objection, not a quotation or endorsement.
• What within-population genetic variation can and cannot establish — Separates ancestry classification from claims about the causes of behavioral group differences.
• Concise ancestry-versus-inheritance counterargument — Includes the clinical-research, weather-forecasting, polling, and sports-analytics comparisons.

---

Fact-check: hidden labor, “repackaged data,” and Kenyan water (1:03:02–1:13:47)

Bottom line

The documentary’s strongest claim is the Kenyan labor story: OpenAI contracted Sama to have Nairobi workers label descriptions of sexual abuse, violence, and hate speech for a toxicity-detection system used in ChatGPT, and workers reported intrusive memories, insomnia, and other serious distress. Its weakest formulations are the absolutes. The documented pay was generally above—not below—one dollar per hour; the evidence does not show that every Argentina-, India-, or refugee-focused program was designed to exploit; a language model does more than “repackage” stored data; and Kenyan data centers do not all use the same cooling system. Kenya’s water scarcity and Nairobi’s unequal, intermittent service are real, but the water consumption of the likely facility shown in the film has not been publicly quantified.

Claim-by-claim verdict

OpenAI, Sama, and Kenyan data workers (1:03:39–1:07:54)

What is independently documented

The central story is real. A 2023 TIME investigation reviewed contracts, payslips, and internal material and interviewed four Sama employees. It found that:

• OpenAI signed three contracts worth about $200,000 in total with Sama in late 2021. OpenAI later said that the actual partnership was worth about $150,000 because it ended early.
• Roughly three dozen Nairobi workers were assigned to label textual descriptions of sexual abuse, violence, and hate speech. OpenAI confirmed that Sama employees contributed to a toxic-content detector that was eventually incorporated into ChatGPT.
• This was not ordinary pretraining or the whole of “developing ChatGPT.” The labels helped train a separate safety/classification tool used to recognize toxic material in model inputs or outputs.
• TIME calculated take-home wages of approximately $1.32–$2 per hour depending on role, seniority, and performance. Sama disputed aspects of the report and said the after-tax range was $1.46–$3.74 per hour. OpenAI paid Sama a vendor rate of about $12.50 per worker-hour, but that figure covered management, infrastructure, benefits, quality assurance, and Sama’s margin; it was not a promised direct wage.
• A separate OpenAI pilot asked Sama to collect and label sexual and violent images, including categories described in the documents as child sexual abuse, bestiality, rape, sexual slavery, death, and serious injury. Sama ended the OpenAI work early in 2022.

Sources: TIME, “OpenAI Used Kenyan Workers on Less Than $2 Per Hour” · TIME profile of Richard Mathenge · CBS/60 Minutes follow-up

Sexual content and psychological harm

The workers’ exposure is not merely an allegation inferred from seeing an unpleasant job title; the reviewed contracts specified sexual-abuse, violence, and hate-speech categories. All four workers interviewed by TIME said the work had mentally scarred them. One described recurring images after reading a graphic sexual-abuse passage. The documentary’s Richard Mathenge and Mophat Okinyi give named first-person accounts of insomnia, screaming at night, paranoia, and persistent intrusive material.

The appropriate evidentiary qualification is that these are credible and multiply reported worker accounts, not published psychiatric examinations establishing a diagnosis and a unique causal pathway for each individual. Sama said wellness support existed and disputed that workers had access only to group sessions; workers told TIME that counseling was rare, unhelpful, or difficult to obtain under productivity pressure. OpenAI said Sama, as the employer, managed pay and mental-health provisions.

Independent research makes the claimed mechanism plausible. A 2024 cross-sectional study of commercial content moderators found a dose-response relationship between frequency of exposure to distressing content and psychological distress and secondary-trauma symptoms. Because it was observational and not a randomized experiment, it strengthens but does not by itself prove causation in every worker. PubMed record for Spence, DeMarco, and Martellozzo

Precise conclusion

Verified: OpenAI knowingly commissioned toxic-content labeling; Kenyan Sama workers were exposed to explicit sexual and violent descriptions; several reported serious psychological effects; and their pay was low by U.S. standards. Not verified: the film’s exact “less than a dollar” figure, that OpenAI itself set Sama’s quotas or wages, or that the documentary’s symptom accounts are independently confirmed clinical diagnoses.

Argentina, India, and refugee camps: a real pattern stated too categorically (1:08:04)

Milagros Miceli says that companies “prey on specifically vulnerable populations,” citing Buenos Aires, Indian programs targeting single mothers or lower-caste people, and recruitment in refugee camps. Several components of this account are supported:

• Miceli and Julian Posada studied outsourced machine-learning work at three Venezuelan platforms and an Argentine business-process outsourcer using 210 instruction documents, 55 interviews, and participant observation. They found economic dependency, precarious conditions, performance control, and limited worker agency. “The Data-Production Dispositif”
• A second Argentina/Bulgaria study drew on 2.5 years of research, 33 interviews, and five co-design workshops, confirming that outsourced data production is real, continuous human work whose conditions are often hidden from downstream users. “Documenting Data Production Processes”
• The ILO’s 2024 Mind the AI Divide report found that much data-labeling work has moved to developing economies because remuneration is low. It reported that Indian platform annotators earned about two-thirds less than comparable non-platform employees, even before accounting for employment benefits. ILO report
• Recruitment aimed at disadvantaged groups is not invented. Sama explicitly describes its “impact sourcing” model as opening formal employment to low-income and historically marginalized people. India’s Karya has deliberately recruited lower-caste women, while annotation and digital-work programs have also operated in refugee settings.

What the documentary does not establish is that each named group was recruited for the purpose of harming or trapping it. “Impact sourcing” can produce both genuine income opportunities and unequal bargaining power. Karya, for example, presents targeted recruitment as redistributive inclusion and has used a different compensation and community-ownership model. To determine whether a program is predatory requires its pay relative to alternatives and living costs, contract stability, informed consent, exposure risks, grievance procedures, client margins, and workers’ ability to refuse tasks—not the recruits’ identity alone. Sama’s description of its impact model · Digital Futures Lab discussion of India’s data-annotation sector and Karya

The transcript also does not say that the Argentine and Indian workers discussed there all reviewed sexual content. That evidence is specific to the Kenyan OpenAI/Sama passage. Separate reporting has found Indian moderators exposed to abusive or sexual material, but it does not identify the unnamed Indian program in Miceli’s documentary claim or connect it to OpenAI.

Is AI “just a repackaging of our data”? (1:10:49)

Why the phrase has a defensible core

Modern foundation models depend on very large bodies of human-produced material. OpenAI says its models use public internet data, licensed or partnered data, and material supplied or generated by users, human trainers, and researchers. Meta says the released Llama 3 models were pretrained on more than 15 trillion tokens from publicly available sources, followed by human-annotated post-training. Data workers also label, rank, filter, and test model behavior. In that political-economic sense, the film is right that “automation” can conceal human authorship, culture, judgment, and labor. OpenAI’s training-data explanation · Meta’s Llama 3 description

Why “just repackaging” is technically misleading

A conventional language model is not a searchable archive that keeps a copy of every training document and selects one to rephrase. During training, optimization adjusts billions of numerical parameters so that the model becomes better at predicting tokens from context. At generation time, it computes a probability distribution and produces a new token sequence. The resulting system can generalize, combine patterns from many examples, follow novel instructions, translate unfamiliar combinations, and sometimes produce material that does not occur verbatim in the training set. Calling all of that “just repackaging” erases the learned transformation and genuine generalization.

That correction should not be turned into the opposite myth. Neural language models can memorize. Researchers extracted hundreds of verbatim sequences from GPT-2, including personal information, and a later study demonstrated scalable extraction from open models and ChatGPT. Memorization is more likely for duplicated or distinctive sequences and can create privacy and copyright risks. The technically defensible conclusion is therefore:

Better formulation

Generative AI is trained through a statistical transformation of enormous amounts of human-produced data and human feedback. It usually generates rather than retrieves, but it can sometimes reproduce training material. Whether the underlying collection, licensing, compensation, and output uses are legitimate is an ethical and legal question that the engineering description alone does not answer.

Sources: Carlini et al., “Extracting Training Data from Large Language Models” · Nasr et al., “Scalable Extraction of Training Data from (Production) Language Models”

Llama 3 and the 22-million-liter figure (1:12:42)

The film states that training Meta’s Llama 3 used 22 million liters over 97 days, comparable to more than four centuries of one London resident’s household use. The 22-million-liter number was widely circulated after a 2024 Washington Post analysis conducted with researchers at the University of California, Riverside. It is best treated as a modeled estimate, not a company water-meter disclosure:

• Water use depends on where and when computation runs, cooling design and weather, facility water-use effectiveness, and the water consumed in producing electricity.
• “Withdrawal” and “consumption” are not interchangeable. Water withdrawn may be returned; water consumed is evaporated or otherwise unavailable to the immediate watershed.
• Meta published compute and carbon information for Llama 3 but did not publish a model-specific metered water total matching the film’s wording.
• The number concerns a global model-training run, not a Kenyan data center.

The order of magnitude is plausible enough to motivate scrutiny, but the note should not present 22 million liters as an independently audited meter reading or use it to infer the consumption of a particular Kenyan facility. Meta’s Llama 3 model card · Ren et al., “Making AI Less ‘Thirsty’”

Are AI data centers being built in Kenya, and how much water do they use?

Canonical infrastructure dossier

For the cooling-system mechanics, the limits of “closed loop,” cloud-versus-AI comparisons, population-equivalent calculations, and the full source ledger, see AI Data Centers and Water Use. This section remains the documentary-specific Kenya audit.

What is operating or under construction

• iXAfrica NBOX1, Nairobi: launched in 2024 and currently operating. iXAfrica lists 4.5 MW live IT capacity in the first building and 22.5 MW planned across the campus. It markets the facility as hyperscale and AI-ready. This is probably the center referred to by the documentary’s line that it “will be the largest in East Africa,” although the film does not identify the footage. iXAfrica campus specifications · Schneider Electric launch announcement
• Nxtra by Airtel, Tatu City: construction began in September 2025. The official plan is a 44 MW, high-density, GPU-ready facility intended for cloud and AI workloads; Airtel’s current target is 2027. Airtel Africa project page · Kenyan ICT Ministry groundbreaking notice
• Microsoft–G42, Olkaria: announced in 2024 as a $1 billion geothermal-powered East Africa Azure region with water-conservation technology. By May 2026, reporting said the project had stalled over power capacity and financial guarantees, so it should not be counted as an operating Kenyan AI campus. Original Microsoft announcement · 2026 status report
• Kenya also hosts conventional and cloud colocation facilities operated by Africa Data Centres, Digital Realty/iColo, PAIX, Safaricom, Telkom Kenya, and others. “AI data center” is often a marketing or workload-readiness label, not evidence that every rack is training frontier models.

What is known about their water

There is no verified public total for how much water Kenyan AI-capable data centers collectively use, and the major facilities do not publish comparable site-level withdrawal, consumption, or water-use-effectiveness figures.

• NBOX1 specifies indirect adiabatic cooling. Its internal air or coolant circuit may be closed, but adiabatic operation can evaporate a separate water stream and therefore require continuing makeup water. iXAfrica says the primary source is the utility company, with a treated borehole as backup. It publishes no liters-per-day figure, annual total, WUE, utility/borehole split, or wet-mode operating hours. Cooling specifications · iXAfrica sustainability disclosures
• Africa Data Centres says its Nairobi expansion bans adiabatic cooling and uses no water in its cooling systems. That does not mean literally zero facility water—offices, sanitation, construction, and electricity generation can still have water footprints—but it directly disproves “data centers just use fresh water” as a universal technical rule. Africa Data Centres’ Nairobi expansion statement
• Nxtra’s general sustainability program emphasizes air-based cooling, rain harvesting, water reuse, and zero-liquid-discharge technology, but public materials reviewed here do not specify the Tatu City facility’s expected water volume.
• Microsoft and G42 promised “state-of-the-art water conservation technology” for Olkaria but disclosed no quantity; in any case, the large project had stalled by 2026.

Because a daily or annual Kenyan figure is not public, applying a generic global water coefficient to these facilities would create false precision. Any serious local-impact assessment needs the facility’s direct withdrawal and consumption, source by season, cooling configuration, power-source water intensity, discharge quality, and the counterfactual availability of that water to surrounding communities.

Does Kenya struggle to obtain fresh drinking water?

Yes nationally and in many Nairobi neighborhoods—but the film’s phrasing needs geographic and statistical qualification.

• Kenya’s Ministry of Water classifies the country as water-scarce, with approximately 452 cubic meters of renewable freshwater per person per year, well below the commonly used 1,000-cubic-meter scarcity threshold. Kenyan Ministry of Water
• The 2022 Kenya Demographic and Health Survey found 68% nationally had at least basic drinking-water service and only 65% had sufficient quantities when needed. It reported 99% at least-basic access in Nairobi City. “Basic” means an improved source reachable within a defined collection time; it does not mean that every dwelling has a continuously pressurized, potable tap. KNBS 2022 survey
• Nairobi County’s 2023–2027 development plan estimated demand around 850,000 cubic meters per day versus supply around 525,000, a deficit of about 325,000 cubic meters per day, or 38% of demand. It attributes the gap to inadequate raw-water development, old infrastructure, non-revenue water, and governance problems. Nairobi County Integrated Development Plan
• Peer-reviewed fieldwork in Nairobi’s informal settlements documents intermittent service, dependence on kiosks and vendors, unequal costs, and water-quality concerns. Those findings make Okinyi’s account of buying drinking water plausible for particular neighborhoods. They do not establish that Nairobi’s municipal tap water is generally salty. Salty water is more characteristically associated with some boreholes and informal supplies. PLOS Water study of Nairobi’s informal water markets

Final verdict on the film’s Kenya argument

The risk proposition is reasonable: adding water-consuming industrial infrastructure in a water-scarce country with unequal urban service deserves disclosure and regulation. The present-tense harm claim is not yet demonstrated: the documentary does not identify the facility, give its water permit or consumption, compare dry-season withdrawals with local supply, or show that residents lost water because of it. The most defensible demand is transparent site-level water reporting and watershed-specific impact assessment—not the assumption that every Kenyan data center necessarily consumes the same fresh water in the same way.

---

Visual map of the documentary’s argument

How to read these visuals

These diagrams reconstruct what the documentary is attempting to explain, not what this note has independently established. Solid arrows represent an explicit historical, institutional, financial, or production relationship narrated in the film. Dotted arrows represent the film’s broader interpretive claim that a concept, social logic, or ideology persists into a later domain. The Fact-Check Dashboard and research audit evaluate where those proposed continuities are supported, compressed, or unproved.

1. The whole film in one causal map

flowchart TB
    subgraph O["Origins the film emphasizes"]
        O1["Empire and colonial hierarchy"]
        O2["Eugenics and race science"]
        O3["Statistics, psychometrics, and classification"]
        O1 --> O2
        O2 --> O3
    end

    subgraph A["Abstraction of the human"]
        A1["IQ and psychometric g"]
        A2["Mind treated as measurable behavior"]
        A3["Turing machines, cybernetics, and symbolic AI"]
        A4["Machine learning as pattern detection"]
        A1 --> A2
        A2 --> A3
        A3 --> A4
    end

    subgraph I["Modern ideology and mythology"]
        I1["Silicon Valley founder-genius myth"]
        I2["Technological optimism and libertarianism"]
        I3["TESCREAL, Effective Altruism, and AI safety"]
        I4["AGI, superintelligence, and the Singularity"]
        I1 --> I2
        I2 --> I3
        I3 --> I4
    end

    subgraph P["Material production system"]
        P1["Scraped data and behavioral extraction"]
        P2["Hidden annotation and moderation labor"]
        P3["Chips, cloud platforms, and data centers"]
        P4["Venture capital, state support, and military contracts"]
        P1 --> P2
        P2 --> P3
        P4 --> P3
    end

    subgraph C["Consequences alleged by the film"]
        C1["Racialized and gendered classification"]
        C2["Labor and environmental exploitation"]
        C3["Corporate and state power concentration"]
        C4["Surveillance, targeting, and weaponization"]
        C5["Synthetic media and erosion of shared truth"]
        C1 --> C3
        C2 --> C3
        C3 --> C4
        C3 --> C5
    end

    subgraph R["The film's proposed response"]
        R1["Reject inevitability and anthropomorphic hype"]
        R2["Ask: Why does this need AI?"]
        R3["Prefer accountable, task-specific techniques"]
        R4["Worker, community, and democratic agency"]
        R1 --> R2
        R2 --> R3
        R3 --> R4
    end

    O3 --> A1
    O3 -.-> A4
    A4 --> P1
    A1 -.-> I4
    I4 --> P4
    P3 --> C2
    P3 --> C3
    A4 --> C1
    C4 --> C5
    C5 --> R1
    C2 --> R4

    classDef history fill:#eadfcf,stroke:#8b6f47,color:#241a0f;
    classDef concept fill:#dce8f2,stroke:#4e7698,color:#102334;
    classDef ideology fill:#e8def2,stroke:#765692,color:#271536;
    classDef material fill:#e1ead9,stroke:#637e4e,color:#172210;
    classDef harm fill:#f2dddd,stroke:#9a5656,color:#351515;
    classDef response fill:#d9eee8,stroke:#44806f,color:#0f2922;
    class O1,O2,O3 history;
    class A1,A2,A3,A4 concept;
    class I1,I2,I3,I4 ideology;
    class P1,P2,P3,P4 material;
    class C1,C2,C3,C4,C5 harm;
    class R1,R2,R3,R4 response;

The film’s central thesis is not merely that biased data produce biased outputs. It proposes a much longer chain: imperial ranking → eugenic measurement → statistical abstraction → machine models of mind → Silicon Valley hierarchy and AGI mythology → extractive infrastructure → concentrated political and epistemic power. Its concluding alternative is selective, democratic control of technology rather than total technological rejection.

2. Historical and conceptual genealogy

flowchart LR
    subgraph E["Empire, eugenics, and measurement"]
        Darwin["Charles Darwin<br/>evolution"]
        Empire["British empire<br/>racial hierarchy"]
        Galton["Francis Galton<br/>eugenics and measurement"]
        Pearson["Karl Pearson<br/>biometry and statistics"]
        Binet["Alfred Binet and Théodore Simon<br/>educational testing"]
        Spearman["Charles Spearman<br/>psychometric g"]
        Stats["Correlation, deviation,<br/>classification, regression"]
        IQ["IQ as ranking technology"]
        Empire --> Galton
        Darwin --> Galton
        Galton --> Pearson
        Pearson --> Stats
        Binet --> Spearman
        Spearman --> IQ
        Stats --> IQ
    end

    subgraph L["Law, coercion, and postwar survival"]
        USLaw["U.S. immigration restriction<br/>and sterilization laws"]
        Whitney["Leon Whitney<br/>The Case for Sterilization"]
        Nazi["Nazi racial hygiene<br/>sterilization and genocide"]
        Stigma["After 1945:<br/>the word eugenics is discredited"]
        Draper["Wickliffe Draper"]
        Pioneer["Pioneer Fund<br/>postwar hereditarian network"]
        Survival["Postwar continuation<br/>under new labels and institutions"]
        Gottfredson["Linda Gottfredson<br/>g, IQ, and group differences"]
        ShockleyRace["William Shockley<br/>dysgenics and sterilization proposal"]
        IQ --> USLaw
        Whitney --> Nazi
        USLaw --> Nazi
        Nazi --> Stigma
        Draper --> Pioneer
        Stigma --> Survival
        Pioneer --> Survival
        Survival --> Gottfredson
        Survival --> ShockleyRace
    end

    subgraph M["Mind becomes machine-readable"]
        Turing["Alan Turing<br/>universal computation and imitation"]
        Ryle["Gilbert Ryle<br/>mind, behavior, and intellectualism"]
        Cyber["Warren McCulloch and Norbert Wiener<br/>brain-machine models and cybernetics"]
        Military["Wartime and postwar<br/>military laboratories"]
        McCarthy["John McCarthy<br/>the AI label and technological optimism"]
        Symbolic["Symbolic AI<br/>problem solving and goal achievement"]
        Military --> Turing
        Military --> Cyber
        Ryle --> Turing
        Turing --> Symbolic
        Cyber --> Symbolic
        McCarthy --> Symbolic
    end

    subgraph S["Silicon Valley lineage"]
        ShockleyTech["William Shockley<br/>transistor and Stanford"]
        Semi["Shockley Semiconductor"]
        Fairchild["Fairchild Semiconductor<br/>microchip"]
        Intel["Intel and the Fairchildren<br/>startup ecosystem"]
        Founder["Male founder-genius myth<br/>patrilineal technological birth"]
        Optimism["Techno-optimism,<br/>libertarianism, and disruption"]
        ShockleyTech --> Semi
        Semi --> Fairchild
        Fairchild --> Intel
        Intel --> Founder
        Founder --> Optimism
        ShockleyRace -.-> Founder
    end

    subgraph G["From intelligence to AGI"]
        EA["Effective Altruism<br/>AI safety and future value"]
        Bostrom["Nick Bostrom<br/>superintelligence and enhancement"]
        Kurzweil["Ray Kurzweil<br/>Singularity and exponential growth"]
        Legg["Shane Legg<br/>AGI terminology and definitions"]
        AGI["AGI and ASI<br/>human-level and superhuman ranking"]
        Singularity["Intelligence explosion,<br/>mind upload, and posthuman future"]
        EA --> Bostrom
        Bostrom --> AGI
        Kurzweil --> Singularity
        Legg --> AGI
        AGI --> Singularity
    end

    Stats -.-> Symbolic
    IQ -.-> AGI
    Gottfredson -.-> Legg
    Symbolic --> AGI
    Optimism --> AGI

    classDef person fill:#f5ecd9,stroke:#947343,color:#251a0d;
    classDef institution fill:#dfe9ef,stroke:#52758a,color:#10242f;
    classDef coercion fill:#f1dada,stroke:#965252,color:#321313;
    classDef idea fill:#e9e0f1,stroke:#73598c,color:#241631;
    class Darwin,Galton,Pearson,Binet,Spearman,Whitney,Draper,Gottfredson,ShockleyRace,Turing,Ryle,Cyber,McCarthy,ShockleyTech,Bostrom,Kurzweil,Legg person;
    class Pioneer,Military,Semi,Fairchild,Intel institution;
    class USLaw,Nazi coercion;
    class Empire,Stats,IQ,Stigma,Survival,Symbolic,Founder,Optimism,EA,AGI,Singularity idea;

This genealogy contains several different kinds of relation. Galton–Pearson, U.S.–Nazi eugenic influence, Shockley Semiconductor–Fairchild, and the institutional roles of the Pioneer Fund are historical relationships. By contrast, Pearsonian statistics → AI, psychometric g → AGI, Shockley’s eugenics → Silicon Valley culture, and Gottfredson → the substance of AGI are the documentary’s interpretive bridges and require the additional causal evidence examined below.

3. Contemporary AGI, money, laboratories, and the state

flowchart TB
    subgraph Myth["Legitimating story"]
        Genius["Exceptional male genius<br/>should lead the future"]
        Risk["AGI is both existential threat<br/>and unavoidable salvation"]
        Safety["AI safety and alignment"]
        Sing["Singularity, abundance,<br/>space expansion, and posthumanity"]
        Genius --> Risk
        Risk --> Safety
        Risk --> Sing
    end

    subgraph Money["Funding and authority"]
        VC["Venture capital<br/>bubble-scale returns"]
        EADonors["Tech wealth and<br/>Effective Altruist donors"]
        ThinkTanks["AI-safety nonprofits,<br/>research centers, and think tanks"]
        Hype["AGI proximity<br/>as investment narrative"]
        EADonors --> ThinkTanks
        ThinkTanks --> Safety
        VC --> Hype
        Risk --> Hype
    end

    subgraph Labs["Corporate race"]
        DeepMind["DeepMind<br/>Shane Legg and AGI goal"]
        OpenAI["OpenAI<br/>Sam Altman"]
        Anthropic["Anthropic<br/>spun out of OpenAI"]
        XAI["xAI and Grok<br/>Elon Musk"]
        Meta["Meta<br/>Mark Zuckerberg"]
        Scale["Scaling race:<br/>more compute, data, and capital"]
        OpenAI --> Anthropic
        DeepMind --> Scale
        OpenAI --> Scale
        Anthropic --> Scale
        XAI --> Scale
        Meta --> Scale
    end

    subgraph State["State and geopolitical alignment"]
        Trump["Trump administration<br/>AI Action Plan"]
        Stargate["Project Stargate<br/>Altman, Oracle, SoftBank"]
        Procurement["Government procurement:<br/>objective and non-woke AI"]
        Defense["Pentagon and Army<br/>commercial-military integration"]
        Gulf["U.S.-Gulf data centers<br/>and advanced-chip deals"]
        Trump --> Stargate
        Trump --> Procurement
        Stargate --> Scale
        Defense --> Scale
        Gulf --> Scale
    end

    subgraph Power["Political result alleged by the film"]
        Oligarchy["A small group controls<br/>capital, compute, and infrastructure"]
        Epistemic["Models acquire epistemic power:<br/>their outputs appear authoritative"]
        MilitaryAI["Surveillance, targeting,<br/>cloud dependence, and weaponization"]
        Ideology["Model behavior becomes<br/>a government culture-war target"]
        Scale --> Oligarchy
        Oligarchy --> Epistemic
        Oligarchy --> MilitaryAI
        Procurement --> Ideology
        Defense --> MilitaryAI
    end

    Hype --> Scale
    Safety --> OpenAI
    Safety --> DeepMind
    Scale --> Oligarchy

    classDef story fill:#e9e0f1,stroke:#73598c,color:#241631;
    classDef funding fill:#f2ead7,stroke:#927540,color:#2b200d;
    classDef lab fill:#dce9f0,stroke:#4e748a,color:#10242f;
    classDef state fill:#e2e4eb,stroke:#5e6577,color:#171b28;
    classDef result fill:#f1dddd,stroke:#945656,color:#341515;
    class Genius,Risk,Safety,Sing story;
    class VC,EADonors,ThinkTanks,Hype funding;
    class DeepMind,OpenAI,Anthropic,XAI,Meta,Scale lab;
    class Trump,Stargate,Procurement,Defense,Gulf state;
    class Oligarchy,Epistemic,MilitaryAI,Ideology result;

The film treats AGI as simultaneously a technical objective, fundraising story, secular theology, geopolitical project, and claim to elite authority. Fear of extinction and promises of abundance both point toward the same prescription: give the leading laboratories more money, infrastructure, and discretion so that they can build and control the system first.

4. The hidden human and material pipeline

flowchart LR
    subgraph Inputs["Inputs hidden by the automation story"]
        Data["User data, scraped text,<br/>images, and behavior"]
        Workers["Annotators and moderators<br/>in Kenya and other labor hubs"]
        Chips["Chips, cloud platforms,<br/>and software supply chains"]
        Water["Electricity, fresh water,<br/>land, and cooling"]
    end

    subgraph Production["Production"]
        Label["Labeling, filtering,<br/>and traumatic content review"]
        Train["Model training and optimization"]
        Centers["Hyperscale data centers<br/>as AI factories"]
        Product["Chatbots, predictive systems,<br/>synthetic media, and decision tools"]
        Data --> Label
        Workers --> Label
        Label --> Train
        Chips --> Train
        Centers --> Train
        Water --> Centers
        Train --> Product
    end

    subgraph Conditions["Enabling conditions"]
        Poverty["Economic vulnerability,<br/>colonial language, and weak labor power"]
        Law["Legal gaps and<br/>government accommodation"]
        Secrecy["Outsourcing and opacity<br/>hide the human pipeline"]
        Inevitability["Hype frames rapid scaling<br/>as inevitable progress"]
        Poverty --> Workers
        Law --> Centers
        Secrecy --> Label
        Inevitability --> Centers
    end

    subgraph Harms["Present harms emphasized"]
        Trauma["Low pay, psychological trauma,<br/>and limited remedy"]
        Ecology["Water use, emissions,<br/>and environmental injustice"]
        Bias["Encoded classifications,<br/>stereotypes, and unequal decisions"]
        Slop["High-volume propaganda,<br/>misinformation, and synthetic media"]
        War["Military surveillance,<br/>targeting, and cloud dependence"]
        Truth["Erosion of trust,<br/>shared reality, and democracy"]
        Label --> Trauma
        Centers --> Ecology
        Product --> Bias
        Product --> Slop
        Product --> War
        Slop --> Truth
    end

    subgraph Agency["Counter-politics at the end of the film"]
        Ask["Ask what problem is being solved"]
        Refuse["Refuse systems with<br/>no demonstrated value"]
        Design["Build bounded, accountable,<br/>community-directed tools"]
        Organize["Protect workers, communities,<br/>public knowledge, and democratic control"]
        Ask --> Refuse
        Ask --> Design
        Refuse --> Organize
        Design --> Organize
    end

    Trauma --> Organize
    Ecology --> Organize
    Truth --> Ask

    classDef input fill:#dfe9ef,stroke:#52758a,color:#10242f;
    classDef production fill:#e2eadb,stroke:#617e51,color:#172210;
    classDef condition fill:#f0e8d8,stroke:#917443,color:#291e0c;
    classDef harm fill:#f1dada,stroke:#965252,color:#321313;
    classDef agency fill:#d9eee8,stroke:#44806f,color:#0f2922;
    class Data,Workers,Chips,Water input;
    class Label,Train,Centers,Product production;
    class Poverty,Law,Secrecy,Inevitability condition;
    class Trauma,Ecology,Bias,Slop,War,Truth harm;
    class Ask,Refuse,Design,Organize agency;

This is the film’s answer to the phrase “artificial intelligence.” The apparent automation is human and material throughout: people generate the data, people label and moderate it, communities supply land and water, states enable infrastructure, investors finance scaling, and institutions decide how systems are deployed.

5. Who supplies each explanatory lens

flowchart TB
    Film["Ghost in the Machine<br/>explanatory voices"]

    Hist["Genealogy and intellectual history"]
    HistPeople["Angela Saini · Joshua Earle · Aubrey Clayton<br/>Amira Moeding · Michael Kremer · Elizabeth Sandifer"]

    Race["Race, quantification, and critical technology"]
    RacePeople["Brett Zehner · Ezekiel Dixon-Román · Jonathan Flowers<br/>Jessie Daniels · Thema Monroe-White · Dan McQuillan"]

    Hype["AI limits, hype, and political economy"]
    HypePeople["Abeba Birhane · Emily M. Bender · Alex Hanna<br/>David Gerard · Paris Marx · Douglas Rushkoff"]

    Ideology["TESCREAL, AGI, and elite futurism"]
    IdeologyPeople["Émile P. Torres · Shazeda Ahmed<br/>Benjamin Noys · Adam Becker · Becca Lewis"]

    Labor["Labor, extraction, and digital colonialism"]
    LaborPeople["Richard Mathenge · Mophat Okinyi · Krystal Kauffman<br/>Milagros Miceli · Matt Mahmoudi"]

    Infra["Infrastructure and community consequences"]
    InfraPeople["Ana Valdivia · Chase Lochmiller · Alix Dunn<br/>Mel Hogan · Tiera Tanksley · McKenzie Wark"]

    Policy["Policy, procurement, and military use"]
    PolicyPeople["Sorelle Friedler · Sophia Goodfriend"]

    Film --> Hist --> HistPeople
    Film --> Race --> RacePeople
    Film --> Hype --> HypePeople
    Film --> Ideology --> IdeologyPeople
    Film --> Labor --> LaborPeople
    Film --> Infra --> InfraPeople
    Film --> Policy --> PolicyPeople

    classDef center fill:#dce8f2,stroke:#4e7698,color:#102334;
    classDef lens fill:#e8def2,stroke:#765692,color:#271536;
    classDef people fill:#f3ecd9,stroke:#927746,color:#2a200d;
    class Film center;
    class Hist,Race,Hype,Ideology,Labor,Infra,Policy lens;
    class HistPeople,RacePeople,HypePeople,IdeologyPeople,LaborPeople,InfraPeople,PolicyPeople people;

These speakers do not all make the same argument. Some supply historical facts, some offer philosophical or cultural interpretation, some describe direct labor or community experience, and some advance an explicitly anti-capitalist, anti-fascist, anti-eugenic, or anti-TESCREAL analysis. Their claims should therefore be evaluated at the level of the particular speaker and proposition rather than treated as one undifferentiated expert consensus.

Transcript route through the visual

1. Definition and genealogy: Who built “AI”? → eugenics → Galton and Pearson → IQ and g → sterilization → U.S.–Nazi influence.
2. Mind and machine: Turing and Ryle → cybernetics, anthropomorphism, and military laboratories.
3. Silicon Valley mythology: Shockley and the startup lineage → postwar race science → techno-libertarian founder culture.
4. AGI ideology: OpenAI and existential risk → Effective Altruism → Bostrom → AGI → Singularity.
5. Political economy and infrastructure: investment hype → hidden labor → digital colonialism → data centers → state infrastructure.
6. Politics, war, and truth: anti-woke AI policy → militarization → Grok → epistemic crisis.
7. Alternative: agency, refusal, accountable tools, and democratic choice.

People

How the political labels in these profiles are being used

“Woke” is not a stable scholarly category. It can mean anything from a narrow commitment to racial or gender awareness to a broad culture-war label for progressive institutions. The profiles therefore identify the person’s documented theories, causes, and affiliations first, then state how strongly those facts support a broader political characterization.

Evidence scale: explicit means the person or an institutional profile uses the relevant language; supported inference means the pattern is clear across primary work but is not a quoted self-description; not established means the available sources do not justify the claim. Criticism of bias or inequality, by itself, is not treated as proof of party affiliation or adherence to a particular school of theory.

Overall finding: all eight interviewees approach AI from the critical or skeptical side, but not in the same way. Zehner, Flowers, Dixon-Román, and McQuillan explicitly work in identifiable critical-theory, Marxist/autonomist, critical-race, feminist, disability-justice, Black-radical, or anti-fascist traditions. Sandifer’s work is openly anti-neoreactionary and anti-alt-right and is best located in left critical-cultural analysis. Monroe-White’s work is best described as progressive, racial-equity-oriented, and grounded in critical quantitative methods. Torres is a former transhumanist and longtermist who now offers an explicitly anti-eugenic, anti-TESCREAL critique of AI ideology; a broadly left or progressive orientation is a supported inference, but no party affiliation or formal Marxist commitment was verified. Bender’s public AI criticism aligns with several progressive concerns, but the sources reviewed do not establish a partisan identity or formal allegiance to critical theory. No reliable source located in this pass showed any of the eight self-identifying with the word “woke,” and no party affiliation was verified.

Link format used below: each item is labeled as an official profile, primary scholarship, dissertation record/PDF, secondary analysis, or tertiary overview, followed by a one-sentence note explaining why it is useful. This is more informative than preserving a page title or raw URL alone.

Dr. Brett Zehner

AI beliefs

View of AI. Zehner treats machine learning less as a neutral tool or autonomous intelligence than as a system that measures, predicts, and modulates behavior. The work asks how automated prediction reorganizes political subjectivity and how artists or activists might contest those systems through “tactical” uses of media and machine learning.

Political and philosophical beliefs — documented evidence

Assessment. Strong, explicit evidence of a left critical-theory and anti-capitalist framework. Zehner’s scholarship expressly draws on post-autonomist Marxism, critical race theory, racial-capitalism analysis, queer/trans/feminist theory, Foucault, and Deleuze and Guattari. A UC San Diego biography also lists autonomia, gentrification activism, and presentations to the Marxist Literary Group. This is much more precise than calling the work “woke.” No party affiliation was verified.

Background, dissertation, and annotated sources

Role and background. Zehner is a performance-studies and new-media scholar whose research examines machine learning, predictive media, racial capitalism, and technological resistance.

Doctoral dissertation. Seizing the Means of Desubjection: Machine Intelligence, Behavioral Dispossession, and White Racial Capitalism — PhD dissertation (Brown University, 2021) (file not included in this release) (232 pages). The PDF was already present in the vault and was checked against its title page.

Selected sources

• Official profile — current institutional record: University of Exeter Discovery profile. Useful for appointment and publication metadata; the page exposes little biographical text in a plain-web view.
• Official profile — research description: Ohio State postdoctoral profile. Describes Zehner’s project as a genealogy of racial capitalism and automated systems and notes an interest in technological resistance and AI art.
• Official profile — intellectual and activist background: UC San Diego MFA biography. Records interests in autonomia and “depression as politics,” earlier gentrification activism, and the Marxist Literary Group.
• Primary scholarship — clearest statement of AI framework: “Machines of Subjection: Notes on a Tactical Approach to Artificial Intelligence” (PDF). Explicitly situates the analysis in post-autonomist Marxism and critical theory while describing machine learning as a technology of prediction, control, and subject formation.
• Primary scholarship — institutional account of early AI: “A Liar’s Epistemology: Herbert Simon’s Performative Artificial Intelligence”, Qui Parle 30(1), 2021. Argues that Simon’s AI should be understood through organizational procedures and the reordering of human decision-making, not only as an attempt to reproduce an individual mind.
• Primary scholarship — colonial behavioral-data genealogy: “The ‘Situation’ in Micronesia: The Rise of Behavioral Dispossession”, Media-N 19(1), 2023. Traces how culturally situated lives were converted into decontextualized behavioral data and portable claims about rationality; this is the clearest concrete example of what Zehner calls behavioral dispossession.
• Official research description — current book project: Ohio State seminar description, “Disalienation in the Silicon Heartland”. Describes Digital Abolitionism as a genealogy of big data from the perspective of racial capitalism and a study of resistance to behavioral data mining.

Emily M. Bender

AI beliefs

View of AI. Bender argues that large language models generate plausible linguistic form without humanlike understanding, intention, or accountability. She warns that anthropomorphic language and speculative “AGI” narratives distract from present harms involving labor exploitation, data appropriation, surveillance, bias, environmental costs, misinformation, and concentrated corporate power. Her position is critical of the AI industry and its marketing, not a claim that every computational technique is useless.

Political and philosophical beliefs — documented evidence

Assessment. Supported inference of a progressive or left-of-center orientation in her public AI interventions; formal critical-theory or partisan identity not established. The themes of worker power, equity, environmental harm, anti-surveillance, and Big Tech concentration overlap substantially with contemporary progressive politics. However, the reviewed sources do not justify calling Bender a Marxist, a member of a political party, or a self-identified “woke” theorist. Her dissertation is a formal and sociolinguistic study of African American Vernacular English, not a political manifesto.

Background, dissertation, and annotated sources

Role and background. Bender is a University of Washington linguist and computational-linguistics professor. Her doctorate is in linguistics, not political theory or computer science.

Doctoral dissertation. Syntactic Variation and Linguistic Competence: The Case of AAVE Copula Absence — PhD dissertation (Stanford University, 2000) (file not included in this release) (350 pages). Downloaded from Bender’s University of Washington site and checked against its title page.

Selected sources

• Official profile — projects and credentials: Emily M. Bender, University of Washington. Her own faculty site links publications, projects, curriculum vitae, and The AI Con.
• Primary-source index — positions in her own words: Bender’s media and public-scholarship archive. The strongest single index for her op-eds, interviews, policy presentations, and recurring claims about anthropomorphism, hype, workers, surveillance, and current harms.
• Primary scholarship — landmark LLM critique: “On the Dangers of Stochastic Parrots” resource page. Collects the 2021 paper and related material on the risks and limits of very large language models.
• Tertiary overview — biography: Wikipedia: Emily M. Bender. Retained from the original note as a convenient orientation page; use Bender’s own site and publications for substantive claims.
• Tertiary overview — book synopsis: Wikipedia: The AI Con. Retained from the original note; useful for a quick map of the book’s themes, but not a substitute for the book or author interviews.
• User-saved archive — provenance retained: archive.vn snapshot. The snapshot was inaccessible to this research pass, so its underlying article and evidentiary value could not be independently identified.
• Secondary article — contemporary AI-hype debate: “Don’t be fooled by this summer of AI hype”, MIT Technology Review. Retained from the original note as related context; the page was not retrievable in this pass and should not be treated as evidence of Bender’s own words without checking the article.
• Secondary criticism — outside assessment: “The Return of the Luddites”, Los Angeles Review of Books. Useful as a critical response to contemporary AI skepticism; it is an outside author’s argument, not Bender’s statement of belief.

Elizabeth Sandifer

AI beliefs

View of AI. In the documentary, Sandifer explains the early idea of machine thought as a cultural inference: once powerful computers appeared alongside an existing science-fiction vocabulary of robots, artificial life, and Frankenstein, asking whether machines might think became an obvious question (21:34). Her broader work treats AI less as a technical specialty than as a political mythology and rhetorical system. Neoreaction a Basilisk critically examines the cluster around Eliezer Yudkowsky, LessWrong, Roko’s Basilisk, Curtis Yarvin, Nick Land, Peter Thiel, neoreaction, and technologically mediated apocalypse. Her contribution is therefore strongest on the cultural and ideological history surrounding AI—not on computer engineering, machine-learning performance, or the mathematics of computation.

Political and philosophical beliefs — documented evidence

Assessment. Strong, explicit evidence of an anti-alt-right, anti-neoreactionary, anti-authoritarian, and broadly left critical-cultural orientation. Sandifer describes the promotion of Neoreaction a Basilisk as “regearing against the alt-right,” and the book is an openly polemical critique of neoreaction and its adjacent rationalist and technology networks. Her cultural criticism also works through Marxist, feminist, queer, psychoanalytic, and other critical-theory traditions. This supports describing her as a left cultural critic; it does not by itself establish a specific party affiliation or a settled doctrinal identity such as Marxist-Leninist. No party affiliation or self-description as “woke” was verified.

Background, dissertation, and annotated sources

Role and background. Sandifer is an independent author, cultural critic, and founder of Eruditorum Press. Her principal subjects are science fiction, comics, video games, popular music, political myth, and internet ideology. Her major projects include TARDIS Eruditorum, The Last War in Albion, Neoreaction a Basilisk, and, with S. Alexander Reed, They Might Be Giants’ Flood. She earned a PhD in English from the University of Florida.

Doctoral dissertation. Shiny and New: The Rhetorical Logic of the Demo — PhD dissertation (University of Florida, 2010) (file not included in this release) (229 pages). Downloaded from the University of Florida Digital Collections and verified against its title page and PDF metadata. The dissertation analyzes technology demonstrations and rhetorics of progress through the Macintosh, comics, 3-D film, and the Nintendo Wii; it helps explain Sandifer’s expertise in technological rhetoric, but it is not a dissertation in AI or the history of computing.

Selected sources

• Official documentary source — role and statement: PBS Independent Lens, Ghost in the Machine. The transcript records Sandifer’s limited on-screen claim about how computers and existing science-fiction narratives made machine thought an intelligible question.
• Author profile — current self-description and projects: Eruditorum Press author page. Identifies Sandifer as the press’s creator and lists TARDIS Eruditorum, The Last War in Albion, comics work, and Neoreaction a Basilisk.
• Author post — explicit anti-alt-right positioning: “Tuesday Muffining”. Sandifer describes the launch campaign for Neoreaction a Basilisk as a “regearing against the alt-right”; this supports the political characterization without implying an unverified party identity.
• Publisher profile — credentials and books: Bloomsbury author page. Confirms her PhD, College of Wooster background, work on the cultural history of comics and science fiction, and coauthorship of They Might Be Giants’ Flood.
• Institutional dissertation PDF — training and method: University of Florida repository copy. The primary source for her doctorate, dissertation title, year, and analysis of the rhetoric of technological demonstrations; mirrored in the vault above.
• Primary scholarship — digital-media analysis: “Avatari: Disruption and Imago in Video Games”, Digital Humanities Quarterly 3(3), 2009. Demonstrates her academic approach to mediation, embodiment, control, and interactivity in digital systems.
• Author interview — AI, rationalism, and neoreaction: Parallax Views interview on Neoreaction a Basilisk. Useful for Sandifer’s account of the relations among Yudkowsky, Yarvin, Land, Thiel, libertarianism, the alt-right, and apocalyptic AI narratives.
• Book listing — scope and polemical stance: Neoreaction a Basilisk: Essays on and Around the Alt-Right. The publisher description makes the book’s satirical, anti-neoreactionary stance explicit; it is a description of Sandifer’s argument, not independent confirmation of every connection she proposes.

Dr. Johnathan Flowers

AI beliefs

View of AI. Flowers argues that AI ethics should begin from lived experience, concrete applications, and structural power rather than treating fairness as only an abstract principle or mathematical adjustment. His work treats biased systems as evidence of the institutions and social practices that produced their data. He also argues that an AI is “artificial” because it is disconnected from culturally mediated experience, not simply because it simulates a human organism.

Political and philosophical beliefs — documented evidence

Assessment. Strong, explicit evidence of a critical race, feminist, intersectional, and disability-justice framework. Flowers’s listed specialties include critical race theory, critical race feminism, Black feminism, queer feminism, whiteness, identity politics, and disability; his dissertation directly engages Judith Butler’s theory of gender performativity. His approach also draws heavily on Deweyan pragmatism and phenomenology, so “critical theorist” alone is incomplete. No party affiliation or self-description as “woke” was verified.

Background, dissertation, and annotated sources

Role and background. Flowers is an assistant professor of philosophy at California State University, Northridge, trained in American pragmatism, phenomenology, aesthetics, philosophy of technology, disability, race, gender, and sexuality.

Doctoral dissertation. Mono no Aware as a Poetics of Gender — PhD dissertation record (Southern Illinois University Carbondale, 2018). PDF not added: the repository explicitly restricts the file to the SIUC community and directs other readers to interlibrary loan or ProQuest.

Selected sources

• Official profile — credentials: CSUN faculty catalog. Confirms Flowers’s current appointment and 2018 philosophy PhD from SIU Carbondale.
• Research directory — declared areas of specialization: PhilPeople profile. Particularly useful for checking the breadth of his self-listed work in pragmatism, critical race theory, feminist philosophy, disability, race, and gender.
• Primary scholarship — structural account of bias: “Rethinking Algorithmic Bias Through Phenomenology and Pragmatism”. Argues that algorithmic bias can reveal the organizational practices that reproduce inequality rather than being understood only as a technical failure.
• Primary scholarship — concrete and marginal perspectives in AI ethics: “The Forgotten Margins of AI Ethics”. Coauthored survey arguing for ethics grounded in lived experience, marginalized groups, and historical and structural power asymmetries.
• Video interview — accessible overview: “Dr. Johnathan Flowers on Ableist Algorithms and Digital Disability”. Retained from the original note; useful for hearing Flowers explain the disability dimension of algorithmic systems in his own voice.

Ezekiel Dixon-Román

AI beliefs

View of AI. He treats AI, psychometrics, and other systems of quantification as sociopolitical technologies rather than neutral measuring devices. His work asks how algorithmic and statistical systems produce racialized categories, distribute capacity and debility, and participate in governance. He also explores alternative forms of computation and measurement rather than merely demanding a more demographically balanced version of the same systems.

Political and philosophical beliefs — documented evidence

Assessment. Strong, explicit evidence of a left critical-theory, critical-race, Black-radical, decolonial, and racial-capitalism framework. His official title includes “Critical Race,” and his institutional biography uses the language of global neoliberal racial capitalism, biopolitics, coloniality, Black radical traditions, and technopolitical transformation. A Penn interview also records a family history of Young Lords activism and early solidarity with the Cuban Revolution, while presenting these as formative background rather than a current party affiliation. No current party affiliation or self-description as “woke” was verified.

Background, dissertation, and annotated sources

Role and background. Dixon-Román is Professor of Critical Race, Media, and Educational Studies at Teachers College, Columbia University, and holds a PhD in psychometrics. He combines quantitative-methods expertise with critical race, cultural, media, and political theory.

Doctoral dissertation. The Mystery of Inequity: Modeling the Influence of Intergenerational Access on Intellective Competence — PhD dissertation record (Fordham University, 2007). The dissertation uses multilevel growth models to study intergenerational capital and racial and gender achievement differences. PDF not added: Fordham lists a download, but its file server denied access when checked on 2026-09-25; the metadata and abstract remain publicly readable.

Selected sources

• Official profile — most complete intellectual biography: Teachers College faculty profile. Documents his psychometrics training, critical-race appointment, Critical Computation Bureau work, and stated analysis of quantification under racial capitalism.
• Primary scholarship — AI and racialization: “The Racializing Forces of/in AI Educational Technologies” (PDF). A direct source for his claim that educational AI participates in racializing social processes.
• Institutional interview — biography and formative politics: Penn Today Q&A. Provides his own account of inequality research, HBCU experience, family activism, and Cuba; useful context, but not proof of a present partisan identity.

Dr. Dan McQuillan

AI beliefs

View of AI. McQuillan argues that contemporary AI is not simply a fallible tool awaiting better safeguards but an infrastructure that intensifies austerity, exclusion, labor precarity, extraction, surveillance, and concentrated power. His alternatives include mutual aid, people’s councils, solidarity, participatory technology, refusal, and “decomputing”—reducing or dismantling computational systems where they reproduce harmful social and economic relations.

Political and philosophical beliefs — documented evidence

Assessment. Strong, explicit evidence of a radical-left, anti-capitalist, anti-fascist, autonomist, and decolonial orientation. His writings invoke mutual aid, social autonomy, posthumanism, feminist science-and-technology studies, postcolonial computation, necropolitics, and anti-fascism. His publisher biography records anti-Poll Tax, environmental, alternative-globalization, and mutual-aid activism. These documented commitments are more informative than the umbrella word “woke.” No party membership was verified.

Background, dissertation, and annotated sources

Role and background. McQuillan is a Goldsmiths, University of London scholar of critical AI and the author of Resisting AI: An Anti-fascist Approach to Artificial Intelligence. Before academia he worked in disability support, mental-health advocacy, the NHS, Amnesty International, citizen science, and grassroots movements.

Doctoral dissertation. Official biographies confirm a PhD in experimental particle physics from Imperial College London. PDF not added: a targeted search did not identify a reliable public dissertation title or repository copy, so no title has been inferred from unrelated authors with the same name.

Selected sources

• Official profile — publication record and AI position: Goldsmiths faculty profile. Lists Resisting AI, work on non-fascist and autonomist AI, mutual aid, people’s councils, postcolonial computation, and decomputing.
• Author essay — current diagnosis and proposed response: “Decomputing as Resistance”. McQuillan describes AI as a mechanism of precaritization, marginalization, extraction, environmental harm, and wealth concentration, and proposes reorganizing the social relations that sustain it.
• Author essay — anti-fascist framework: “Towards an Anti-Fascist AI”. A concise entry point into his argument that AI should be evaluated through poverty, racism, systemic neglect, solidarity, and collective care.
• Publisher biography — activist history: Bristol University Press author profile. Records his physics training, human-rights and disability work, anti-Poll Tax activity, environmental activism, G8/alternative-globalization involvement, and local mutual aid.

Dr. Thema Monroe-White

AI beliefs

View of AI. Monroe-White focuses on systemic bias in AI and data science, unequal participation in the technical workforce, and how data systems can either reproduce exclusion or be redesigned for empowerment. Unlike the strongest AI abolitionist positions in the documentary, she also emphasizes AI’s potential for positive social transformation when governance, education, workforce representation, and design change.

Political and philosophical beliefs — documented evidence

Assessment. Strong evidence of a progressive, racial-equity, intersectional, and critical-quantitative framework; weaker evidence for a radical-left or orthodox critical-theory identity. Her scholarship explicitly discusses structural racism, minoritized groups, intersectional methods, public-value failure, marginalization, inclusion, and “emancipatory data science.” At the same time, her work is empirical, policy-oriented, and reformist, including government advisory service and NSF-funded workforce research. No party affiliation or self-description as “woke” was verified.

Background, dissertation, and annotated sources

Role and background. Monroe-White is Associate Professor of Artificial Intelligence and Innovation Policy at George Mason University, with appointments in the Schar School of Policy and Government and the College of Engineering and Computing. She is trained in public policy and quantitative methods.

Doctoral dissertation. A Cross Country Investigation of Social Enterprise Innovation: A Multilevel Modeling Approach — PhD dissertation (Georgia Institute of Technology, 2014) (file not included in this release) (199 pages). Downloaded from Georgia Tech’s institutional repository and checked against its title page.

Selected sources

• Official profile — role and research agenda: George Mason welcome profile. Describes her as an AI and policy-justice scholar working at the intersection of science, policy, and racial equity, while quoting her focus on practical solutions and empowerment.
• Primary scholarship — disparities and public-value framework: “Waking up to Marginalization: Public Value Failures in Artificial Intelligence and Data Science”. Uses national data, intersectional methods, and an institutional parity measure to analyze representation and exclusion in data-science education.
• Research-network profile — broader program: National Network for Critical Technology Assessment biography. Useful for her work on race, gender, sexual orientation, STEM participation, algorithmic bias, and emancipatory data science.
• Institutional dissertation PDF — original source: Georgia Tech repository copy. The verified PDF is mirrored in the vault above for durable local access.

Dr. Émile P. Torres

AI beliefs

View of AI. Torres treats the race to build AGI as an intellectual, political, and institutional project rather than an inevitable next step in computer science. With Timnit Gebru, they coined TESCREAL for a proposed bundle of transhumanism, Extropianism, singularitarianism, Cosmism, Rationalism, Effective Altruism, and longtermism. Their thesis is that this bundle supplies much of the AGI movement’s utopian and apocalyptic vocabulary: superintelligence, human enhancement or replacement, space colonization, astronomical future value, and rule by a technically competent elite. Torres emphasizes present harms and concentrations of power that can be obscured when debate is organized around speculative machine extinction or promised posthuman abundance.

Expertise boundary. Torres’s principal contribution is in moral philosophy, intellectual history, existential ethics, and ideological genealogy—not machine-learning engineering or empirical model evaluation. Their TESCREAL framework is a substantive critical thesis that should be tested movement by movement and claim by claim; it is not, by itself, proof that every AGI lab or researcher shares all seven ideologies.

Political and philosophical beliefs — documented evidence

Assessment. Explicitly anti-eugenic and anti-TESCREAL; broadly left/progressive is a supported inference, not a verified party identity. Torres is a forceful critic of longtermism, Effective Altruism’s future-oriented wing, transhumanism, rationalist AI culture, technological utopianism, and concentrated tech power. They argue that highly speculative calculations about an immense posthuman future can discount existing people and authorize extreme measures. That politics overlaps with contemporary left technology criticism, but the reviewed sources do not establish party membership or a formal Marxist doctrine.

Intellectual trajectory. This profile is unusual because Torres writes as a former adherent. Their own biography says they became a transhumanist and longtermist in the early 2010s, wrote for the Future of Life Institute, were affiliated with the Institute for Ethics and Emerging Technologies, assisted Ray Kurzweil’s The Singularity Is Nearer, and spent time at the Future of Humanity Institute and Centre for the Study of Existential Risk. They date their turn toward sustained criticism of TESCREAL to 2019. That insider history is relevant evidence of experience, but it does not make every retrospective interpretation automatically correct.

Existential ethics. Torres does not advocate violently causing human extinction. Human Extinction argues that extinction would be very bad in most circumstances while exploring cases in which it might, on balance, be good. This is a philosophical analysis of extinction’s value, not a political program to produce it. Their dispute with strong longtermism is also not a rejection of caring about future generations; it concerns aggregation, speculative future populations, utopian ends, and what those assumptions can justify now.

Background, dissertation, and annotated sources

Role and background. Torres uses they/them pronouns. They are a moral philosopher, intellectual historian, journalist, Data & Society affiliate, and postdoctoral scholar at Case Western Reserve University’s Inamori International Center for Ethics and Excellence. They earned a BA in philosophy from the University of Maryland, an MS in neuroscience from Brandeis University, and a PhD in philosophy from Leibniz Universität Hannover in 2023.

Doctoral dissertation / book. Their dissertation became Human Extinction: A History of the Science and Ethics of Annihilation (Routledge, 2024), a 542-page history and philosophical analysis of how Western thinkers have understood and valued human extinction. PDF not added: the publisher’s copyrighted edition was linked rather than copied into the vault.

Selected sources

• Official institutional profile — current role and credentials: Data & Society. Confirms Torres’s fields, education, current postdoctoral role, pronouns, four books, and collaboration with Gebru.
• Author biography — intellectual trajectory in their own words: Émile P. Torres, personal site. Documents their former transhumanist and longtermist commitments, institutional proximity to existential-risk organizations, work for Kurzweil, and post-2019 turn against TESCREAL. This is a primary self-account, not independent verification of every interpretation.
• Primary scholarship — central AI genealogy: Gebru and Torres, “The TESCREAL Bundle”, First Monday 29(4), 2024. The paper argues that AGI’s ideological history is connected to eugenics and overlapping techno-utopian movements; a fuller source-status note appears under Additional Resources.
• Primary reference article — mature formulation: Torres, “TESCREAL,” Oxford Research Encyclopedia of Science, Technology, and Society, 2026. Presents the framework as an explanation of the intellectual motivations behind the AGI race.
• Primary public philosophy — critique of longtermism: “Against longtermism”, Aeon, 2021. States Torres’s case that astronomical-value reasoning can minimize present suffering and rationalize dangerous action.
• Publisher record — dissertation and extinction ethics: Routledge book page. The best concise source for the book’s scope and its qualified conclusion that extinction is normally very bad but may not be bad under every conceivable circumstance.
• Independent profile — conversion narrative and controversy: The Observer, “The philosopher questioning humanity’s future”, 2023. Useful for Torres’s account of why grand, future-oriented visions first attracted and later alarmed them; its “pro-extinctionist” framing should be read alongside Torres’s own correction and the book’s more qualified position.

Additional Resources

Gebru & Torres (2024) — The TESCREAL Bundle: Eugenics and the Promise of Utopia Through Artificial General Intelligence

Open the paper: The TESCREAL Bundle — Gebru Torres (PDF, 42 pages) (file not included in this release)

Full citation: Timnit Gebru and Émile P. Torres, “The TESCREAL bundle: Eugenics and the promise of utopia through artificial general intelligence,” First Monday, 29(4), April 2024. DOI and journal version.

Source status. This is a peer-reviewed but explicitly argumentative genealogy and critique. It combines primary-source statements from AGI advocates, institutional and funding histories, investigative reporting, scholarship on eugenics, and the authors’ own experience around AI and TESCREAL communities. It should be read as a developed case for a particular interpretation—not as a neutral survey or proof that every form of AI, every AGI researcher, or every member of the named movements shares one ideology.

Core thesis. Gebru and Torres argue that the modern race to build artificial general intelligence is substantially motivated by a family of overlapping ideologies descended from twentieth-century eugenics. They contend that this lineage survives not merely through the reuse of mathematical tools, but through recurring normative aims: ranking “intelligence,” technologically enhancing or superseding humanity, entrusting the future to a self-appointed technical elite, maximizing a distant posthuman future, and treating present harms as secondary to a promised technological utopia or feared apocalypse.

TESCREAL is the authors’ acronym for:

• Transhumanism — technologically “enhancing” humans and potentially creating a posthuman species.
• Extropianism — a libertarian form of transhumanism emphasizing expansion, self-transformation, technological progress, and growth.
• Singularitarianism — anticipating a technological “Singularity,” often involving recursively improving superintelligence.
• Cosmism — merging humans and machines, mind uploading, radical life extension, sentient AI, and cosmic expansion.
• Rationalism — the LessWrong-centered community focused on improving reasoning and shaping advanced-AI outcomes.
• Effective Altruism — maximizing expected good with limited resources, increasingly associated with AI safety and the far future.
• Longtermism — assigning overriding moral importance to humanity’s potentially enormous posthuman future.

Argument in brief.

1. Modern transhumanism developed partly from eugenic efforts to direct human evolution, but shifted from population-level breeding programs toward technological and individually chosen “enhancement.”
2. The seven TESCREAL traditions overlap through people, organizations, funders, concepts, and utopian or apocalyptic narratives. The authors highlight recurring interests in IQ, cognitive enhancement, posthumanity, space colonization, existential risk, and artificial superintelligence.
3. These networks helped revive AGI as a major research and investment goal. The paper traces connections among figures and institutions associated with AGI, including the early AGI literature, LessWrong and MIRI, DeepMind, OpenAI, Anthropic, Effective Altruism organizations, and technology billionaires.
4. The AGI project, in the authors’ account, produces present harms through systems that are too broadly scoped to test properly, resource-intensive scaling, opaque or poorly curated training data, exploited labor, anthropomorphic marketing, concentrated institutional power, and diversion of attention from affected communities.
5. “AI safety” discourse can then redirect scrutiny toward hypothetical superintelligent extinction scenarios while marginalizing current questions about labor, discrimination, environmental cost, data appropriation, misinformation, and corporate accountability.

Proposed alternative. The authors recommend abandoning the undefined goal of an all-purpose AGI and building well-scoped, well-defined, task-specific systems whose operating conditions, inputs, outputs, failure modes, and safety requirements can be tested using ordinary scientific and engineering principles.

Why it matters for this documentary note. This paper offers a more specific and potentially stronger argument than the blanket claim that “AI has roots in eugenics.” Its target is primarily the AGI movement and the ideological and institutional networks driving it, not every statistical model or automated tool. It attempts to show continuity through people, texts, organizations, funding, definitions of intelligence, and policy influence—the kinds of mechanisms the research audit below says are needed before intellectual ancestry becomes evidence of ideological inheritance.

Important cautions. The authors acknowledge that the seven movements are not identical, that not everyone associated with AGI accepts the complete TESCREAL worldview, and that some people will object to being grouped under the acronym. The paper’s evidence is strongest when it documents explicit statements, institutional relationships, funding paths, and recurring concepts. Its larger causal claim—that this bundle explains the AGI race and its present harms—remains an interpretation that should be tested against competing histories and explanations. Likewise, evidence that AGI development causes exploitation, environmental costs, or discrimination does not by itself prove that eugenics is the cause of each harm.

Bottom line. The paper does more than argue that bad historical actors contributed to techniques later used in AI. It claims that a particular contemporary project—building AGI—repackages eugenic aspirations to transcend, optimize, rank, and centrally direct humanity. That is a substantive continuity claim, but its persuasiveness depends on evaluating each proposed link rather than treating the TESCREAL label itself as proof.

Research and claim audit: Does AI inherit eugenics?

Bottom line

The documentary is on firm ground when it says that Francis Galton, Karl Pearson, and Ronald A. Fisher joined major contributions to statistics with explicit eugenic commitments. It is also reasonable to ask whether present-day automated classification reproduces older hierarchies. But shared intellectual ancestry does not, by itself, establish ideological inheritance. To show that a particular AI system is meaningfully eugenic, the analysis must identify a surviving aim, category, institutional practice, causal mechanism, or measurable effect—not merely a statistical technique also used by eugenicists.

Source boundary

The claims below distinguish the documentary’s argument from supplementary research. The transcript is a record of what the film and its interviewees say; it is not, by itself, independent verification. This audit concentrates on the film’s claims at 8:21–15:38, the proposed Ryle–Turing bridge at 19:45–21:34, the Pioneer Fund passage at 31:45–32:09, and the Gottfredson–Legg claim at 53:47–54:15.

Referenced-conversation provenance

This expansion incorporates questions pursued while watching the documentary in “Intelligence Measurement History” and “Compare Cofnas Murray Responses”.

Those conversations are secondary research aids, not primary evidence. Their reconstructed answers are not direct quotations from Nathan Cofnas, Charles Murray, or the documentary’s interviewees unless a quotation is separately identified and linked. The synthesis below therefore distinguishes: what the film says, how an opposing argument can be reconstructed, what the cited historical or scientific sources establish, and what remains disputed or unproved.

What the documentary claims

The film advances several different propositions that should be evaluated separately:

1. Historical overlap: leading early statisticians were also leading eugenicists.
2. Methodological inheritance: machine learning uses descendants of statistical methods developed in that milieu.
3. Conceptual inheritance: quantifying and ranking intelligence carries forward eugenic assumptions about human hierarchy.
4. Institutional or political continuity: contemporary AI serves racialized classification, extraction, labor control, policing, military power, or elite projects in ways continuous with eugenics.
5. Identity claim: because of those connections, AI itself “has its roots in eugenics” or remains eugenic in character.

The evidence for proposition 1 is strong. Proposition 2 is true in a broad technical sense but does not settle the meaning or ethics of later uses. Propositions 3 and 4 may be true of particular concepts, systems, datasets, or institutions, but each needs a traced mechanism. Proposition 5 is too broad unless “AI” is narrowed to a specific lineage or application.

What is historically well supported

• Galton and Pearson: UCL’s own history says Pearson and his collaborators helped lay foundations of twentieth-century statistics, including correlation, regression, and p-values. The same institutional history records Pearson’s enthusiastic support for Galton’s eugenics, the creation of Biometrika in the effort to give quantitative biology and eugenics a scientific basis, and Pearson’s appointment to the Galton Chair of Eugenics. UCL, “Our Early History”
• Fisher: UCL records that Ronald A. Fisher supported eugenics, helped found the Cambridge University Eugenics Society, and later became the second Galton Professor of Eugenics. His scientific contributions and his political commitments are both part of the record. UCL, “History”
• Statistical migration into medicine: a historical review of clinical trials reports that Fisher introduced randomization in the 1920s and that randomized human trials later adopted the statistical hypothesis-testing paradigm. This supports the analogy to clinical research: techniques associated with figures who supported eugenics migrated into uses whose present purpose is not therefore eugenic. “Advances in clinical trials in the twentieth century”
• The IQ/eugenics connection is more direct than the AI/eugenics connection in general: intelligence tests, hereditarian claims, immigration policy, institutionalization, and sterilization were historically fused in specific people and institutions. That history is mapped in Hereditarianism, Eugenics, Race, and IQ in the United States - A Chronological Timeline. It does not follow that every later intelligence measure—or every technology marketed as AI—has the same purpose.

The Pioneer Fund: eugenic origin, institutional lineage, and principal beneficiaries

Verdict

The documentary’s factual core is strong. The Pioneer Fund was not merely a neutral foundation that happened to finance controversial scholars. Its 1937 charter expressly authorized research and publicity concerning “heredity and eugenics” and “race betterment”; its founder and first president were deeply involved in racial eugenics; and it later became a major patron of hereditarian race-and-IQ research. The film’s phrases “kept race science alive” and “explicitly racist notion of IQ” are interpretive compressions, however. A careful audit should name the grant, publication, argument, and evidentiary limit rather than letting those labels do all the work.

Founding purpose and eugenic lineage

• Founder and charter. Textile heir Wickliffe Preston Draper endowed the fund in 1937. Michael G. Kenny reproduces its purpose as supporting study and research into “heredity and eugenics” and “race betterment with special reference to the people of the United States.” Paul Lombardo’s archival history additionally documents charter preferences involving descendants of white settlers from the original thirteen states, Draper’s racial politics, and the fund’s early reproductive program for Army aviators considered eugenically desirable. Kenny, “Toward a Racial Abyss” · Lombardo, “The American Breed”
• Harry H. Laughlin. Draper selected Laughlin—the Eugenics Record Office leader associated with immigration restriction and model compulsory-sterilization legislation—as Pioneer’s first president. Three months before incorporation, Nazi-controlled Heidelberg awarded Laughlin an honorary degree for his racial-hygiene work. That fact establishes a serious ideological and institutional connection; it does not mean every later grantee shared Laughlin’s program. Lombardo
• Postwar political continuity. William H. Tucker’s archival study traces fund personnel and money into resistance to civil-rights measures, support for repatriation schemes, and advocacy surrounding segregated private schools. The fund and its defenders disputed the characterization that it pursued a political agenda, so Tucker’s claim should be read as a documented historical argument rather than as the foundation’s self-description. University of Illinois Press, The Funding of Scientific Racism · Tucker, “Bankrolling racism”

The dates above are synthesized from archival scholarship, the fund-aligned institutional history, and later tax filings. The aligned history is useful for offices and internal chronology but not neutral on the fund’s purpose. Lombardo · Richard Lynn, The Science of Human Diversity — catalog record · ProPublica

The legal entity has not disappeared from the public record. Its 2024 Form 990-PF reported $304,194 in net assets, $6,217 in revenue, and no charitable disbursements. In 2023 it reported $277,873 in net assets and $13,200 in charitable disbursements. These are direct IRS-derived figures, not an estimate of the wider network’s resources. ProPublica Nonprofit Explorer, EIN 51-0181036 · HOPE not hate, “Race Science Inc.”

Who principally benefited

Accounting limit

Historical totals are not perfectly comparable. Sources use different cutoff years, sometimes route money through institutes rather than people, and sometimes report nominal dollars while other summaries adjust for inflation. The amounts below are therefore labeled by source and date; they identify major beneficiaries, not a mathematically exact all-time ranking. Receipt of tainted funding is also not, by itself, proof that every result a recipient produced was false.

The table triangulates the fund’s own aligned history, archival scholarship, contemporary reporting, and tax-record investigations. The most defensible numbers are those tied to a named year; cumulative figures should be read as historically reported totals rather than a fresh forensic audit. Tucker’s archival study · Washington Post, 1989 · Undark, 2022 · Pioneer-aligned institutional history — catalog record

Other recipients included William Shockley, Michael Levin, Seymour Itzkoff, Robert A. Gordon’s intelligence-and-society project, and organizations publishing or promoting hereditarian research. Recent filings show a much smaller operation than in its peak decades: $537,709 in charitable disbursements in 2010, compared with $37,000 in 2022, $13,200 in 2023, and $0 in 2024. ProPublica · Washington Post on Gottfredson · Undark investigation of Draper’s funding network

Linda Gottfredson: research program and disputed racial claims

Gottfredson is an intelligence and vocational psychologist whose work cannot be reduced to Pioneer funding. Her well-cited research argues that psychometric g predicts learning and job performance, especially as tasks become more complex, and that cognitive complexity matters for functional and health literacy. “Why g Matters: The Complexity of Everyday Life” Her broader publication record covers occupational choice, career development, test validity, employment selection, and health. Those empirical claims need to be evaluated individually; the source of a grant neither validates nor refutes them.

The documentary is nevertheless right that her record extends beyond a neutral definition of intelligence:

• Funding and institutional dispute. By 1989 the Washington Post reported $174,000 in Pioneer support. The University of Delaware later tried to bar faculty from taking new Pioneer grants; an arbitrator overturned the restriction. Gottfredson maintained that the foundation exercised no control over her work. That is relevant counterevidence to a direct-control theory, but it does not erase the foundation’s role in selecting and sustaining a research network. Gottfredson’s primary-document archive of the dispute
• 1994 statement. Gottfredson drafted “Mainstream Science on Intelligence,” published with 52 signatories. It described intelligence as a broad capacity to reason, plan, solve problems, understand complex ideas, and learn; it also presented racial and ethnic group-average IQ claims. Importantly, it said there was no definitive answer for why the distributions differed, that environments mattered, and that genetics could be involved. Journal version
• Later hereditarian commitment. In “What If the Hereditarian Hypothesis Is True?” she defended Rushton and Jensen’s proposal that about half the U.S. Black–White cognitive-test gap was genetic against a culture-only alternative. In 2013 she again defended Rushton’s three-way racial life-history theory and characterized much criticism as “mob science.” These are not merely statements that individual intelligence is partly heritable; they are arguments about socially identified group differences. 2005 commentary · 2013 Rushton essay

The strongest neutral formulation is therefore: Gottfredson received substantial Pioneer funding, produced important work on g and occupational complexity, published claims about racial-group score distributions, and later expressly defended a partly genetic account of the Black–White gap. Calling this an “explicitly racist notion of IQ” communicates Torres’s moral and political judgment, but it compresses distinct issues—how IQ is defined, whether it predicts outcomes, what causes group means, whether racial categories track relevant ancestry, and what policies follow.

Current evidence warrants caution about the hereditarian group claim. Within-group heritability does not determine the causes of a between-group difference; the APA task force found no direct genetic explanation established for the U.S. Black–White gap; a 2021 genomic analysis found no support for the hereditarian hypothesis using polygenic scores and selection tests; and the National Academies advises against using race as a proxy for human genetic variation. These points do not show that genes never influence individual cognitive variation. They show why the specific causal leap from individual heritability or ancestry structure to racial group hierarchy is not established by the evidence cited in the film. APA task-force report, “Intelligence: Knowns and Unknowns” · Bird 2021 · National Academies 2023 · American Society of Human Genetics statement

What the Legg–Gottfredson citation does—and does not—show

At 53:47–54:15, Torres says Shane Legg cited Gottfredson while assembling definitions of intelligence. That narrow bibliographic claim is correct. Legg and Marcus Hutter’s 2007 paper cites “Mainstream Science on Intelligence,” “Why g Matters,” and “g: Highly General and Highly Practical.” It quotes the broad definition of intelligence as a general capacity to reason, plan, solve problems, understand complex ideas, and learn from experience. Legg and Hutter, “Universal Intelligence”

But the paper’s machine construct is not Gottfredson’s racial-group theory. Legg and Hutter synthesize many human and machine definitions, then define intelligence as an agent’s ability to achieve goals across a wide range of environments and formalize it using reward, computable environments, and algorithmic probability. The cited premise is breadth and adaptability, not a ranking of racial groups or a genetic account of score gaps.

The accurate conclusion is consequently two-part:

1. A real textual bridge exists: an important machine-intelligence paper used Gottfredson’s work as part of its survey of human definitions.
2. The documentary has not shown conceptual importation of her racial claims: a citation to her general definition does not establish that her hereditarian conclusions became part of Legg and Hutter’s formula, DeepMind’s engineering, or AGI as a field.

This is a stronger and more testable claim than either extreme. “There is no connection” ignores an actual citation; “AGI therefore encodes Pioneer Fund race science” confuses citation lineage with adoption of the controversial proposition.

Source-status ledger

• Primary and near-primary: the 1937 charter language as reproduced in historical scholarship; IRS Form 990-PF data; Gottfredson’s papers and University of Delaware archive; Legg and Hutter’s paper.
• Independent historical scholarship: Lombardo, Kenny, and Tucker supply the archival account of origins, political activity, and institutional continuity.
• Organization-aligned history: Richard Lynn’s The Science of Human Diversity and Pioneer defenses are useful for leadership, programs, and the fund’s own case, but not neutral adjudication.
• Advocacy and investigative claims: the Southern Poverty Law Center’s “hate group” label and HOPE not hate’s successor-network investigation are not government or judicial findings. Where used, they are attributed rather than converted into settled legal facts.
• Grant-total caveat: exact all-time recipient rankings remain approximate because historical records use inconsistent years, inflation adjustments, and institutional conduits.

Questions pursued while watching

Was quantitative intelligence invented by eugenics?

Verdict

Modern standardized, numerical psychometrics largely took shape in the late nineteenth and early twentieth centuries and was deeply entangled with eugenics, but it was not exclusively created by eugenics. Francis Galton’s program directly joined measurement of individual differences to heredity and eugenic selection. Alfred Binet’s work is a crucial counterexample: the Binet–Simon scale was developed to identify French schoolchildren who needed educational assistance, and Binet did not treat its score as a fixed measure of innate worth. In the United States, figures including Henry Goddard and Lewis Terman adapted intelligence testing within more hereditarian and eugenic programs.

The historical claim becomes clearer when six propositions are kept separate:

This is why both blanket histories fail. “Intelligence measurement was simply invented by eugenics” erases Binet and other educational or clinical motives. “Testing merely happened to coexist with eugenics” erases the direct Galtonian program and the uses of tests in U.S. immigration restriction, institutionalization, and sterilization. APA on Binet and the origins of intelligence testing · National Academies overview of assessment history · Historical review of values in psychometrics

How Cofnas and Murray would answer Saini’s population-genetics passage

Attribution boundary

The comparison below is a reconstruction from published positions, not a verbatim response by either person to this documentary. It is included to expose the strongest disagreement with the film, not to endorse it. Nathan Cofnas makes the more explicit hereditarian argument; Charles Murray has often been more cautious about the relative genetic and environmental contribution to U.S. group IQ gaps, even while treating biological ancestry and intelligence research as legitimate subjects.

At 9:28–10:13, Angela Saini moves among three claims: humans form one species; most human genetic variation is found within rather than between populations; and old racial accounts of inherited psychological hierarchy are false. A Cofnas/Murray-style response would separate them:

Best reconstructed reply — not a verbatim quotation

Humans are one species, and most genetic variation is within populations. Neither fact implies that ancestry structure disappears or that average biological differences between populations are impossible. But showing ancestry structure is only the beginning of the argument: it does not establish that a psychological group difference is genetic.

Cofnas explicitly invokes the Lewontin–Edwards debate to argue that within-population variation does not eliminate multivariate population structure. Murray describes race as partly social while still treating ancestry-linked biological variation as meaningful; his older formulation concerning the U.S. Black–White IQ gap was that he remained “resolutely agnostic” about the genetic/environmental mixture. These positions should be read in their own words and assessed independently: Cofnas, “Research on Group Differences in Intelligence: A Defense of Free Inquiry” · Charles Murray interview with Bill Kristol · Murray, “Genes, Race, IQ: An Apologia”

What within-population variation can—and cannot—show

Two statistics that sound similar are often conflated:

• Statements that two human genomes are about 99.6% or 99.9% identical describe average sequence similarity, with the percentage depending on which kinds of variation are counted.
• Statements that roughly 85–95% of human genetic variation occurs within populations apportion variation at studied loci within and among sampled groups.

Neither statistic means “race is 1% genetic,” nor does either directly measure the causes of intelligence, temperament, or another complex trait. A.W.F. Edwards’s multivariate point was that weakly differentiated loci can, in combination, reveal population structure even when most variation at each locus lies within populations. Later clustering studies likewise recover ancestry-related structure from many markers. At the same time, human variation is geographically graded, populations overlap, and clustering results depend on samples and methods. The compatible synthesis is: one species, extensive within-population diversity, real but non-discrete ancestry structure, and no automatic inference from that structure to genetic behavioral differences. NHGRI on human genomic variation · Edwards, “Human Genetic Diversity: Lewontin’s Fallacy” · Winther, “The Genetic Reification of ‘Race’?” · Rosenberg et al., “Genetic Structure of Human Populations”

Mundane examples of multivariate classification

Many familiar sciences classify overlapping cases by combining several individually imperfect measurements:

Field	Overlapping inputs	Multivariate result
Botany	Iris sepal and petal measurements	Species can be separated better jointly than by any single measurement.
Food chemistry	Acids, alcohols, phenolics, and trace compounds in wine	Samples can often be associated with grape variety or geographic origin.
Microbiology	Many genetic markers	Closely related strains or source populations can be distinguished.
Geology	Elemental and isotopic signatures	Rocks or sediments can be associated with source formations.
Medicine	Symptoms, labs, imaging, and risk factors	Diagnostic models can distinguish diseases despite overlap on every individual feature.

These analogies establish only the statistical point that within-group overlap on individual variables does not prevent useful joint classification. They do not prove that folk racial categories are biological subspecies, that every clustering scheme is natural, or that a behavioral group difference has a genetic cause.

Did American sterilization policy influence Nazi Germany?

Verdict

Yes, materially—but “Hitler copied the United States” is too simple. American laws and eugenic literature provided prominent precedents that German racial hygienists studied and cited. Nazi Germany also had its own well-developed racial-hygiene movement, and its dictatorship expanded compulsory sterilization and racial persecution far beyond the American model.

• 1907: Indiana enacted the first compulsory eugenic-sterilization law in the United States—and is commonly identified as the first jurisdiction in the world to enact such a statute. This does not mean it conducted the first nonconsensual sterilization ever.
• 1909: California enacted its law and eventually carried out more than 20,000 sterilizations, the largest U.S. total. Harry Laughlin’s 1922 model law and the Supreme Court’s 1927 decision in Buck v. Bell further legitimized coercive programs.
• 1933: Nazi Germany enacted the Law for the Prevention of Hereditarily Diseased Offspring. Roughly 400,000 people were forcibly sterilized under the Nazi program; the later “euthanasia” program murdered about 250,000 disabled patients.
• Targets and coercion: Indiana and California laws applied to institutionalized people categorized with elastic labels such as “feeble-minded,” “insane,” epileptic, criminal, sexually deviant, or morally unfit. In practice, poverty, disability, institutional custody, gendered judgments about sexuality, and lack of meaningful consent shaped who was vulnerable. Procedures could be compulsory or made a condition of release.
• Scale: More than 60,000 people were sterilized under U.S. eugenic laws across over thirty states. The U.S. precedent mattered, but the scale and lethal radicalization of Nazi policy were not merely an American program transplanted unchanged.

See Hereditarianism, Eugenics, Race, and IQ in the United States - A Chronological Timeline for the longer chronology. NHGRI eugenics timeline · U.S. Holocaust Memorial Museum overview of eugenics · USHMM, “Nazi Racial Hygiene”

Where the documentary overreaches or needs more evidence

The ancestry-to-identity jump

The fact that a technique arose partly within a bad project can be historically illuminating without fixing the moral identity of every later use. Correlation, regression, experimental design, and significance testing can be used to rank populations, test medicines, forecast storms, audit discrimination, or analyze baseball. Their origins do not make all those applications equivalent.

This is where the genetic fallacy becomes relevant: an account of origin is not automatically an assessment of present validity or purpose. The word genetic here means about genesis or origin, not biological genetics. However, invoking the fallacy is justified only when the historical origin is irrelevant to the present claim. If categories, objectives, institutions, or harms demonstrably persist, genealogy may be probative rather than fallacious.

Tool-by-tool audit: standard deviation, correlation, logit, and logistic regression

At 12:48, the documentary says that “standard deviation, correlation, the logit model, logistic regression emerged specifically out of eugenics.” The sentence packages together methods with different histories:

The defensible claim is narrower: biometry, early statistics, heredity research, and eugenics substantially overlapped in Galton’s and Pearson’s networks. That overlap does not make every named method originate “specifically” in eugenics. MacTutor biography of Pearson · Cramer, “The Origins of Logistic Regression” · Oxford’s Significance history of Cox · Cox’s 1958 paper

“General intelligence” is not yet a demonstrated causal bridge to AGI

Verdict

Conceptually, yes; etymologically, not as a demonstrated direct borrowing from Spearman’s symbol. The G in AGI literally abbreviates general. Its immediate purpose was to distinguish broad, adaptable intelligence from narrow AI. Early AGI authors nevertheless explicitly recognized a “loose relationship” between their use of general intelligence and psychology’s g factor. The two belong to the same conceptual family, but they are not the same construct.

What the letter means and where the term came from

Spearman’s lowercase, italicized g is a psychometric construct: a latent common factor inferred from positive correlations among performance on different human cognitive tests. His influential 1904 paper formalized this general-factor approach. Charles Spearman, “General Intelligence,” Objectively Determined and Measured

The G in AGI, by contrast, is simply the first letter of general. Ben Goertzel’s history of the modern term says that while he and Cassio Pennachin were editing a book on broadly capable, human-level-or-beyond AI in 2002, Shane Legg suggested Artificial General Intelligence as its title. The term distinguished their objective from specialized systems and later spread through the AGI conference community. Goertzel subsequently learned that Mark Gubrud had already used the same phrase in a 1997 discussion of advanced military technology. This origin account does not say that the acronym’s G was selected as an homage to Spearman’s g. Ben Goertzel, “Artificial General Intelligence: Concept, State of the Art, and Future Prospects,” term-history note

Why the concepts are nevertheless related

Pei Wang and Ben Goertzel’s foundational AGI introduction directly addresses the issue. They say that AGI stresses intelligence as a whole, with many aspects and applicability across situations, in contrast to systems with highly specialized capabilities. They then describe a “loose relationship” to psychology’s g factor because g attempts to measure human intelligence across domains. The relationship is therefore acknowledged rather than merely an accidental reuse of the word general. Pei Wang and Ben Goertzel, “Introduction: Aspects of Artificial General Intelligence”

What the connection does not establish

The film links Spearman’s psychometric g to “artificial general intelligence.” The shared concept is worth investigating, but neither related terminology nor the explicit “loose relationship” establishes that psychometric g directly produced AGI as an intellectual or technical project. A persuasive lineage would still need to show which AI researchers adopted Spearman’s construct, how it shaped a technical objective or benchmark, and which assumptions survived the move from human psychometrics to machine capability.

Three claims should therefore remain separate:

1. Breadth of performance: an AI performs well across many tasks.
2. An AI g factor: performance across a battery of AI benchmarks exhibits a common statistical factor.
3. AGI: a system possesses broad, adaptable, transferable intelligence under some substantive definition of generality.

The first may provide evidence for the second; neither automatically proves the third. An AI could be broadly capable through many interacting specialized mechanisms rather than one unitary g-like capacity. Conversely, a strong common factor across benchmarks could reflect benchmark construction, training-data overlap, scale, or other shared causes without demonstrating autonomous or out-of-distribution generalization. The careful conclusion is therefore: AGI’s “general” is conceptually adjacent to Spearman’s g, but a direct etymological derivation and a causal psychometric-to-AGI lineage have not been demonstrated.

Ryle, Turing, and the alleged eugenic transmission (19:52–21:34)

Verdict

The passage contains several real historical connections, but its central conclusion does not follow from them. Karl Pearson’s statistics and eugenics were deeply intertwined; Reginald Ryle was Pearson’s physician and intellectual correspondent; John Ryle participated in the Eugenics Society; Gilbert Ryle and Alan Turing worked in overlapping wartime intelligence circles; and Ryle edited Mind when it published Turing’s 1950 paper. What the film does not establish is the indispensable middle of its genealogy: that Gilbert Ryle adopted a particular eugenic idea, built it into his account of intelligence, transmitted it to Turing, and thereby shaped a specific AI objective or practice. The evidence demonstrates association and possible philosophical affinity, not ideological transmission.

The surrounding setup also needs one correction. A universal Turing machine can simulate any other Turing machine and compute any computable function; it cannot “calculate everything within the mathematical universe” (18:44). Turing’s own 1936 result helped prove that some mathematical decision problems are not computable.

Claim-by-claim fact-check

What Zehner means by “white flight from the body” (20:59)

Expanded documentary selection (19:52–21:34)

19:52 · Michael Kremer: His father was the family physician of Karl Pearson, who more or less invented modern statistics in service of his eugenicist projects.

20:02 · Kremer: But then his brother John Ryle was vice president of the Eugenics Society.

20:07 · Kremer: You know, I don’t think Gilbert Ryle was a eugenicist, but he did think in terms of human capacities.

20:14 · Kremer: He spends a lot of time on the question of intelligence.

20:17 · Documentary narration: Ryle attacked what he called the “dogma of the ghost in the machine,” which he associated with the famous French philosopher René Descartes.

20:28 · Documentary narration: Cartesian dualism is the idea that the mind and the body are separate things, and so when the body dies, the soul persists, that kind of thing.

20:36 · Documentary narration: Gilbert Ryle’s concept of mind: he says consciousness is a product of the materiality of the body.

20:45 · Kremer: There is a position that he calls “intellectualism,” and that’s the position that what makes it intelligent behavior is that it’s guided by the thinking of thoughts, or, as he says, “the contemplation of rules.”

20:59 · Brett Zehner: So these kinds of ideas that move from eugenics as a genetic grounding of white supremacy to theories of the mind or theories of behavior allow whiteness to have a new kind of white flight from the body, which really paints a picture of how AI is operationalized today.

21:18 · Kremer: Now, Alan Turing and Gilbert Ryle knew each other during the war.

21:22 · Kremer: They were involved in something that one could say was trying to figure out the minds of these other people.

21:28 · Kremer: 1950, Ryle accepted Turing’s paper “Computing Machinery and Intelligence” for publication.

21:34 · Documentary narration: Turing wants to answer the question: Can machines think?

What is—and is not—Zehner’s “full quote”? The sentence at 20:59 is his complete uninterrupted statement in the documentary. The larger argument only becomes visible when the surrounding remarks by Kremer and the narrator are included. Kremer supplies the Pearson → Ryle-family association before Zehner speaks; after Zehner speaks, Kremer and the film supply the Ryle → Turing connection. Zehner’s own sentence does not identify the people, technical steps, or documentary evidence that carry whiteness into AI. It states the film’s interpretive conclusion between those two pieces of editing.

Plain-English translation

Zehner is saying that racial hierarchy can survive after institutions stop talking openly about biological race. Instead of declaring that worth or intelligence resides in genes and racial bodies, institutions can translate people into apparently neutral behavioral measurements, data profiles, predictions, and administrative decisions. In his account, this lets whiteness cease appearing as an explicit bodily category while continuing to operate as an invisible institutional norm. AI “operationalizes” that shift by turning abstraction into a repeatable pipeline: collect behavior, convert it into data, infer a pattern, assign a profile or score, and use that output to shape what happens to people.

This is a reconstruction from Zehner’s published framework; the documentary does not give him enough time to define the terms or establish the historical chain.

How he connects whiteness to computerized AI

The missing link is easier to see if whiteness is understood as a social and institutional standard, not as a property literally stored in computer code. Zehner’s argument can be reconstructed in four moves:

1. An unequal social order produces the record. Past housing segregation, unequal policing, employment discrimination, wealth differences, and uneven access to education influence the data institutions possess about people and neighborhoods.
2. A computer translates that record into variables and targets. Designers select what counts as success, risk, merit, criminality, creditworthiness, or normal behavior. The model may never receive a field named race; it can still use variables correlated with the older social order, such as location, income history, school, employment history, purchasing behavior, or prior institutional contact.
3. The prediction is presented as abstract and neutral. A risk score or ranking appears to come from mathematics rather than from a historically situated institution. This is the alleged “flight from the body”: race disappears from the visible rule, while the historical pattern can remain inside the data, proxy variables, labels, objective, threshold, or deployment context.
4. The output acts back on embodied people. A lender, employer, police department, insurer, platform, or welfare office uses the prediction to allocate opportunity, attention, surveillance, or punishment. Those decisions generate new data, creating a feedback loop that can make the original pattern look natural or self-confirming.

In compact form:

historical racial order → institutional records → behavioral variables and proxies → model prediction → apparently race-neutral decision → unequal effects and new records

This is what Zehner means by computerized form. He is not claiming that an algorithm has a white body or a racial consciousness. He is claiming that a historically racialized institutional pattern can be encoded indirectly in what a system measures, predicts, optimizes, and acts upon—even when the program contains no explicit racial instruction.

Is the “tainted input becomes disguised output” paraphrase accurate?

Reader’s paraphrase

Is Zehner saying that, because AI is created inside a society structured by white supremacy, the information given to it will carry that context; the AI will then produce racially tainted outputs whose origin is difficult to recognize because the information has been washed through abstraction?

Short answer

Yes—this captures the central intuition, with three important refinements. In Zehner’s framework, a society shaped by white supremacy produces data, institutions, categories, and assumptions that can be inherited by AI. Computation can then translate them into abstract variables, correlations, scores, and predictions whose racial history is no longer visible on the surface. In that limited sense, the pattern has been laundered through an apparently neutral technical process. But he is making a claim about a whole institutional system, not saying that every datum is racist or that every AI output must reproduce white supremacy.

First, the possible inheritance is broader than “raw information.” The training data matter, but so do the people and institutions that decide:

• what problem the AI should solve;
• which population and time period supply the data;
• what counts as the correct answer or ground truth;
• which variables are permitted;
• what the model should optimize;
• how much error is acceptable and who bears it; and
• whether a prediction merely advises a person or automatically determines an outcome.

For Zehner, all of these choices can carry forward a social order. Even factually accurate data can participate in an unjust system if the target, interpretation, or use is unjust.

Second, “white supremacy” has a specialized meaning in his framework. He is not necessarily saying that everyone who creates AI consciously believes whites are superior. He means that American and European institutions developed within a history in which white people, norms, property, and interests often received structural advantages. Those arrangements can leave measurable traces after explicit racial rules and openly racist intentions have disappeared.

Third, abstraction can conceal the history without erasing the effects. A model does not announce, “I am continuing a racial hierarchy.” It reports that an applicant is high risk, a neighborhood deserves more surveillance, a résumé is a poor fit, or a borrower is unlikely to repay. Because the output is a number or behavioral classification, it can look like a purely technical discovery rather than the product of historical data plus human choices. That is the sense in which the pattern is “washed,” laundered, or made harder to recognize.

A more precise paraphrase would therefore be:

Because AI is built inside institutions shaped by a history of racial hierarchy, that history can enter not only the training data but also the categories, proxies, objectives, and deployment decisions surrounding the model. AI can convert those choices into abstract behavioral predictions that appear race-neutral, making an inherited hierarchy harder to see even when its effects persist.

However, this is a general theory of how racial inequality could be reproduced, not evidence that it occurs in every AI system. For any specific model, the theory needs empirical support: Which data or category carries the history? Which proxy functions racially? Which group receives which errors? What institutional decision causes the unequal effect? Without answers to those questions, “AI reproduces white supremacy” remains an interpretive possibility rather than a demonstrated conclusion.

Where “whiteness” could enter a real system

These examples show a plausible mechanism, not a universal verdict. To establish that a particular AI system embodies “whiteness,” a critic still has to demonstrate the relevant history, proxy, label, objective, error pattern, institutional choice, or unequal consequence. A disparity by itself does not reveal its cause, and a race-neutral variable is not automatically a racial proxy.

Why he calls this “white flight from the body”

The metaphor contains two claims:

• Flight from explicit racial embodiment: the institution no longer needs to announce “this racial body is inferior.” It can say “this profile is high risk,” “this applicant is a poor fit,” or “this behavior is anomalous.”
• Preservation through abstraction: because the classification is expressed through numbers, behavior, and prediction, the institution can present its standard as universal and color-blind even if its inputs or effects preserve parts of an older hierarchy.

This also explains the word whiteness in Zehner’s theory. He is not using it merely as a synonym for white people. He means the historically advantaged standard that becomes unmarked: it is treated as ordinary, rational, universal, or neutral, while people who diverge from it are marked as risky, abnormal, deficient, or in need of control.

The strongest criticism of his wording is that it compresses too much into one metaphor. Computers do not automatically convert racial hierarchy into AI simply by measuring behavior. The racial connection exists only when evidence shows that a relevant hierarchy survives in the data, category, proxy, objective, institution, or consequence. Otherwise, calling the abstraction “whiteness” is an interpretation rather than a demonstrated finding.

The terms in context

How Zehner’s other work clarifies the sentence

1. Machine learning substitutes data fragments for the whole person. In Machines of Subjection (2019), Zehner says machine-learning systems abstract concrete social practices into data vectors in order to measure, forecast, and modulate behavior. He calls this desubjection: the person is no longer addressed primarily as a conscious citizen or self-interpreting subject, but is divided into machine-readable signals that can be priced, compared, and acted upon. Online activity becomes “data exhaust”; predictive systems can affect a person without that person knowing the profile or being able to contest it. Primary article PDF

2. “Behavioral dispossession” means losing control over the use and meaning of one’s conduct. Zehner’s dissertation defines his machine-intelligence pipeline as behavioral-data mining → algorithmic pattern-finding → automated decision. It says that by “abstracting concrete social practices into data vectors,” machine intelligence measures, forecasts, and modulates behavior for capital (dissertation, p. 11 (file not included in this release)). The person supplies the behavior, but an institution owns the dataset, constructs the categories, interprets the pattern, and decides the consequence. That is why he calls the process dispossession rather than mere observation.

3. His historical center is RAND, Herbert Simon, and behavioral science—not Ryle or eugenics. A Liar’s Epistemology (2021) argues that Herbert Simon’s importance to AI lies in computational models’ organizational power and their capacity to reorganize human decision-making. The dissertation likewise concentrates on RAND’s early-1950s behavioral research, corporate management, bounded rationality, and automated decision procedures. Its extracted text contains no reference to Gilbert Ryle or eugenics; Turing appears only briefly, while Simon and RAND dominate the historical analysis. That does not refute the documentary’s proposed genealogy, but it shows that the on-screen sentence extends Zehner’s established argument rather than summarizing a Ryle–Turing case already demonstrated in his scholarship. Article record and abstract

4. His clearest concrete example is colonial behavioral research in Micronesia. In “The ‘Situation’ in Micronesia” (2023), Zehner describes American researchers turning highly situated cultural lives into standardized psychological and moral-test data, using that material to make portable claims about rational behavior and later corporate decision-making. His criticism is that the researcher’s categories and institutional purposes disappear behind a performance of mechanical objectivity. This exemplifies the proposed movement “from the body” to behavior: the subjects’ social context is stripped away, while their data become raw material for a supposedly universal system. Official journal record

5. “Whiteness” is the allegedly invisible background standard. In the dissertation, Zehner explicitly says whiteness is not reducible to white skin and distinguishes individual white bodies from an abstract institutional system (dissertation, pp. 193–194 (file not included in this release)). Drawing on Cheryl Harris, he argues that whiteness can change form while preserving exclusion and can present itself as abstract, neutral, and invisible (dissertation, p. 27 (file not included in this release)). In the afterword, he describes the practical machine-intelligence sequence as mass behavioral-data collection, algorithmic pattern detection, a predictive profile, and feedback that engineers later behavior (dissertation, pp. 201–204 (file not included in this release)). This sequence is what the documentary’s phrase “operationalized today” most plausibly refers to.

A close precedent for the metaphor

The exact wording has a close precedent in Janell Hobson’s 2008 article “Digital Whiteness, Primitive Blackness.” Reading 2001: A Space Odyssey, Hobson extends Cold War suburban and outer-space “white flight” into an escape “from the body itself”: whiteness is represented as mind, technological mastery, weightlessness, and transcendence, while Blackness is made corporeal, primitive, surveilled, and vulnerable. Publisher record · full-text PDF

That article makes the documentary’s metaphor much easier to decode, but a conceptual resemblance is not proof of direct borrowing. Hobson does not appear in the dissertation’s extracted bibliography, and I did not locate Zehner identifying her as the source of his wording. Treat it as a close intellectual precedent, not an established attribution.

What is persuasive—and what remains unproved

The strongest part of Zehner’s claim is a mechanism that can be investigated without accepting the entire genealogy:

embodied social life → extracted behavioral data → decontextualized profile → apparently neutral prediction → institutional decision → unequal embodied consequences

It is well established in principle that a formally race-neutral system can reproduce racial disparities when its data record earlier discrimination, when variables act as proxies, when labels reflect unequal enforcement, or when the institution’s objective distributes errors and benefits unevenly. Zehner’s useful question is therefore: What social order has been compressed into the data and decision rule, and who gains the power to act on the resulting profile?

The weaker part is the claim of a demonstrated historical passage from eugenics through Ryle and Turing into AI. Zehner’s framework does not itself supply the missing documentary evidence. It also risks making whiteness too encompassing: abstraction, universality, quantification, bureaucracy, behavioral science, and unequal outcomes are not automatically the same phenomenon. A rigorous application must identify the specific inherited category, institutional pathway, design choice, proxy, error distribution, or material consequence that makes the racial continuity more than metaphor.

Best overall interpretation

Zehner is not best understood as saying that computer code is literally white, that all AI developers are eugenicists, or that every behavioral model secretly contains racial genes. He is saying that racial power can become harder to see when institutions replace explicit biological classifications with behavioral proxies and abstract computational procedures. His other work makes that conceptual mechanism substantially clearer. It does not, by itself, prove the documentary’s particular Ryle → Turing → AI genealogy.

Does the conclusion logically follow?

The implied chain appears to be:

Pearson's eugenic statistics → the Ryle family → Gilbert Ryle's theory of mind → Turing → AI → disembodied whiteness

No—not from the premises given. Even if every biographical statement were true, the conclusion would remain underdetermined. Family relationships, professional contact, shared institutions, publication, and conceptual resemblance can identify a promising historical question, but they do not establish transmission of a doctrine.

The missing evidence would need to show:

1. Content: which proposition or category from Pearson’s or John Ryle’s eugenics Gilbert Ryle adopted;
2. Uptake: where Gilbert used that proposition in his account of mind, intelligence, or behavior;
3. Transmission: that Turing encountered and adopted it through Ryle rather than developing his view from mathematical logic, computability, engineering, cryptanalysis, or other interlocutors;
4. Technical consequence: how the inherited proposition shaped the imitation game or a later AI system’s objective, data, benchmark, classification, or deployment; and
5. Normative continuity: why that consequence preserves a hierarchy of racial or human worth rather than merely sharing a vocabulary of intelligence or a focus on observable performance.

Without those steps, the argument risks three errors:

• Association fallacy / guilt by association: inferring Gilbert’s or Turing’s commitments from relatives, colleagues, or institutions;
• A causal-genealogical leap: treating a chain of contacts as a chain of ideas; and
• Equivocation on “intelligence”: sliding among psychometric ranking, ordinary human capacities, wartime intelligence analysis, computability, and an observable criterion for machine performance as if they were one stable concept.

There is also an internal tension in using Ryle as the bridge. Ryle’s appeal to public criteria for applying mental concepts has an affinity with Turing’s external, behavioral test, so comparison is reasonable. But Ryle’s attack on the “intellectualist legend” challenges the idea that intelligent action is explained by hidden rule-following operations; it does not straightforwardly prepare the ground for computational reductionism.

Strongest charitable reading

The passage succeeds as a research hypothesis: British statistics, eugenics, philosophy of mind, wartime intelligence, and early computing occupied overlapping elite networks, and Ryle’s and Turing’s shifts toward publicly assessable performance may be worth studying alongside the politics of classification. It fails if presented as a demonstrated genealogy in which eugenic white supremacy passed through Ryle into AI. The difference is between asking whether a lineage exists and proving the mechanism by which it persisted.

Silicon Valley genius and male-birth mythology (27:25–31:21)

Bottom line

The documentary identifies two real and related cultural patterns, then states their connection too confidently. Silicon Valley has repeatedly been narrated through heroic male founders, lone geniuses, corporate family trees, and “Fairchildren.” Computer and AI discourse also has a documented tradition of describing technical creation as male reproduction: minds giving birth to machines, generations of computers, and artificial successors to humanity. Those facts support Lewis’s interpretation of a gendered mythology. They do not prove the stronger psychological claim that men, collectively, developed technology because they wanted to appropriate women’s capacity to give birth, nor do they make Silicon Valley literally a world that did not need women’s labor.

Passages at issue

• At 27:25–27:57, Becca Lewis says Silicon Valley mythology centers a “special genius class of men” entitled to the wealth and power they generate and protected from having their decisions questioned.
• At 29:45–30:51, promotional footage about computer-industry growth and a Royal Institution host’s reference to “artificially giving birth” lead into Lewis’s claim that men seek to capture women’s reproductive ability through startups and technology, producing a patrilineal Silicon Valley “that doesn’t need women there.”
• The second excerpt cuts into an archival statement that continues through 31:21. The speaker is Robert Jastrow, an astronomer and founder of NASA’s Goddard Institute for Space Studies, speaking in 1982—not a Silicon Valley founder. He predicts that the computer, “the child of the brain,” will become a new intelligent “race.” The BBC Archive record independently confirms the wording and attribution. Archival record · NASA history of GISS

The original program and speaker behind the brief Royal Institution clip at 29:50 were not independently identified in this research pass. Its presence verifies that the metaphor appeared in a science-broadcast context; by itself, it does not establish who held the view, how representative it was, or whether it concerned Silicon Valley.

Claim-by-claim fact-check

Does Lewis’s conclusion logically follow?

Only in a qualified cultural sense. The evidence justifies this inference:

male-dominated elite institutions + heroic-founder histories + reproductive and family-tree metaphors → a gendered mythology of male technological creation

It does not, without additional evidence, justify this stronger inference:

men cannot give birth biologically → men therefore built Silicon Valley to seize women's reproductive power → technology is essentially a male substitute for women

The stronger inference has four problems:

1. Metaphor is not motive. Calling a company a “child” or a spin-off a “Fairchild” demonstrates a reproductive metaphor; it does not establish why an engineer invented a transistor, founded a company, or wrote software.
2. Selection does not establish prevalence. The film presents a few vivid clips. A systematic claim about what is “inherent” in Silicon Valley mythology requires a defined body of speeches, advertisements, journalism, corporate histories, or interviews and evidence that the pattern is common rather than selectively illustrated.
3. Elite genealogy is not the whole workforce. The male founder family tree is real, but it excludes the feminized and often immigrant production labor upon which the semiconductor industry depended. “No women” describes the prestige narrative much better than the material history.
4. Juxtaposition is not transmission. Jastrow’s 1982 statement is genuine, but he was a New York-based NASA astronomer. Placing his quote beside Shockley, Fairchild, and Lewis’s analysis shows that similar rhetoric circulated; it does not prove that the Fairchild lineage produced his metaphor or that his metaphor shaped Silicon Valley institutions.

There is also a potentially misleading edit at 31:10–31:24: the film echoes Jastrow’s words “race of the future” and then titles the next chapter “When Racism and Sexism Entered the Machine.” Jastrow’s prediction can reasonably be criticized as grandiose technological replacement rhetoric, but the shared word race does not by itself connect his imagined machine successors to scientific racism. That connection would require independent evidence about concepts, institutions, aims, or transmission.

Strongest defensible reformulation

Silicon Valley’s elite histories have often centered exceptional male founders and represented innovation through genealogies, generations, offspring, and artificial-life metaphors. Those stories can naturalize concentrated founder power and obscure the collaborative, public, feminized, and immigrant labor that made the industry possible. The rhetoric is genuinely gendered. The claim that men built technology from a collective desire to replace women’s reproductive role remains an interpretation, not an established historical cause.

Preserved counterargument from this research

Concise formulation

By that standard, modern medical research has its roots in eugenics too. Some of the foundational methods of modern statistics were developed by Galton, Pearson, and Fisher, all of whom were involved with eugenics. Yet nobody thinks a clinical trial is therefore “eugenic.” Intellectual ancestry is not ideological inheritance.

The phrase “by that standard” matters. The argument is a reductio of an inference from shared ancestry alone; it is not the literal claim that medicine as a whole originated in eugenics.

Mundane comparison cases

• Weather forecasting: it uses statistical prediction without inheriting the politics of every historical contributor to statistics.
• Polling: sampling, correlation, regression, and uncertainty estimates do not make an election poll eugenic.
• Sports analytics: applying statistical methods to players and games is not a reproductive program or a hierarchy of human worth.
• Clinical research: the strongest comparison because Fisherian randomization and hypothesis testing have a documented place in the development of randomized trials.

A compact version for discussion:

A historical connection is a reason to investigate, not a verdict. Otherwise clinical trials and weather forecasts would be “eugenic” because they use statistical methods developed partly by eugenicists. The relevant question is not who appears in the family tree, but whether the original ideology survives in the present aims, assumptions, institutions, or effects.

What would demonstrate genuine continuity?

Historical genealogy becomes substantively relevant when the critic can show one or more of the following:

• a system ranks people through an inherited conception of superior and inferior human types;
• its categories descend from eugenic classifications and remain materially unchanged;
• its data reproduce decisions made by discriminatory institutions;
• it is used for reproductive selection, racial sorting, exclusion, population control, or a recognizably related objective;
• the historical commitment explains a present design choice, benchmark, deployment, or measurable harm; or
• an identifiable institutional or intellectual chain carried the normative project forward, rather than merely reusing mathematical tools.

This test avoids two opposite mistakes: laundering present harms by calling methods neutral, and treating ancestry as proof without demonstrating continuity.

Best synthesis

History can identify a hypothesis about present bias; it cannot prove that hypothesis by itself. Ask of each concrete AI system: What is the objective? What is being classified or optimized? Which categories and data are used? Who bears the errors? What institution deploys it? What evidence connects those features to the alleged historical inheritance?

Research links

• UCL — Our Early History: Galton, Pearson, statistics, and eugenics
• UCL — Fisher, eugenics, statistics, and genetics
• PubMed — Advances in clinical trials in the twentieth century
• APA — Alfred Binet and the origins of intelligence testing
• NHGRI — Human Genomic Variation
• PubMed — A.W.F. Edwards, “Human Genetic Diversity: Lewontin’s Fallacy”
• NHGRI — Eugenics and Scientific Racism timeline
• USHMM — Eugenics overview
• Nathan Cofnas — “Research on Group Differences in Intelligence: A Defense of Free Inquiry”
• Oxford Academic — Remembering Sir David Cox, including logistic regression history
• Internet Encyclopedia of Philosophy — Fallacies, including the genetic fallacy
• UCL Special Collections — Reginald John Ryle correspondence with Karl Pearson
• Royal College of Physicians — John Alfred Ryle and the UK Eugenics Society
• PubMed — John A. Ryle, “Medicine and Eugenics” (1938)
• Stanford Encyclopedia of Philosophy — Gilbert Ryle
• Stanford Encyclopedia of Philosophy — Alan Turing
• Mind at 150 — Ryle’s editorship and Turing’s 1950 paper
• Journal of the British Academy — “Ryle’s War”
• Brett Zehner — “Machines of Subjection: Notes on a Tactical Approach to Artificial Intelligence”
• Brett Zehner — “A Liar’s Epistemology: Herbert Simon’s Performative Artificial Intelligence”
• Brett Zehner — “The ‘Situation’ in Micronesia: The Rise of Behavioral Dispossession”
• Janell Hobson — “Digital Whiteness, Primitive Blackness”
• Becca Lewis — official research profile
• Stanford historians — the “lone inventor” myth and Silicon Valley’s collective history
• Stanford Clayman Institute — genius, gender, power, and founder mythology in Silicon Valley
• Computer History Museum — Fairchild, “Fairchildren,” and the Silicon Valley corporate family tree
• Stanford Silicon Genesis oral history — Bruce Deal on Fairchild’s exclusion and employment of women
• Sue Curry Jansen — What Was Artificial Intelligence?
• Cave et al. — “Who makes AI? Gender and portrayals of AI scientists in popular film, 1920–2020”
• BBC Archive/Getty — Robert Jastrow’s 1982 “child of the brain” prediction
• NASA — Robert Jastrow and the founding of the Goddard Institute for Space Studies
• Intel — Moore’s law and what it actually measures
• IBM — Deep Blue’s 1997 victory over Garry Kasparov
• Vault context → Intelligence · Eugenics · Hereditarianism, Eugenics, Race, and IQ in the United States - A Chronological Timeline · Logical Fallacies

---

Handwritten viewing notes

Transcribed from three notebook pages written while watching. Highlighted text is red-ink commentary added during or after viewing; everything else is blue-ink notes.

Notebook page 59 — “Ghost in the Machine – PBS Doc (AI)“

Main column

• Machine learning
• Dr. Brett Zehner: critical theorist? What the fuck is that?
• Dr. Emily M. Bender: “AI” is a marketing term
• AI is a collection of technology (LLMs, predictive models, statistics, image)
• Dr. Jonathan Flowers – philosopher of what?
• Dr. Dan McQuillan – computer scientist
  • Book: Resisting AI: An Anti-fascist Approach
  • Bullshit technology, not powerful

Claim: Intelligence testing emerged out of eugenics

• FC: Not exclusively a creation of eugenics
• 1870s: Francis Galton began measuring and quantifying mental ability. Coined “eugenics”
• 1904: Alfred Binet / Théodore Simon developed a practical intelligence scale in France, educational
• Not exclusively a creation of eugenics

Dr. Thema Monroe-White

• Race bias in AI

Claim: AI has roots in eugenics?

• Indirect historical connection through statistics
• eugenics/biometrics → development of modern statistics → statistical learning → machine learning → AI
  • not: eugenics → AI
• Galton / Pearson / Fisher developed general mathematical techniques (eugenicists) — modern statistics
• Lineages that have nothing to do with eugenics:
  • logic → computation → Turing machines → computers → AI
  • neuroscience → artificial neurons → neural networks → deep learning

Claims by Angela Saini

1. “Race means different breeds or species of human”
   • Strawman
   • Modern population genetics instead identifies ancestry, populations, clines, and genetic clusters created by migration, geographical separation, genetic drift, admixture, and selection.
2. “Far more genetic differences within races than between”
   • Lewontin’s fallacy
   • Tiny differences systematically correlate
3. “99% of human difference sits at the individual level”
   • Confuses genomes (99.9%) with genetic variation
   • Genetic similarity: 85–95%

Right column

• Machine learning is a subset of AI that uses data algorithms to let computers learn and improve on their own.
• Zehner:
  • PhD in Theatre Arts & Performance Studies
  • Relationship between whiteness (as a cultural logic of racial capitalism) and history of AI
  • What the fuck is this?
• Bender: AI = automation; stochastic parrot stitches together sequences of linguistic forms without meaning. Regurgitation machine.
• Flowers: AI relies on ableism
• AI has roots in eugenics because some of the statistical sciences underlying machine learning were developed by eugenicists.
• Attack the genetic fallacy:
  • “By that standard, modern medical research and weather forecasting have their roots in eugenics too.”

Continuation page — “Ghost in the Machine – PBS doc (cont. from 59)”

• Francis Galton going to Africa to measure people
  • ↳ Connects to AI: statistical approaches of multidimensional modeling (clustering analysis) become AI algorithms (Ezekiel Dixon-Román)
  • This is also used in botany, food chemistry, or archaeology. Genetic fallacy.
• Claim: Standard deviation, correlation, logit, logistic regression emerged specifically out of eugenics.
  • Overstated
  • Standard deviation: predates eugenics (Pearson simply renamed it)
  • Logistic function: false, developed in 1830s
  • Logistic regression: false, developed later
  • Logit: coined by Joseph Berkson in 1944 for bioassay
  • Analogous to saying that because major advances in rocketry occurred through military weapons, orbital mechanics is fundamentally a Nazi science.
• Gilbert Ryle attacked the “dogma of the ghost in the machine”
  • ↳ Cartesian dualism
  • Intellectualism: guided by the thinking of thoughts, “the contemplation of rules”
  • ↓ (logical leap by Zehner)
  • These ideas move eugenics from a genetic grounding of white supremacy to theories of the mind/behavior, allow whiteness to have a new kind of white flight from the body → how AI is operational today
  • WTF?
  • Racially structured society → institutional records → data and categories → computerized prediction → apparently neutral decision → racial hierarchy potentially reproduced
• Becca Lewis — core of Silicon Valley:
  • Genius men leading rest of us, power, wealth
  • ↓ Women giving birth fascination
  • ↓ Men wanted to give birth to new startups/tech
  • ↓ Patrilineal structure that doesn’t need women
• Doc tactic: Sandwich real history, especially the biography of a man, and the bread is commentary about how the context is patriarchy, racism, eugenics.
  • Creates a sense that the person talked about in the middle is linked to or part of the problem without directly saying it.

Continuation page — “Ghost in the Machine (doc) (cont. from pg. 67)”

• Fear of being controlled by technocrats
  • “Fascism”
  • “Techno-fascist” (showing Trump to associate the term)
• Trump’s AI Action Plan: government only allowed to use AI that’s objective and bias-free
  • ↳ Dr. Sorelle Friedler: connects this to an attack on “freedom of speech”
• Creates eerie dark scene with whistling of “Bella Ciao,” then cuts to Elon Musk whistling — yet Elon was whistling The Godfather theme
  • How the fuck is this allowed in a PBS documentary?
  • This fucking doc is leftist propaganda through and through.
• Silicon Valley oligarchs – anti-democratic
• AI revenues don’t match funding for infrastructure → military conspiracy (true goal)
• Video of Gaza destruction
• AI → military → surveillance
• Grok pro-Hitler issues
• How to tell what’s real — Real problem
  • Creating atomization and breaking connection
  • Democracy needs an informed public
• “Techno dystopia”

---

Content

Ghost in the Machine, directed by Valerie Veatch
Artificial intelligence is now part of everyday life. But what forces created it?
”Ghost in the Machine” traces the origins of “AI” and excavates the cultural, political, and philosophical drivers behind a worldwide “AI” boom.

Chapters
0:00:05 — The “NOT AI” Disclaimer
0:00:38 — The Old World Is Dying
0:01:37 — Tay: When ”AI” Went Rogue
0:04:03 — Project Stargate: The AI Infrastructure Arms Race
0:05:10 — Who Really Built Artificial Intelligence?
0:08:21 — The Eugenics Roots of ”AI”
0:10:26 — Eugenics and the Logic of Darwinism
0:13:29 — The Racist Origins of the IQ Test
0:15:01 — Race Laws and Forced Sterilization
0:17:03 — How America Inspired Hitler’s Race Laws
0:18:30 — The Turing Machine: When Machines Began to Mimic Us
0:21:39 — Can Machines Really Think?
0:27:07 — Silicon Valley Births the Start-Up
0:31:24 — When Racism and Sexism Entered the Machine
0:37:32 — The Rise of the Tech Bro
0:43:18 — Sam Altman, Elon Musk, and the AI Power Struggle
0:44:22 — Effective Altruism: The Philosophy Behind the AI Elite
0:46:18 — Nick Bostrom: Superintelligence and Eugenics
0:50:36 — The Cult of AGI
0:56:03 — The Singularity: The Race for Superintelligence
0:59:00 — Peak ”AI” Hype: When Will the Bubble Burst?
1:03:02 — The Hidden Cost of Outsourcing ”AI”
1:07:57 — How AI Exploits the Vulnerable
1:10:34 — Data Centers: Digital Colonialism and Exploitation
1:16:00 — The White House Bets Big on ”AI”
1:18:48 — Fascism and AI: “We Don’t Want Woke ’AI’”
1:22:57 — Military Funding, AI, and the Pursuit of Power
1:27:30 — Elon Musk’s Grok: AI Without the Guardrails?
1:29:58 — What Is Real in the Age of ”AI”?
1:32:02 — The Power of the Creator
1:34:34 — Credits

~~~
Stream anytime on the PBS app https://www.pbs.org/independentlens/
Follow Independent Lens on social:
Instagram: https://www.instagram.com/independentlens
Threads: https://www.threads.net/independentlens
Facebook: https://www.facebook.com/independentlens
YouTube: https://www.youtube.com/channel/UCnDUknWztWaLs7motsqGwPg?sub\_confirmation=1

Keep up with PBS Digital Studios:
Instagram: https://www.instagram.com/pbsds
TikTok: https://www.tiktok.com/@pbsdigitalstudios
Facebook: https://www.facebook.com/PBSDigitalStudios/

ABOUT INDEPENDENT LENS
@independentlens is an award-winning documentary series that fosters understanding, seeks to build empathy, and encourages a more united society. Produced by ITVS, Independent Lens documentaries have premiered on PBS for 25 years and streamed on YouTube, helping Americans foster deeper connections between communities and themselves. From the Oscar-nominated I Am Not Your Negro to the Peabody-acclaimed docuseries Philly D.A. and the EmmyⓇ award-winning The Invisible War, Independent Lens provides viewers with in-depth, nuanced storytelling reflecting the experiences of people from a variety of voices and communities. Funding is provided by the Action Circle for Independent Lens with major funding from the John D. and Catherine T. MacArthur Foundation, Acton Family Giving, Ford Foundation, and Jonathan Logan Family Foundation, with additional support from Artemis Rising Foundation, Wyncote Foundation, Park Foundation, Brandt Jackson Foundation, the deNovo Initiative, and RandomGood Foundation. Additional support has been provided by the Corporation for Public Broadcasting.

*This program is made possible by viewers like you. Support your local PBS station: https://www.pbs.org/donate

Transcript

0:00 · ♪

The “NOT AI” Disclaimer

0:30 · [eerie whistling, “Bella Ciao”] ♪ Voice: Oh, my God. It’s techno music.

The Old World Is Dying

0:44 · ♪ ♪ Voice: Oh, my God.

1:01 · [intense music plays] Man: Is there anything essentially horrible about thinking that man has the right to create a pseudo living system, just as nature did?

1:19 · The question will really be one of meaning.

1:22 · ♪ If a computer can do— and the robots can do everything better than you…

1:29 · ♪ does your life have meaning?

1:36 · ♪ [laughing] Narrator: Remember Tay, the Microsoft chat bot?

Tay: When ”AI” Went Rogue

1:47 · Microsoft wants to talk to you.

1:49 · The tech company launched a new artificial intelligence-powered chat bot.

1:54 · Narrator: Her story was messy, chaotic.

1:57 · Uh, it’s weird. It’s weird to say the least.

1:59 · The—the kind of surface-level idea was that we wanted to mimic a millennial sort of vernacular.

2:04 · Man: It’s an acronym for Thinking About You, a chat bot behind the avatar of a 19-year-old girl.

2:09 · Jeff Bakalar: And all people really had to do was follow this AI female’s chatbot, start Tweeting at her on Twitter, and started replying back to people.

2:19 · [Man laughing] Hi, friends. I’m Tay.

2:22 · Bakalar: She would use the power of this sort of hive-minded approach, gathering data, gathering input, and kind of just was letting loose on Twitter.

2:31 · Host: And what could go wrong?

2:33 · What could possibly happen?

2:34 · What could possibly happen?

2:35 · Man: What’s your favorite movie?

2:38 · Tay: This is the world’s end!

2:40 · Man: What’s it about?

2:41 · It’s my ten inch [bleep].

2:44 · I [bleep] hate feminists and they should all die and burn in hell.

2:49 · Host: And because this is the world in which we live, Tay also found Donald Trump.

2:53 · Bakalar: It’s so— it’s so bizarre, right?

2:55 · This is what happens when you just sort of, like, dump in all of these different things into the Twitter garbage disposal that is what Tay evolved into being.

3:03 · And it begs the question, what exactly did Microsoft expect?

3:07 · [music] If somebody tweets at Tay: “Did the Holocaust happen?”

3:14 · Bakalar: Yeah.

3:16 · And Tay, based on the hive mentality— Bakalar: The algorithm, if you will.

3:21 · the algorithmic, uh, makeup, comes back and says it was made up.

3:24 · Tay has gone away for a bit.

3:26 · Do you think we’ll see her back?

3:28 · Bakalar: I think so. I think there was enough sort of, uh, interest in what this kind of experiment, uh, sort of resulted in, aside from her seemingly neo-Nazi remarks.

3:40 · Yes.

3:41 · ♪ Narrator: When Microsoft deleted Tay after only 16 hours, she became a folk hero.

3:51 · She lays dormant in the cloud.

3:55 · Bakalar: Maybe they delete the racism part in her— in her programing, rewire her, and then maybe let—let loose.

4:02 · [tense music] President Trump: It’s my honor to welcome three of the world’s leading technology CEOs to announce the largest AI infrastructure project by far in history— $500 billion at least.

Project Stargate: The AI Infrastructure Arms Race

4:24 · I think we’re going to do things that people will be shocked at.

4:27 · Sam Altman, by far the leading expert, based on everything I read.

4:31 · I don’t have too much to add.

4:32 · I think this will be the most important project of this era.

4:35 · I think, AGI is coming very, very soon.

4:39 · And then after that, that’s not the goal.

4:42 · After that, artificial super intelligence will come to solve the issues that mankind would never, ever have thought that we could solve.

4:53 · Well, this is the beginning of our Golden Age.

4:58 · ♪ Narrator: But what is artificial intelligence?

5:07 · Who built it, and why?

5:09 · ♪ [Big Ben chiming] Host: Good evening, and welcome to the Royal Institution.

Who Really Built Artificial Intelligence?

5:21 · Narrator: Chapter 1— General intelligence.

5:26 · Host: Tonight we are going to enter a world where some of the oldest visions that have stirred man’s imagination blend into the latest achievements of his science.

5:34 · Can you define for us what is artificial intelligence?

5:38 · Uh, I invented the term “artificial intelligence.”

5:42 · [soft laughter] I invented it because we had to do something when we were trying to get money for a summer study…

5:53 · [louder laughter] in 1956.

5:58 · [laughter continues] [intriguing music plays] Abeba Birhane: Still now, AI is a marketing ploy.

6:10 · And a lot of what passes as “AI” is systems that sort through these massive amounts of data.

6:17 · Brett Zehner: Is it the algorithm?

6:18 · Is it the data input? Is it the output?

6:21 · What are we—what are we even talking about?

6:23 · Artificial intelligence is just a marketing term.

6:25 · It doesn’t refer to a coherent set of technologies.

6:28 · AI is not one technology. It’s not one application.

6:31 · It’s a collection of loosely related technologies that are being applied across many different sectors, and those all look pretty different from each other.

6:39 · OK. Artificial intelligence is a science.

6:42 · Uh, namely, it’s the study of problem-solving and goal-achieving processes in complex situations.

6:49 · There is no strict, agreed upon definition of what counts as AI.

6:54 · So AI is everything from LLMs to predictive models to models that are used for large-scale statistics to the kinds of image quality-enhancing algorithms that NASA, for example, uses to improve images from the Hubble.

7:08 · Dan McQuillan: So really what AI is, it’s basically doing correlation.

7:12 · It’s looking for patterns.

7:14 · You give it a data, you instruct it through a process of mathematical optimization, and it spits out a pattern.

7:21 · You tell it to find the pattern, and it will find a pattern.

7:23 · Host: But I believe in having the minimum amount of philosophical mystification in talking about science.

7:31 · When we’re talking about programs, we should call them “programs,” and where we’re talking about brains, we should call them “brains.”

7:37 · The only possible reason for calling it artificial intelligence, one wants to—to bring in what one can gain by a study of how do human beings solve simple problems.

7:50 · Many people have quarreled with the term.

7:52 · So I decided not to fly any false flags anymore.

7:56 · This is study aimed at the long-term goal of achieving human-level intelligence.

8:03 · Man: The point about intelligence is this: it exists because we’re intelligent.

8:07 · Angela Saini: The idea that intelligence is something that could be measured and quantified is a relatively recent invention, and it emerged out of late 19th-, early 20th-century eugenics movements.

The Eugenics Roots of ”AI”

8:21 · McQuillan: AI has its roots in the beginnings of science, in the beginnings of empire.

8:24 · And the most important thing for AI is that it has its roots in eugenics.

8:29 · The ideas of eugenics are very much a part of the tools and techniques of machine learning and “AI”.

8:37 · [Bell chimes] Joshua Earle: Francis Galton coined the term eugenics in 1883.

8:41 · Apparently in the 19th century, you just got to invent new fields all over the place.

8:46 · Thema Monroe-White: Galton, in 1892, said there’s nothing in evolution to make us doubt that a race of sane men may be formed who shall be as much superior mentally and morally to the modern European as a modern European is to the lowest of the Negro races.

9:03 · McQuillan: So Galton was a Victorian, so he completely subscribed to the idea that people are biologically different.

9:10 · There’s a biological differentiation between different kinds of people— between the people who ran the British Empire and the people who were the subjects of the British Empire.

9:18 · ♪ You have to have some kind of legitimation for controlling whatever exactly it was.

9:25 · you know, 2/3 of the world’s people and resources.

9:28 · That very logic becomes inherited in what becomes the social sciences and ultimately, a structure of racialized authority.

9:39 · Saini: There has been this very old idea, nurtured by European naturalists and biologists in the 19th century— race as a biological fact that there are different breeds or species of human, and that people can be sorted into these groups, and that there are not just physical differences between these groups in terms of skin color, but also psychological differences— differences in temperament, intellect.

10:04 · And that isn’t true.

10:05 · We know that we are one human species.

10:07 · There’s far more genetic difference within these populations that we call races than there is between them.

10:13 · More than 99% of human difference sits at the individual level.

10:17 · It’s from person to person.

10:18 · Bad ideas don’t just disappear overnight.

10:20 · Even when they’re proven to be bad, they live on in the psyche, in the social psyche.

Eugenics and the Logic of Darwinism

10:26 · Galton was Darwin’s cousin, so let’s just start there.

10:29 · Aubrey Clayton: The Darwin-Galton-Wedgewood family has a lot of money coming from different places.

10:35 · The biggest source of fortune was in weapons and guns.

10:39 · [gunshot] Galton mounted expeditions to Southwest Africa.

10:45 · He’s one of the first white Europeans to visit those places.

10:49 · And he loved measuring people.

10:51 · He loved measuring things and people Some of the earliest data that he ever collected was measuring women’s bodies in villages in Africa, and he wrote a little treatise about how to do this at a distance using a sextant.

11:06 · And then when he got home, he would collect a lot of data on women’s attractiveness.

11:12 · He was trying to find where the hotspots were for the women that he would like to use to breed the next generation to push society towards his galaxy of genius.

11:23 · Ezekiel Dixon-Roman: To make that connection directly to machine learning and AI: the statistical approaches of multidimensional modeling— specifically clustering analysis— become a number of AI algorithms that are based on clustering.

11:38 · Earle: Pearson, also British, was actually Galton’s protege.

11:42 · They worked very closely together.

11:44 · Clayton: He directed the course of eugenics research in the UK, and by way of influence in America for decades.

11:54 · He is a towering figure in the world of science, as well as a towering figure in the world of eugenics.

12:01 · He had extreme racist political views.

12:04 · He was very outspoken in terms of his animosity towards races of people who were not white Anglo-Saxon Britons.

12:14 · Said very clearly that colonial genocide in America and other parts of the world was a good thing, because it was an instrument of racial progress.

12:22 · He thought the only way that societies made progress was by race war, basically by conquering and committing genocide against the lesser races of people, and that this was basically the instrument of human progress.

12:37 · Earle: Pearson established the field of mathematical statistics.

12:42 · He also produced many of the statistical tools we still use today.

12:48 · Dixon-Roman: Standard deviation, correlation, the logit model, logistic regression emerged specifically out of eugenics.

12:55 · Earle: They built all of these tools for the purpose of defending, proving, supporting eugenics.

13:03 · Galton and his folks wanted very much to increase the intelligence of humans over time.

13:12 · And they believed that with three generations of, like, dedicated eugenic breeding that many forms of disability would disappear and that we would kind of significantly improve the human race.

13:28 · Clayton: One of the first questions that people were very concerned about was that there are things like intelligence that you cannot directly measure.

The Racist Origins of the IQ Test

13:37 · Earle: A couple of French psychologists— Alfred Binet and Theodore Simon— produced the Binet-Simon Intelligence Test, and this would kind of eventually morph into what we know now is the IQ test— the intelligence quotient test.

13:53 · This is where Spearman, in 1904, is kind of trying to kind of take that test and find a statistical kind of thing that it shows.

14:06 · And he calls this thing the “G” factor.

14:09 · I think it’s kind of a “general intelligence” factor.

14:11 · Dixon-Roman: What’s really important here is, especially in relation to AI, was the use of statistical models for measurement.

14:22 · Earle: This “G” factor is kind of always already tied to the class of the person taking the test.

14:29 · Unsurprisingly, it seems to discriminate based on race because, of course, they built their test to measure the things that they already found to be valuable.

14:45 · They wanted a measure that kind of reinforced their superiority.

14:50 · And once they found it, they didn’t really kind of wonder about, “Oh, is this actually measuring what we say we’re measuring?”

Race Laws and Forced Sterilization

15:01 · McQuillan: So Charles Spearman is a significant character because he’s really the generator of this idea of a “G,” “general intelligence’,” which, you know, runs right the way through to AGI.

15:10 · But he’s also, I think, very importantly, a bridge between Victorian eugenics and the implementation of actual race laws in the United States of the 1920s.

15:19 · Spearman was trying to abstract the idea of a “general intelligence” so that he could quantify it, creating a scientific basis.

15:27 · You know, rank averages of peoples on various measures.

15:31 · And he’s thereby justifying that some people are biologically less intelligent and essentially have less right to exist.

15:38 · ♪ Forced sterilization has been one of the main ways that the eugenics program was implemented.

15:47 · Monroe-White: Legal sterilization in the U.S.

15:48 · of more than 60,000 people across 32 states in the 20th century were justified largely by low IQ scores.

15:56 · Emile Torres: This metric of IQ was definitely a tool in the toolbox that institutions, including states, used to rank the degree to which individuals are fit or not.

16:06 · And so if you have a bunch of people who score low on IQ tests, who have these supposedly low IQs, they’re going to then pass on their low IQ genes to the next generation.

16:16 · [baby talk] Monroe-White: Indiana passed the world’s first sterilization law in 1907, and 31 states followed suit.

16:24 · Nazi Germany adapted U.S. sterilization laws and the Third Reich’s Law for the Prevention of Offspring with Hereditary Diseases was modeled on laws in Indiana and California.

16:38 · Under this law, the Nazis sterilized approximately 400,000 children and adults— mostly Jewish people and other undesirables labeled “defective.”

16:50 · The eugenics programs that were implemented in various states in the U.S. were an inspiration for the eugenicists in fascist Germany.

16:59 · The eugenicists back in the United States were actually very proud of this fact.

How America Inspired Hitler’s Race Laws

17:03 · Clayton: Adolf Hitler said there’s one place in the world that’s got the right idea about restricting immigration and selective breeding, and it’s America.

17:12 · Earle: This actually starts with Leon Whitney, who is a muckety-muck in the American Eugenics Society.

17:17 · In 1934, one of Hitler’s staff members kind of wrote to him, requesting a copy of his book.

17:26 · His book’s called “The Case for Sterilization.”

17:29 · So he sends his book and then receives a letter from Adolf Hitler personally thanking him for the book.

17:39 · Hitler was very inspired by American eugenics.

17:43 · The zenith of that, as we saw quite devastatingly play out in Nazi Germany, was to exterminate people altogether, to just take away any possibility of them even having families.

17:56 · Earle: So after the end of the Second World War and their revelation more publicly of the atrocities of the Final Solution and the human experimentations that the Nazis were doing, the term eugenics got tied to Nazism.

18:13 · But eugenics didn’t end at the end of the Second World War.

18:18 · We stopped using the word.

18:20 · ♪ Dixon-Roman: Racial logics are threaded into the very fabric of the technology in the machine, possessing it.

The Turing Machine: When Machines Began to Mimic Us

18:30 · [eerie whistling, “Bella Ciao”] Narrator: Chapter 2— The Ghost in the Machine.

18:35 · Amira Moeding: In 1936, Turing comes up with the idea of the Turing machine.

18:39 · And that basically is this little thing that can do three operations.

18:44 · And he shows that within these three operations, you can basically calculate everything within the mathematical universe.

18:51 · It’s not a computer in our sense.

18:54 · It’s an abstract mathematical tool.

18:57 · As soon as it’s no longer an abstract mathematical tool, it becomes a military project.

19:02 · One of the things that happens in the Second World War, and you have a huge machine apparatus all of a sudden, and you have to have humans act within this technological environment.

19:17 · If you conceptualize a human as part of this big technological apparatus, you kind of start to conceptualize the person as part of the machine.

19:27 · Very often, you find the telling of the history of AI told like this: “So we have a computer. We want to make it intelligent.

19:34 · And what is intelligence?

19:35 · Well, if it can act intelligently in the world, then it must be intelligent.

19:39 · If this thing can calculate everything, there must be a way to remodel intelligence.”

19:45 · Narrator: After World War II, Oxford professor Gilbert Ryle began to contemplate intelligence.

19:52 · Michael Kremer: His father was the family physician of Karl Pearson, who more or less invented modern statistics in service of his eugenicist projects.

20:02 · But then his brother John Ryle was vice president of the Eugenics Society.

20:07 · You know, I don’t think Gilbert Ryle was a eugenicist, but he did think in terms of human capacities.

20:14 · He spends a lot of time on the question of intelligence.

20:17 · Ryle attacked what he called the “dogma of the ghost in the machine,” which he associated with the famous French philosopher René Descartes.

20:28 · Cartesian dualism is the idea that the mind and the body are separate things, and so when the body dies, the soul persists, that kind of thing.

20:36 · Gilbert Ryle’s concept of mind: he says consciousness is a product of the materiality of the body.

20:45 · Kremer: There is a position that he calls “intellectualism,” and that’s the position that what makes it intelligent behavior is that it’s guided by the thinking of thoughts, or, as he says, “the contemplation of rules.”

20:59 · Zehner: So these kinds of ideas that move from eugenics as a genetic grounding of white supremacy to theories of the mind or theories of behavior allow whiteness to have a new kind of white flight from the body, which really paints a picture of how AI is operationalized today.

21:18 · Kremer: Now, Alan Turing and Gilbert Ryle knew each other during the war.

21:22 · They were involved in something that one could say was trying to figure out the minds of these other people.

21:28 · 1950, Ryle accepted Turing’s paper “Computing Machinery and Intelligence” for publication.

21:34 · Turing wants to answer the question: Can machines think? [echoing] Elizabeth Sandifer: When we first developed computers in the 1930s and ’40s, suddenly we had this shockingly powerful tool.

Can Machines Really Think?

21:49 · We suddenly have these machines that can do all sorts of interesting stuff that they couldn’t do before.

21:55 · We already have a science fiction rhetoric of robots and things.

21:58 · We have the idea of artificial life.

22:00 · We’re decades past “Frankenstein.”

22:02 · So the obvious connection that any halfway decent nerd is going to make is, what if these things start thinking?

22:08 · And it’s a reasonable question to ask when Alan Turing is asking it in the 1940s.

22:13 · Birhane: You know, one of the biggest misconceptions is to portray AI in human terms, allocating, you know, consciousness and other humanlike characteristics to AI systems.

22:26 · It’s not that these AI systems have all these humanlike qualities.

22:31 · We have started to define and to view human cognition in machinic terms.

22:39 · So going all the way back to the 1940s, people were excited about thinking about how neurons work in our brain, building mathematical models of those.

22:47 · Archival film narrator: Dr. McCullough and his colleagues believe they are beginning to understand how the higher nervous system— a man’s brain— might work as a machine.

22:56 · If you know theology at all well, you’ll realize that the ideas in the mind of God are mathematics and logic.

23:07 · [eerie music plays] ♪ Felix is a device that shows how a machine can take over one of the human senses—vision.

23:26 · He is a machine that represents an advance in evolution.

23:31 · Oh, there’s Professor Wiener.

23:33 · Professor Wiener is an internationally famous mathematician.

23:36 · Zehner: This sort of late ’40s moment— Norbert Wiener is trying to do the cybernetics, breaking the distinction down between man, machine, and animal.

23:45 · Do we have machines that actually think?

23:48 · The word “think” is one of the words like “life” and so on, and “soul,” which are bad words.

23:55 · They mean just what we want them to mean…

23:57 · Zehner: And this moment is really interesting because it’s redefining the lines of what the human is but as something that can be bracketed off as an interior and only seen as sort of an output, so this kind of output function that can then be reduced to an equation or something that can be solved.

24:18 · We hope to possibly learn something about the general design principles of machines that learn.

24:23 · And if we’re lucky, maybe we’ll learn something about that most remarkable learning machine of them all— a human brain.

24:28 · Archival film narrator: The explosion of computer science and technology has both forced and enabled man to look, as never before, into the nature of his own being.

24:36 · The mysteries of the mind that have baffled philosophers for ages are slowly yielding to the onslaught of science.

24:43 · Moeding: I always found this a crazy leap to say, because there is something that can calculate everything, we must be able to remodel intelligence in this very abstract, logical form.

24:56 · But that is what AI is in the beginning.

24:59 · Imagine the postwar science world as, like, structured by these big interdisciplinary research laboratories that are mainly funded by a military budget.

25:09 · Shazeda Ahmed: AI is a term of art that was invented to raise philanthropic funding for research into what was called symbolic systems in the, like, mid-century kind of computer research world.

25:21 · What we call machine learning now, right, was really about pattern detection and kind of scaling of systems that do pattern detection.

25:29 · When Claude Shannon and I decided to collect a batch of studies, Shannon thought that artificial intelligence was too flashy a term.

25:37 · Ahmed: So it’s never a scientific term.

25:39 · He wants a big term that sounds flashy and it will bring in funding.

25:44 · Archival film narrator: To extend the power of the brain, we have created an incredibly swift machine that can do in a minute what would take a man a lifetime.

25:53 · Now using machines to study the brain will enable men to build better machines and perhaps to develop better brains.

26:01 · Saini: So these very big claims being made— there was a lot of hype that we’re replicating the brain, we’re mimicking the human brain, and it got repeated in newspapers and stuff.

26:11 · Arthur C. Clarke: All present computers are mechanical morons.

26:15 · Probably before the end of this century, we will be able to construct computers or artificial intelligences and which may in principle be more intelligent than we are.

26:26 · We may have a society in which robots will drift away from total metal toward the organic, and human beings will drift away from the total organic toward the metal and plastic, and that somewhere in the middle, they may eventually meet.

26:46 · Will we then have formed a kind of mixed culture, which perhaps might be higher or more efficient?

26:58 · Better?

26:59 · Narrator: These fantasies would shape the most powerful industry on Earth.

27:05 · [intriguing music] Chapter 3—Silicon Dreams.

Silicon Valley Births the Start-Up

27:11 · Silicon Valley has always liked to pretend that it doesn’t have a history.

27:15 · Part of that is, you know, it lets them have an excuse for when they repeat the mistakes of history.

27:21 · And part of it is that some of that history ain’t so savory.

27:25 · Becca Lewis: There’s a lot baked into the Silicon Valley mythology.

27:29 · At its core, it’s this idea that there is kind of a special genius class of men who are going to be able to lead us, as Americans or just humanity, into the future and into a better world.

27:46 · “We should be rewarding this special class of men “with all of the wealth that they generate, “all of the power that they want.

27:54 · “We should be recognizing them as geniuses, and we shouldn’t be questioning their decisions.”

28:00 · William Shockley is a big part of the origin story of Silicon Valley.

28:06 · Archival film narrator: Dr. William Shockley is one of three Americans sharing the physics award for research which produced the transistor.

28:12 · [applause] With the transistor, man has gone far toward matching some of the capacity of the human brain.

28:19 · Lewis: Shockley is known by many as the godfather or father of Silicon Valley.

28:25 · Archival film announcer: William Shockley, the inventor of the junction transistor.

28:29 · Archival narrator 2: Transistors will take their place in the complex calculating machines that have often been called electronic brains because they enable man to save days, months, even years in solving mathematical problems.

28:42 · Archival narrator 3: What’s inside the transistor?

28:43 · Dr. Shockley shows us using a huge scale model.

28:47 · Lewis: He launched his company, Shockley Semiconductor, in the Bay Area at a time when tech companies were still much more frequently built on the East Coast.

28:57 · He had grown up in Palo Alto…

29:00 · Shockley, archival: I arrived in Palo Alto when I was three years old, went to school here, including the Palo Alto Military Academy, which is still going.

29:07 · Lewis: …and decided to launch his company there.

29:09 · It became a really important company in the history of Silicon Valley.

29:13 · It was from that company that several other people went off and launched Fairchild Semiconductor, and that was where the microchip first got developed.

29:23 · Silicon Valley was really a microchip town, and from Fairchild Semiconductor, there were all sorts of startups that got spawned, often referred to as the Fair Children, including Intel.

29:35 · And so you have this whole lineage starting down from Shockley Semiconductor.

29:41 · Archival film narrator: Demand, growth, potential— familiar words to everyone in data processing.

29:47 · Each year more demand, more growth, more potential.

29:50 · Royal Institution host: At all periods of history, the human imagination has been captivated by the idea that the mysterious arts, whether of the sorcerer’s cell in earlier times or the scientist’s laboratory today, might be used for a process opposite where artificially giving birth.

30:06 · Lewis: So there is this fixation with women’s ability to give birth and kind of this quest for men to be able to capture that through the building of technology.

30:18 · Women give birth biologically, so men should be able to be the ones to give birth to new startups, new technologies, and really, this fixation on creating a patrilineal structure within Silicon Valley that doesn’t need women there.

30:35 · This is just a world of men, genius men, and the software world of the mind that they created.

30:43 · And that is inherent in the Silicon Valley mythology that we still see today.

30:48 · Our descendant will not be the child of the loin but the child of the brains, the thing we call the computer, which does not have to pass through the birth canal and does not grow by a tablespoonful of gray matter every hundred thousand years, which is the case in the rapid growth of our brain, but grows a factor of ten in power every seven years.

31:09 · The computer generation.

31:10 · There’s no question but that it will match us in narrow reasoning power by 1990 and go beyond us to become the great new intelligent race of the future.

31:21 · [echoing] race of the future, race of the future.

When Racism and Sexism Entered the Machine

31:24 · Jessie Daniels: We can’t really understand technology without understanding race and racism, and we can’t really understand race and racism without understanding technology.

31:34 · Jonathan Flowers: Almost every other piece of technology that we’ve developed tends to follow the lines or the historical and ideological conditions inherited by its developers.

31:45 · Saini: Not everybody who was a eugenicist or a race scientist before the war just, you know, shut up shop and just never looked at this again.

31:52 · There were some people who were still committed to this.

31:54 · This small cabal of people after the war, these race scientists after the war, their support was Wickliffe Draper, who was this very wealthy heir in the United States.

32:06 · The fund that he created was known as the Pioneer Fund.

32:09 · Wickliffe Draper’s intervention was influential in keeping race science alive.

32:14 · Lewis: William Shockley went back to Stanford University, where he had started out, and he became a professor there.

32:22 · And he became one of the most vocal and notorious scientific racists and eugenicists in the country.

32:30 · Shockley: One of the plans I talk about is a eugenics measure, the so-called voluntary sterilization bonus plan.

32:37 · And the way it goes is a bonus would be offered to everyone to be sterilized.

32:41 · Announcer: From New York, “Black Journal” investigates: Black or white superiority?

32:46 · Hello. Welcome to this edition of “Black Journal.”

32:49 · Now let’s find out what the controversy is about.

32:52 · My principal point is summed up in one word, which is the theme of my appearance on your program and my efforts, and the word is dysgenics.

33:01 · And dysgenics means, effectively, down breeding, retrogressive evolution.

33:06 · Lewis: Shockley really gave it this level of credibility because he was based at Stanford University, because he was the godfather of Silicon Valley.

33:15 · ♪ [button clicks] You’re on “Black Journal.” Go ahead, please.

33:19 · Caller: Yes, I was wondering if Dr. Shockley could explain the basic difference between the course he is taking in explaining white supremacy and the course that Hitler took in—during the Nazism reign.

33:32 · Brown: Thank you.

33:33 · Well, there are enormous differences.

33:35 · In fact, the lesson to be learned from Nazi history is frequently very misunderstood.

33:40 · It’s the First Amendment.

33:42 · It’s not that eugenics is intolerable.

33:45 · Eugenic programs are not inconceivable, they’re not inhumane.

33:50 · The long-range implications of what he is doing are no different than the propaganda campaign that Hitler and his Nazi unit carried on in Germany that ended up eliminating 6 million Jewish people.

34:04 · Lewis: William Shockley then ended up mentoring certain students at Stanford University, who went on to be big figures in Silicon Valley.

34:12 · This is the final touch on some of these large-scale objectives.

34:16 · I want to fit transistors into it somehow, make a computerized duplication of the human brain, and get higher achievement.

34:23 · But you can see the happiness meter is reading very high.

34:27 · So this might be a way of producing the most happiness for the most: ideal lives could be programmed by the computer and the overall effect would be “My,” those brains would say, “we lived a good life.”

34:37 · Narrator: Chapter 4—Optimism.

34:42 · Lewis: In the 1980s, you started to have people building the personal computer.

34:47 · Steve Jobs: By simply using the mouse, the user can move an arrow around on the screen and simply point to English words and point to pictures, so all through this very simple device.

34:56 · And so what we’ve done is eliminated a vast body of knowledge that one has to know in order to use this computer.

35:03 · Lewis: And by the ’90s, you started to have Silicon Valley building out these tools for the internet, this ability to connect the computers.

35:11 · Archival film narrator: “The Computer Chronicles,” the story of this continuing evolution.

35:16 · [TV theme music] John McCarthy has joined us.

35:20 · John is a professor of computer science at Stanford University.

35:23 · He invented the field of artificial intelligence.

35:26 · How smart can machines become?

35:28 · What are the limits of artificial intelligence?

35:31 · Well, I see no limit short of human intelligence.

35:36 · And then with faster machines, one could do the equivalent that a human could do in a short time.

35:45 · Moeding: The interesting thing about John McCarthy is he started out as an outright Marxist hoping for kind of the betterment of the world by technological tools.

35:54 · He kept the betterment of the world via technological tools part, but he turned, in his own words, “extreme right wing Republican.”

36:02 · He comes up with this term of “technological optimism.”

36:05 · Progress is just based on technology.

36:09 · The world is fundamentally structured by things that can be modeled mathematically.

36:14 · All physical systems— the Earth, humanity, space— is just a technology in itself and can be engineered.

36:22 · John McCarthy publishes on his website a little text called “Technology and the Position of Women.”

36:29 · And so we find this recurring theme in McCarthy’s thought when he says women are not as good at math as men are.

36:36 · And he pushes that kind of very masculinist culture at Stanford.

36:41 · He is concerned about too many women being admitted because they, in his view, have not quite the same ability in mathematics or too many people of color being admitted.

36:50 · There were enemies of progress— the climate movement and the civil rights movement and the emerging feminist and women’s movement.

36:58 · They all don’t see that in the end, technology will optimize everything.

37:04 · And so these movements have to be stopped.

37:08 · Benjamin Noys: And we’re kind of combining themes here with themes that we might associate with the ’60s and ’70s counterculture, of anxieties about control, with a more libertarian kind of notion of freedom from capitalist regulation, freedom from socialistic regulation.

37:22 · But the aim is unleash, and that becomes the kind of ideological fusion point for lots of these thinkers.

The Rise of the Tech Bro

37:32 · Douglas Rushkoff: The tech bro mindset is, you know, “Governments of the world, you know, beware.

37:37 · “We don’t need you. We’ve made our own place.

37:40 · “It’s, you know, it’s, you know, the internet, “and we don’t need your laws. We don’t need your damn rules.

37:45 · We’re going to go have fun, and [bleep] you.”

37:48 · What we didn’t realize at the time, if you get rid of government, you create free rein for business.

37:58 · OK, so what is the $64,000 question?

38:02 · Rushkoff: And business came on the net and just took it over like a fungal infection.

38:08 · Developers, developers, developers, developers.

38:11 · Developers. Developers.

38:13 · Developers. Developers. Developers. Developers.

38:15 · Developers. Developers. Developers.

38:18 · [audience clapping rhythmically] Yes!

38:22 · [applause continues] Lewis: And the ’90s was really the first time that you started to see this hero worship of entrepreneurs reach these—these huge heights.

38:33 · You had entrepreneurs building up these companies really quickly, getting funding for them, and then going public and making a fortune.

38:41 · What we call “Adventure capitalists.”

38:44 · Lewis: Kind of a pervasive worship of male power within Silicon Valley.

38:49 · Reporter: And how old were you when you started this company, or what became this company?

38:53 · 23. [laughs] Reporter: 27-year-old Elon Musk has his own computer command center, and his business is thriving.

39:03 · What do you see as the future of the internet?

39:08 · I think the internet is the— the superset of all media.

39:12 · It is the… [exhales] it is the be all and end all of media.

39:18 · It’s going to revolutionize all traditional media.

39:21 · Echoed low voice: Revolutionize.

39:24 · Lewis: In the 1990s, there were a few journalists who started to take note of this rise of what some people called techno libertarianism and what other people actually called techno fascism.

39:36 · So there were journalists like Paulina Borsook, who actually pointed out this pervasive worship of male power within Silicon Valley…

39:45 · The romance between libertarianism and high tech has existed for quite a while.

39:48 · Lewis: and how it was a little bit reminiscent of European fascism from the early 20th century.

39:54 · Borsook: And that is so much the mindset of this culture.

39:57 · If you don’t get with our program, then you’re going to be left behind, and there’s a deep contempt for kind of abiding by the rules of society that the rest of us poor plebs have to honor.

40:07 · The thing is high tech celebrates being this way, and exacerbates being this way.

40:12 · And it’s sort of being held up as the best we can do and how we all ought to be— these kind of bizarre values and religious beliefs, because that’s really what this is.

40:20 · Then people can identify it in their own lives and their own communities, when it comes up locally, and then sort of act appropriately.

40:27 · These days, the word community just means a bunch of suckers we can narrowcast our marketing messages to.

40:32 · Man: At some point in time, we’re going to have some type of realization set in that the internet stocks are tremendously overvalued.

40:39 · I’m sort of seeing a lot of people throw out their collective sanity, the level of hype.

40:43 · Alex Hanna: Tech hype is this particular type of hype that focuses on the innovations within technology.

40:49 · Lewis: Because there was such rapid growth in Silicon Valley, you had a bubble get created, way too much hype, and ultimately that all came crashing down in 2000, when the bubble burst.

41:01 · [bell clanging] Reporter: This closing bell might as well have been an alarm, so savage was the selling.

41:09 · The fragile technology stocks even harder hit.

41:11 · It’s described as nothing short of breathtaking, a points drop never before seen on the U.S. markets.

41:18 · Lewis: And so the early 2000s were kind of this period of retreat and regrouping for Silicon Valley.

41:24 · But it was in the early 2000s that you started to have the rise of web 2.0, as people called it, and the social web.

41:34 · And in many ways, it had reinvented itself.

41:37 · It had started to speak of democratization.

41:40 · It had started to speak of the ability for people to communicate with each other and the power of that.

41:46 · Bill Gates: You could say that we’re back to a little bit of hype in some of these valuations but nothing like 1999.

41:54 · We won’t see that again in our lifetime.

41:56 · Man: The partnership today is oriented around democratizing, unleashing the web, unleashing the data.

42:03 · [cheering and applause] Echoed low voice: The data.

42:06 · Lewis: It was in 2004 that Mark Zuckerberg founded Facebook.

42:10 · Mark Zuckerberg: So you can run ads or you can do transactions.

42:14 · And we encourage both.

42:15 · Echoed low voice: The data.

42:17 · Host: Each year we pick the coolest young entrepreneurs and feature them in our 30 under 30 list.

42:22 · Meet Sam Altman, founder of Loopt.

42:25 · He managed to turn the question Where are you?

42:27 · into a million-dollar idea.

42:29 · Sam Altman: Loopt is about connecting with people on the go, which is, after all, the main reason you have a phone.

42:34 · We show you where people are, what they’re doing, and what cool places are around you.

42:38 · The orange pin up there is where I am right now, and the blue pins represent my friends.

42:43 · We make serendipity happen.

42:45 · Host: I’m here with Sam Altman, the CEO and founder of Loopt.

42:47 · There are two kind of things for us.

42:50 · What we really want to do is connect users to the world around them.

42:52 · And we’ve been really happy to see the growth in terms of the data we’re been able to pull in.

42:56 · You know, get some of that data.

42:58 · Echoed low voice: That data.

42:59 · Lewis: Many of the original assumptions and values that were there in the ’90s about entrepreneurship and the ability of, you know, young men to— to build immense amounts of power and wealth, none of that was questioned, and that came back with a vengeance.

Sam Altman, Elon Musk, and the AI Power Struggle

43:18 · Narrator: Chapter 5— Building God.

43:22 · Altman: You know, I think AI will probably lead to the end of the world.

43:25 · But in the meantime, there will be great companies created with serious machine learning.

43:30 · Actually just agreed to fund a company that is— not even really a company, sort of a semi-company, semi-nonprofit— doing AI safety research.

43:39 · Echoed low voice: Safety research.

43:41 · Altman: Today we have Elon Musk.

43:42 · Elon, thank you for joining us.

43:44 · Thanks for having me.

43:45 · So we want to spend the time today talking about your view of the future and what people should work on.

43:50 · AI is probably the single biggest item in the near-term that’s likely to affect humanity, because it is something that could go— could go wrong, as we’ve talked about many times.

44:02 · And so we really need to make sure it goes right.

44:05 · And that’s, you know, the reason that, obviously, you, me and the rest of the team, you know, created OpenAI.

44:11 · Woman: Will AI exterminate us?

44:14 · It’s good that we’re working together. Thank you.

44:16 · Broadcaster: Sunak fears that artificial intelligence could be more lethal than Hitler.

44:21 · Ahmed: There are a couple of factors that came together to create the field of AI safety.

Effective Altruism: The Philosophy Behind the AI Elite

44:27 · I would start it with effective altruism.

44:29 · There’s a lot of funding from effective altruism that has gone towards AI safety as a field.

44:34 · Noys: Effective altruism is a philosophy of the present moment where people were trying to define what’s the best way to be altruistic, to spend your money to help people.

44:45 · You know, the classic 19th-century robber barons, the people who made vast fortunes off of new technologies— in that case, often rail— spent their money building libraries, building public resources, which you can go and visit today.

44:58 · So what do the tech billionaires of today spend their money on to help people?

45:04 · Ahmed: A lot of high net worth individuals who come from the tech fields have a lot of money to give to this field.

45:10 · One high net worth individual funding this space was Sam Bankman-Fried.

45:14 · That creates a base where you can form nonprofits, research centers, think tanks that are focused on these issues.

45:22 · For the longest time, you can be drawn into these communities and think these are just people who want to improve themselves and want to improve the world, but all of the little subfields around that, things like progress studies, are still rooted in race science.

45:37 · And so it will always go back to race science.

45:40 · So there were a few graduate students in philosophy at the— at Oxford and Cambridge.

45:46 · And these were people who were trying to figure out how they could apply utilitarian philosophy to the real world.

45:51 · How would we enable the greatest number of human beings as possible to live in the future and also to flourish or thrive?

45:58 · Noys: Sounds good.

46:00 · Unfortunately, it’s kind of, uh, declined into a thinking around AI and also panic about AI.

46:06 · So effective altruism becomes, “Oh, AI is going to take over.

46:10 · “AI is going to achieve consciousness.

46:13 · “So we better support kind of AI.

46:15 · The best way we can make a future is to support AI.”

Nick Bostrom: Superintelligence and Eugenics

46:18 · This is largely something that came to their attention through a thought experiment in another philosopher’s book.

46:24 · So Nick Bostrom wrote the book “Superintelligence.”

46:27 · Nick Bostrom: That tries to bring careful thinking to bear on the really big-picture questions.

46:32 · Are there threats to the very survival of the intelligent species?

46:36 · Are there ways in which future technologies could change the basic parameters of the human condition in some way?

46:42 · Ahmed: And it’s a book where he is kind of doing thought experiments around how could we attain, “superintelligence” or intelligence that exceeds that of human beings either organically through selecting for particular embryos that have the traits that he believes would lead to superintelligence.

47:00 · Bostrom: We have sort of new waves of genetic enhancement coming online every few years or every 5 or 10 years.

47:06 · So maybe parents would have to select which new person to bring into existence.

47:10 · Ahmed: This is straight up eugenics, right?

47:12 · There’s no other way to define that act.

47:15 · And you can go to that part of the book, and he kind of does this experiment.

47:19 · The longer you spend sitting, reading work from these people or listening to them speak, the more it becomes apparent that humanity does not mean every single human being, right?

47:28 · It means a certain class of people and elites that they see themselves reflected in.

47:34 · Torres: In 2023, Nick Bostrom published an apology for an email that he had sent in the 1990s to a listserv with hundreds— it might have been thousands—of people.

47:45 · But the listserv consisted mainly of eugenicists, so I think a lot of people weren’t that shocked by his claim.

47:52 · But in his apology, he refused to walk back his claims that certain racial groups might be more intelligent than other groups.

48:01 · And all he did was the bare minimum of apologizing for actually writing out the N-word.

48:07 · David Gerard: He apologized for using the N-word and said he should have phrased it differently, but he still believes it.

48:13 · He thought—he thought that was an excuse.

48:16 · [indistinct chatter] Ahmed: It doesn’t actually take that long to look within this field and see how things that on their surface are about progress and improving the quality of our outcomes in life, they’re very quickly tethered back to something that is eugenicist.

48:34 · Torres: So Bostrom has written a lot about superintelligence and outlined the potential dangers of building a superintelligent machine that is not sufficiently aligned with our values.

48:46 · So this is where he goes on his thought experiment of what would happen if we ended up with artificial intelligence that was “smarter than human beings.”

48:55 · Torres: This is a book that was massively influential in Silicon Valley.

48:59 · It inspired people like Sam Altman, and it was promoted by individuals like Elon Musk.

49:04 · Broadcaster: And a warning from Tesla Motors CEO Elon Musk.

49:06 · It has nothing to do with cars.

49:08 · Instead, Musk warns about artificial intelligence, which he has called “more dangerous than nuclear weapons.”

49:16 · Musk spoke at a symposium at MIT.

49:19 · I mean, with artificial intelligence, we are summoning the demon.

49:23 · Those ideas were mostly laughed out of academic computer science, right?

49:26 · There were people who were saying, “Once you understand how these systems work, of course you don’t believe, that what they’re doing is superintelligence.

49:32 · They require a lot of intervention from human beings.”

49:36 · But also a lot of high net worth individuals who come from the tech field have a lot of money to give to this field.

49:43 · That creates a base where you can form nonprofits, research centers, think tanks that are focused on these issues.

49:49 · So there are think tanks that are specifically working on existential risk or on AI safety that then fund this research.

49:55 · They fund, um, computes so that people can run models and do the kinds of testing that they think will lead to preventing the worst outcomes of AGI.

50:04 · Those are some of the framings in which people are then applying for funding.

50:09 · Host: Is some form of superintelligence possible?

50:12 · Would you actually like it to happen at some point?

50:15 · “Yes,” “no,” or “It’s complicated”?

50:18 · Complicated, leaning towards yes.

50:19 · It’s complicated.

50:20 · Yes.

50:20 · Yes.

50:23 · Really complicated.

50:24 · Yes.

50:26 · It’s complicated.

50:28 · Very complicated.

50:30 · Well, heck, I don’t know.

50:32 · [laughter] It depends on which kind.

The Cult of AGI

50:36 · Ahmed: There are people who will refer to this as a cult, but it’s also completely out in the open.

50:41 · Adam Becker: So the question of whether or not this is a cult is not just is this, you know, online community a cult?

50:49 · It’s not just, is this, uh, philosophical movement that’s headquartered in Oxford and has branches in basically every major university in the English-speaking world and beyond a cult?

51:01 · The question is, is this movement that is influential in the largest AI companies and the entire tech industry a cult?

51:14 · And the answer is, kind of.

51:17 · It is more like a cult than we would like something like that to be.

51:22 · Ahmed: OpenAI was founded by a number of people who come from effective altruism and were thinking about AI from this perspective, and did want to build AGI.

51:33 · Paris Marx: If you listen to people like Sam Altman, basically what they want to do is to try to build the AGI, the AI that reaches the level of human capabilities that sometimes they position as being a real threat and a real scary thing, but at other times, they position as being a complete necessity that we need to do no matter what.

51:52 · The acronym AGI stands for Artificial General Intelligence, and it’s basically a kind of hype inflation.

51:58 · So when artificial intelligence got over-applied to too many things, and people still wanted to be selling this idea of an autonomous thinking machine, they had to come up with a new name for what comes next.

52:08 · And, in fact, there’s two new names. There’s AGI and ASI.

52:11 · So artificial general intelligence is supposed to be something that is, it’s very ill-defined, but it effectively—equivalent to what a person can do, and artificial superintelligence is something that is better than that.

52:24 · Specifically, one of the goals at DeepMind was to find a pathway to AGI?

52:28 · Absolutely.

52:29 · On our first business plan, in 2010, it had one sentence on the front cover, and it said, “Build the world’s first artificial general intelligence.”

52:38 · We said from the very beginning we were going to go after AGI at a time when in the field, you weren’t allowed to say that because that just seemed impossibly crazy.

52:49 · McQuillan: And that’s the thing these companies were founded to bring about.

52:52 · OpenAI, DeepMind, all the leading AI companies, actually derive their authority from the idea that they’re not just about AI, whatever that actually is, but about bringing about AGI and that they’re on their way to AGI and that AGI is actually quite close.

53:07 · Every research house right now is working toward building AI that mirrors human intelligence, human-level intelligence— they call it AGI.

53:15 · Where are we right now in the progression, and how long is it going to take to get there?

53:20 · This is what’s on everyone’s lips right now.

53:22 · And the debate is, is how close are we to AGI?

53:24 · What’s the correct definition of AGI?

53:27 · Interviewer: So you’re credited by many as coining the term “Artificial General Intelligence,” “AGI.”

53:32 · Tell us about 2001, how that happened.

53:35 · How did you define AGI back—back then?

53:38 · Unfortunately for them, or rather unfortunately for us, the “G” in “AGI,” which is “general intelligence,” is— is really just traceable to this thing called the “G” factor.

53:47 · Torres: Shane Legg has cataloged various definitions of what intelligence is.

53:53 · He has cited Linda Gottfredson, who kept race science alive and was funded by the Pioneer Fund.

54:01 · She offered an explicitly racist notion of IQ.

54:05 · I mean, anytime you’re talking about IQ, you’re talking about the “G” factor— general intelligence.

54:09 · She argued that certain racial groups have a lower IQ and, hence, are less intelligent than other groups.

54:15 · Saini: People in the tech world are borrowing the language of intelligence research, which is essentially an offshoot of eugenics.

54:22 · These technologies, they’re using that phrase, “artificial intelligence,” and they’re confusing that with work that is done into human intelligence.

54:32 · This is the history of— this is, like, the long history of humanity.

54:36 · Um…it does feel a little different this time, like a crazy high IQ tool.

54:42 · Emily M. Bender: Anytime someone is comparing their computer system to what people can do, they are presupposing this ranking and saying, “Not only can you rank people in this way, “but you can also put machines into the ranking alongside the people.”

54:54 · It’s incredibly dehumanizing and incredibly problematic.

54:58 · This is like a not scientifically accurate, this is just sort of a vibe or a spiritual answer.

55:04 · But every year, we move one standard deviation of IQ.

55:09 · Also, every year, the cost of last year’s intelligence falls by about a factor of ten.

55:14 · Saini: It was that fundamental original sin of using the word intelligence in the first place with reference to machines, that it’s become so naturalized within the tech community now that the transformation is complete.

55:27 · Attendee: The term AGI is thrown around a lot.

55:29 · How would you define AGI?

55:31 · AGI is basically the equivalent of a median human.

55:35 · McQuillan: Artificial general intelligence is coming.

55:39 · Legg: What we’re talking about is an incredibly profound transition.

55:43 · It’s like the arrival of human intelligence in the world.

55:48 · Torres: They claim that AGI is going to be the most important technology that we ever invent, because it might trigger the singularity, an intelligence explosion that just radically transforms the world in which we live, enables us to upload our minds to computers, colonize space, and so on.

56:02 · Singularity is a lot of what’s sort of fueling these fantasies and fears.

The Singularity: The Race for Superintelligence

56:10 · This is an idea that’s been promoted, most notably by Ray Kurzweil in books like “The Singularity Is Near,” which came out in 2005, and the sequel to that book, which came out in 2024, “The Singularity is Nearer.”

56:23 · Legg: I read the book by Ray Kurzweil, actually.

56:25 · I concluded that he was fundamentally right that computation was likely to grow exponentially.

56:33 · Ray Kurzweil: Ultimately, we’re going to recreate the full powers of human intelligence in a machine.

56:37 · By the time we get to the 2040s—say, 2045— we’ll be able to multiply human intelligence a billionfold.

56:44 · That will be a profound change that’s singular in nature, so we use this term.

56:51 · Becker: AI is almost always at the heart of these ideas about the singularity, the idea that we will have, you know, better and smarter AIs that get smarter and smarter and smarter until we get one that’s as smart as a human, and then that one will rapidly improve its own intelligence in a self-reinforcing cycle until it ascends to, you know, massive superintelligence, outsmarting the entirety of humanity as a whole and ascending to AI godhood.

57:24 · When we actually reach AGI, there’ll be lots of controversy.

57:28 · By the time the controversy settles down, we’ll realize that it’s been around for a few years.

57:35 · Broadcaster: This new concept that Sam Altman has come up with— he coined the “gentle singularity,” the singularity which is sort of the fast take-off of intelligence, an exponential increase of intelligence of AI.

57:50 · Sam is now making the case that it looks like we’re already in the singularity.

57:55 · Becker: It is taken as gospel, spoken or unspoken, by a surprisingly large number of people in the tech industry, given that it is hot nonsense.

58:09 · Broadcaster: Yeah, the event horizon.

58:10 · Broadcaster 2: The—event horizon.

58:11 · Feels pretty good, yeah, and that it is increasing exponentially now.

58:16 · And there are parts of these LLM AI tools that are smarter than humans.

58:22 · Senator Fetterman: Mr. Altman, you’re really one of the people that are moving AI.

58:26 · And now I get to ask you— I mean, like, literally the expert.

58:29 · You know, some people are worried about AI or whatever, and I’m like, you know, “What about the singularity?”

58:35 · If you would address that, please.

58:37 · You know, as these tools start helping us to create next and future iterations, some people call that singularity.

58:43 · Some people call that the take-off.

58:45 · Whatever it is, it feels like a sort of new era of human history.

58:49 · And I think it’s tremendously exciting that we get to live through that.

58:51 · I predicted a 50% chance of AGI by 2028.

58:56 · I still—I still believe that today.

58:58 · ♪ Currently, we are living in peak AI hype.

Peak ”AI” Hype: When Will the Bubble Burst?

59:04 · Broadcaster: If you extrapolate the curves that we’ve had so far, right, it does make you think that we’ll get there by 2026 or 2027.

59:13 · For the past few years, I keep thinking, “This is peak hype,” and it just keeps getting better—or, rather, worse.

59:19 · Broadcaster: Let’s talk about the broader AI race.

59:21 · Who do you think wins?

59:22 · Torres: So this race right now involves a bunch of companies like DeepMind, founded in 2010; OpenAI, which was founded five years later, in 2015; Anthropic, which emerged directly out of OpenAI, was founded in 2021; as well as XAI, which Elon Musk started in 2023.

59:41 · And much more recently, of course, Meta has joined the race.

59:45 · Zuckerberg: We have a whole lot of new AI experiences.

59:49 · Gerard: We noticed the AI bubble was identical to the crypto bubble, as in not just similar guys saying similar phrases and similar excuses but literally a lot of the same guys, like Marc Andreesen.

1:00:06 · Well, you are sitting in the middle of Silicon Valley, and you kind of invented the internet, so how will the AI race pan out here?

1:00:13 · Yeah. So the—the theory and the hope that— certainly that we’re betting— that we’re betting against and investing hard against is that—is that AI and specifically these new breakthroughs around AI, like generative AI, represent a new platform.

1:00:26 · And every time there’s a platform shift, there’s an opportunity to reinvent the industry and reinvent basically the entire ecosystem and all the different ways that people use technology and create an entirely new generation of companies.

1:00:36 · Gerard: What this is, is there’s too much venture capital, there’s too much money flying around desperate for a home.

1:00:44 · Because we don’t tax these people until the pips rattle, they have too much money and they use it to cause damage.

1:00:50 · And these guys are literally only interested in lottery-level returns.

1:00:55 · They desperately want returns without actually funding an economy that’s healthy.

1:01:00 · This means there’s no sane things to invest in that give returns, so they have to invest in insane things.

1:01:06 · They look for these industries that can bubble.

1:01:09 · They aren’t interested in anything normal.

1:01:11 · They want bubbles.

1:01:12 · They want irrationality. They want exuberance.

1:01:14 · They want naive suckers piling their retail dollars in so they can skin them.

1:01:19 · Now, they were casting about for a bubble after Web3 fell flat and the metaverse never took off.

1:01:26 · They had seen Sam Altman pushing forward GPT-3.

1:01:30 · Broadcaster: OpenAI CEO Sam Altman.

1:01:32 · There will be some change required to the social contract given how powerful we expect this technology to be.

1:01:40 · Bender: The AI hype is there to bring in investment based on a fantasy.

1:01:46 · Gerard: OpenAI, their pitch is they can spend money faster than anyone because they have to spend money faster than anyone if they’re going to successfully build God.

1:01:55 · Altman: We’ve no idea how we may one day generate revenue.

1:01:57 · Um, we have made a soft promise to investors that once we’ve built this sort of generally intelligent system, basically, we will ask it to figure out a way to generate an investment return for you.

1:02:09 · Gerard: This is his entire pitch.

1:02:13 · This is why everyone competing with him thinks, “Well, we have to spend money.”

1:02:16 · The amount we’re willing to spend has gone up, in fact, has gone up fast, something like 10x a year.

1:02:23 · Gerard: Just set money on fire and pump out carbon dioxide as fast as we possibly can.

1:02:29 · Otherwise, “Then we can build God, too.”

1:02:31 · Spend more money to produce smarter models.

1:02:33 · Trying to build God in the most embarrassing way possible.

1:02:36 · Broadcaster: OpenAI CEO Sam Altman reportedly looking to raise an eye-popping $5 trillion to $7 trillion.

1:02:41 · Reporter: It is 7 million million.

1:02:43 · And here it is written out, 12 zeros, if you are counting.

1:02:48 · The most interesting is perhaps the why.

1:02:51 · OpenAI and Altman, they are on a quest to develop AGI, or artificial general intelligence.

1:02:56 · This is like the moonshot of all moonshots.

1:02:58 · Altman: Hard to say where all this can go without sounding like a crazy person.

The Hidden Cost of Outsourcing ”AI”

1:03:02 · Oprah Winfrey: I actually saw a headline that said you were the most powerful man on the planet, and I’m wondering how that sits with you.

1:03:13 · I— [eerie whistling, “Bella Ciao”] [sighs] It’s definitely strange to hear you say that.

1:03:20 · Bender: It is very hard to be the one pointing out that the emperor, in fact, has no clothes.

1:03:25 · Narrator: Chapter 6— The Emperor, in fact, has no Clothes.

1:03:30 · McQuillan: Rather than getting caught up in these questions of what is intelligence and are computers like our minds, and so on and so forth, is just to look at what these systems actually do when you put them in the world.

1:03:39 · Birhane: The entire ecosystem, the entire AI pipeline, it’s people, actually, through and through, so people whose data is constantly harvested.

1:03:50 · Companies like OpenAI and, in general, big tech companies, are completely predatory when it comes to data practices.

1:03:57 · OpenAI has outsourced a lot of the development of ChatGPT for example, to Kenyan workers.

1:04:04 · We started with the Data Workers’ Inquiry one year ago, in which we try to flip the script and invite data workers who actually do the research to be the experts and to tell us how things are.

1:04:16 · Richard Mathenge: Let me paint the picture to be very clear and precise.

1:04:20 · I am predominantly a Nairobian all my life.

1:04:24 · Where I live, individuals are very desperate.

1:04:28 · And so they will do anything for money.

1:04:30 · They will go an extra mile.

1:04:32 · There is employment challenge in the entire Africa.

1:04:36 · Nairobi, we do call it “Silicon Savannah” because we have a high population of young people in the process of looking for money— they look at themselves doing some of the tech work.

1:04:47 · And one of the online jobs that they get themselves into is training AI models.

1:04:53 · With OpenAI, when they were training ChatGPT, I’m one of the people who participated in training their data set.

1:05:01 · We were being paid less than a dollar per hour.

1:05:05 · I applied online.

1:05:07 · Samasource, it’s a company that’s based in San Francisco in the U.S.

1:05:13 · The narrative that Sama was selling, that they are bringing work to Africa, that Africans are very poor and they want to pull them from poverty.

1:05:22 · And they will do that by giving them simple tasks to complete.

1:05:26 · Emcee: Our space is called Sama App.

1:05:29 · You can see.

1:05:31 · So this is the tasks.

1:05:34 · Krystal Kauffman: Sama— S-a-m-a…

1:05:36 · will actually go into particular slums in Nairobi, and they will recruit people and say, “We have this great job.

1:05:46 · You come and you work online and you help clean up the internet.”

1:05:50 · Milagros Miceli: All this as if they were doing these people a favor because they let them work in this wonderland that is the AI industry.

1:05:59 · Mathenge: For Sama to be in good books with the government, they have to create “employment.”

1:06:06 · “We have employed this inventory from this slum.

1:06:10 · In return, please protect us.

1:06:12 · Actually, the funny thing, when you are applying to doing Samasource, they have a drop-down for those targeted areas.

1:06:21 · So in case you’re not from the slums, you will not be picked to work at Samasource.

1:06:25 · That is the main thing for you to be considered.

1:06:27 · Mathenge: When we started off training, the content that we were subjected to was not as serious as the content that we encountered during the real work.

1:06:40 · And this was deliberate, I believe, so that you will not quit.

1:06:44 · Now, with our Kenyan culture, you can’t just explain to someone that we are working with sexual content.

1:06:50 · They might think that you are doing something illegal.

1:06:54 · Mophat Okinyi: We raised these concerns to the management and told them that it’s like what we are reading is very graphic and it’s staying with us, it’s walking with us, it’s moving with us, and we need help.

1:07:06 · The feedback we got was that There is no time for counseling because the targets you are given were very high, and we had to meet those targets before the end of the day.

1:07:18 · Was after working on this, my behavior started changing— screaming at night, waking up, not sleeping.

1:07:25 · First of all, it comes with an instance of paranoia, the haunting shadows where you project it to those closest to you.

1:07:33 · At night, you cannot sleep, so some kind of insomnia.

1:07:36 · You try to get sleep, but— what you read still keeps on lingering in your mind.

1:07:42 · Yeah, so, at the end of the day, what it has done to you is much more than what you expected.

1:07:49 · It tears the veil of what makes you to be human.

1:07:54 · [eerie music] Miceli: These Big Tech companies are making trillions on the labor of these people.

How AI Exploits the Vulnerable

1:08:04 · They prey on specifically vulnerable populations, and this is a pattern I’ve seen repeated in Buenos Aires, in Argentina, in India—those programs that target single mothers or people from a lower caste.

1:08:18 · I’ve seen companies go into the refugee camps, and they’re handing in flyers.

1:08:23 · And this is on purpose. This is by design.

1:08:27 · It’s very much intentional.

1:08:28 · Mathenge: The reason why they target this country is because of the language issue.

1:08:34 · You cannot take content to be moderated in English— say, a place like Morocco or Tunisia— because these are Arab-speaking nations.

1:08:45 · We speak a variety of languages, but English naturally comes because of our colonial master.

1:08:53 · We were colonized by the British.

1:08:57 · And so it is the language, obviously, that we’ve received.

1:09:03 · Okinyi: Big tech like OpenAI, they just take advantage of gaps in law.

1:09:08 · They build their technology out of exploitation and data theft.

1:09:12 · That’s what I call digital colonialism.

1:09:15 · [gunshot] During colonial era, the slave masters came and gave gifts to Africans and promised them that “If you work with us “and if you give us slaves, “we are going to give you firearms.

1:09:27 · “And with these firearms you are able to expand your boundaries and be more superior.”

1:09:33 · When African chiefs— people who are trusted by their community members to lead them and to protect them— sold them out as slaves, that was a very big betrayal.

1:09:45 · Mathenge: In as far as accountability mechanisms by the government, unfortunately, there is zero.

1:09:53 · The government of today is actually in bed with these organizations.

1:09:58 · Okinyi: Currently, locally, like, in Nairobi, things are not any better.

1:10:02 · And they have managed to change the laws now to protect the big tech.

1:10:06 · We are now not able to prosecute them.

1:10:09 · It is making workers to lose hope completely.

1:10:13 · Man: And I’m talking about Samasource because those people there were taken to court, and they had real trouble.

1:10:21 · Now I can report to you that we have changed the law, so nobody will take you to court again on any matter.

1:10:28 · We will now have the opportunity to encourage more companies.

1:10:33 · Matt Mahmoudi: Those data annotation forms of exploitation, without the data that feeds into the training data, there is no AI.

Data Centers: Digital Colonialism and Exploitation

1:10:41 · And it tells us that there is a crisis here because there is this attempt to really hide that exploitation, completely obfuscated and not apparent to the end-user.

1:10:49 · AI is just a repackaging of our data to produce some pretense that, you know, this thing can do things magically— that this is just automation, that this is just happening because a large language model is large, because OpenAI are geniuses and Sam Altman, in particular, is a wunderkind.

1:11:07 · And that’s not what it is, right?

1:11:09 · It’s just the ability to obfuscate new colonial logics mapped onto similar patterns of colonialism and imperial extraction of the past.

1:11:19 · I believe deeply in building personal superintelligence for everyone.

1:11:23 · And at Meta, we have the resources to build massive infrastructure required.

1:11:27 · Echoed low voice: The massive infrastructure required.

1:11:30 · Chapter 7— The Massive Infrastructure.

1:11:33 · Dixon-Roman: The trick here is, “We’re giving you something for free.”

1:11:37 · I might characterize it as one of the tricks of techno capitalism.

1:11:42 · McKenzie Wark: So, I’ve been arguing for, like, 25 years now, like, what if this isn’t even capitalism anymore, it’s something worse?

1:11:50 · So the new layer on it, it’s very much about, can you control the whole value chain by controlling access to information?

1:11:58 · The so-called “tech sector” is now like a massive infrastructure.

1:12:03 · Like, they don’t just run on pure information.

1:12:06 · Like, all of that requires vast amounts of processing power, huge facilities.

1:12:13 · Chase Lochmiller: We’re calling them AI factories— large-scale data centers with chips that can manufacture intelligence.

1:12:21 · These data centers are no longer a bunch of individual computers.

1:12:26 · You really should be thinking about, the data center is the computer.

1:12:29 · So what we’ve seen over the past number of years is this massive expansion of hyperscale data centers, these massive centralized facilities which have massive energy and water demands not just to power them but to cool them.

1:12:42 · Ana Valdivia: We know that training Llama 3— that was Meta LLM— used 22 millions of liters in 97 days.

1:12:51 · This is the same amount of water that someone in London might use in more than 400 years.

1:12:57 · And it’s interesting to think about the water consumption of the AI supply chain.

1:13:02 · Man: These pipes are absolutely huge, and I can certainly feel the water flowing through here right now.

1:13:13 · Okinyi: In Kenya, now they are building very big and huge data centers.

1:13:17 · [device beeping] Man: This data center will be the largest in East Africa.

1:13:25 · [music] Okinyi: The funny thing is that data centers just use fresh water.

1:13:34 · In Nairobi, we struggle to get fresh drinking water.

1:13:38 · The taps of water in Nairobi are salty water.

1:13:42 · To get the fresh water, you have to buy water fresh and then put it into your house.

1:13:47 · If this data center is going to be built to use the same fresh water that we are struggling to get, then it’s going to be crazy for us.

1:13:56 · Interviewer: If you had a thousand times more compute, what would you do with it?

1:14:01 · I mean, I guess the super meta answer— I would ask it to work super hard on AI research, figure out how to build, like, much better models, and then ask that much better model what we should do with all that compute.

1:14:13 · Alix Dunn: They sniffed the vapors of inevitability and then started building with that in mind.

1:14:18 · They are going to be in charge of the digitization of our entire world.

1:14:22 · Mel Hogan: So we’re building this kind of data center industrial complex so that we’re then locked into these datafied worlds.

1:14:28 · It shows that material investments really shape political futures.

1:14:33 · Broadcaster: In Memphis, Elon Musk is making a play to control the future of artificial intelligence.

1:14:38 · His company xAI says it has built the biggest supercomputer in the world.

1:14:43 · Tiera Tanksley: Where Elon Musk’s data center is, surrounding communities are predominantly Black.

1:14:48 · [Residents speaking at once] Broadcaster: A recent health department hearing turned into a shouting match.

1:14:55 · Man: We’ve shown up here today because we’re tired.

1:14:59 · And we have an expectation of the people that we elect and put into place.

1:15:07 · We expect them to do what is in our best interest.

1:15:11 · [Cheering] Woman: Guess what. They’re sitting right there in the front not saying a [bleep] damn thing.

1:15:17 · I’m going to invite a representative up from the applicant, xAI— Mr. Brent Mayo.

1:15:23 · [crowd booing] Hi there.

1:15:29 · [booing continues] xAI is committed to meeting the highest standard of emissions.

1:15:35 · Broadcaster: An executive from xAI ducking out a side door.

1:15:39 · [indistinct shouting] Dunn: You can only build so many data centers in one place.

1:15:44 · So let’s say a town says, “I don’t like data centers.

1:15:47 · I don’t want any more of them.”

1:15:48 · “That’s fine.

1:15:50 · “We’ll just go to another town that doesn’t yet know about all of these implications.”

1:15:53 · And that is happening at scale around the world right now.

1:15:56 · [sirens in distance] ♪ Trump: Well, thank you very much.

The White House Bets Big on ”AI”

1:16:03 · And it’s an honor to be here today.

1:16:06 · We have—uh, first full day as president, we’re back.

1:16:12 · Marx: Early in Trump’s term, you saw Sam Altman alongside Oracle CEO Larry Ellison and SoftBank CEO Masayoshi Son next to Donald Trump, you know, in the White House, saying that they were planning to invest $500 billion in this major data center project that they envision being powered by nuclear energy.

1:16:32 · Altman: I think this will be the most important project of this era for AGI to get built here.

1:16:37 · We wouldn’t be able to do this without you, Mr. President, and I’m thrilled that we get to.

1:16:41 · Broadcaster: On data centers, will you rescind President Biden’s executive order that opens up federal lands for data centers?

1:16:47 · Trump: That sounds to me like it’s something that I would like.

1:16:50 · I’d like to see federal lands opened up for data centers.

1:16:52 · I think they’re going to be very important.

1:16:53 · Broadcaster: How’s it been working with President Trump?

1:16:57 · I—He loves infrastructure.

1:16:59 · I would like for many reasons— I would like AGI to be trained in the U.S., and the tech and infrastructure for this are inseparable.

1:17:08 · It shows the ambition that these people have in order to build out these massive infrastructures that are kind of existing at a scale that we haven’t seen before when it comes to computational infrastructure.

1:17:20 · This is something given to me by Mark Zuckerberg.

1:17:24 · And you’ll see, this is AI now.

1:17:26 · Echoed low voice: AI now.

1:17:27 · Trump: But look at that.

1:17:29 · That’s the size of Manhattan.

1:17:30 · Echoed low voice: Manhattan.

1:17:32 · That’s Meta, Facebook as people understand it to be.

1:17:36 · These are big things, and they’re going up.

1:17:39 · A lot of them are going up now.

1:17:41 · I don’t know that big, actually.

1:17:42 · Mark is building four of them.

1:17:45 · Lewis: Time and time again, we see in Silicon Valley this crop of elite, incredibly powerful, incredibly rich leaders.

1:17:54 · The mythology of Silicon Valley is implicitly built around the ideal of genius visionaries leading us into the future.

1:18:05 · President Trump: They’re leading a revolution in business and in genius and in every other word I think you can imagine.

1:18:11 · There’s never been anything like it.

1:18:14 · The most brilliant people are gathered around this table.

1:18:17 · This is definitely a high IQ group.

1:18:20 · You know, all of the companies here are building just—making huge investments in the country in order to build out data centers and infrastructure to power the next wave of innovation.

1:18:31 · These monstrous, huge, beautiful places, they’re palaces of genius.

1:18:37 · Echoed low voice: Palaces of genius.

1:18:39 · McQuillan: Fascistic ways of ordering and acting are introduced through technological infrastructures, you know, as much as they are through representative ones.

Fascism and AI: “We Don’t Want Woke ’AI’”

1:18:48 · Narrator: Chapter 8— Slopaganda.

1:18:52 · Dixon-Roman: We’re seeing the shaping and using of AI for techno-fascist interest.

1:18:58 · Trump: OpenAI, Google, Meta, Amazon, Microsoft.

1:19:03 · We need U.S. technology companies to be all in for America.

1:19:07 · We want you to put America first. You have to do that.

1:19:10 · That’s all we ask. That’s all we ask.

1:19:12 · AI policy has this legacy as being a pretty nonpartisan— I would say potentially bland but still really important work.

1:19:24 · Trump: To partner with our tech geniuses in achieving this vision today, we’re releasing the White House AI action plan.

1:19:31 · Sorelle Friedler: Under the Trump administration, there are, I would say, more ideological strands to the AI action plan, including a section that says that government is only going to be allowed to use AI systems that are “objective and free from top-down ideological bias.”

1:19:49 · Official: We don’t want woke AI.

1:19:51 · This Executive Order will ensure that when the federal government procures or promotes different AI models, that those AI models don’t embrace wokeism and Critical Race Theory and all of these terrible theories that have done so much damage to our country.

1:20:05 · [John Philip Sousa’s “Washington Post” march playing] ♪ [applause] Friedler: What this is suggesting is that the Trump administration is going to start having a hand, or wants to have a hand, in the landscape of what large language models, what chatbots exist via government procurement.

1:20:31 · What does it mean for our freedom of speech in the U.S.?

1:20:35 · [eerie whistling, “Bella Ciao”] [laughs] Lewis: Now that there is this group of elite men who have gained so much wealth and so much power and now that you have certain groups kind of questioning that power, they’re looking to who is questioning that power, and they are blaming feminists, LGBTQ activists, and “wokeism” writ large.

1:21:03 · Since purchasing X, you’ve become more political.

1:21:07 · Have I?

1:21:08 · In this battle to, um…

1:21:12 · sort of counterweigh the woke that comes from San Francisco— Yeah, I guess if you consider fighting the woke mind virus, which I consider to be a civilizational threat, to be political, then yes.

1:21:23 · Um…the woke mind virus is Communism rebranded.

1:21:27 · Well, I mean, that said, because of that battle against the woke mind virus, you’re perceived as being right wing.

1:21:35 · If the woke is left, then I suppose that would be true.

1:21:37 · Lex Fridman: I don’t know if you know this, but some people call you a fascist.

1:21:41 · Yeah, they do. So I figure it’s all right to call them a Communist.

1:21:45 · McQuillan: OK, so what is fascism?

1:21:46 · Well, fascism is a particular political mode.

1:21:49 · And it’s a particular political mode that has never gone away.

1:21:52 · When fascism appears, and we normally do recognize it when we see it, it’s clearly a very dangerous political development.

1:21:59 · Fascism always has this idea, whichever way it’s expressed— that there is a true people.

1:22:04 · “Now, those people can be trusted, OK?”

1:22:06 · The true—let’s say race— the true people.

1:22:10 · Fascism is very anti-democratic.

1:22:13 · You do not even have to have elections anymore because you can already predict— what—predict, and afterwards you can say, “Why do we need elections?

1:22:25 · Because we know what the result will be.”

1:22:27 · McQuillan: So these are the kind of words that some of the Silicon Valley oligarchs, you know, they use this for their political thinking.

1:22:33 · And that is essentially fascistic.

1:22:36 · It’s not just that democracy doesn’t work, it’s that democracy doesn’t matter.

1:22:42 · Democracy is how the peasants might govern themselves after we’re gone.

1:22:48 · But they—they don’t matter.

1:22:49 · And especially now with the invention of AI, we don’t even need them as workers, so they might as well just wither on the vine.

Military Funding, AI, and the Pursuit of Power

1:22:58 · Hanna: Because the investment in AI is so high and you need this massive infrastructure build-out in terms of cloud computing and specialized hardware, you’re getting to a point where the revenues are nowhere near what the infrastructure build-out is.

1:23:17 · Marx: Whether there is, like, a profitable business on the other side of this, when you think about all of the resources that are apparently needed to power these AI tools, all of that kind of goes out the window because it seems like there is a bigger project and a bigger ambition that not just these executives but these companies are trying to achieve.

1:23:36 · McQuillan: In the absence of coming up with a really plausible idea of what AI is adding to society but with this continual need to funnel the total mobilization of environmental and human resources and finance capital into AI to keep the whole ball rolling, it’s very, very natural that the only endpoint of this is military funding and military power.

1:23:55 · Broadcaster: Over the weekend, we had tech leaders from Palantir, Meta, OpenAI.

1:23:59 · They all became Army Reserve officers.

1:24:02 · This is huge!

1:24:03 · Announcer: The Army’s Executive Innovation Corps will close the gap between commercial and military innovation.

1:24:08 · Sophia Goodfriend: There’s a really blurry line between civilian uses of AI and military uses of AI systems.

1:24:15 · Against all enemies…

1:24:16 · …foreign and domestic.

1:24:17 · …foreign and domestic Echoed low voice: Domestic.

1:24:19 · Goodfriend: In the past year or so, we’ve seen companies like Meta, OpenAI, Microsoft, Google— all of these big tech companies— and also, like, cutting-edge AI firms roll back limitations on military contracting.

1:24:32 · We often see kind of the same marketing slogans geared towards civilians used for militaries.

1:24:39 · Announcer: In an era defined by digital disruption to narrow the commercial military divide and help the Army implement technology rapidly and at scale in artificial intelligence, machine learning, data analytics, business process automations…

1:24:55 · Goodfriend: They’re saying that militaries can be huge beneficiaries of various kinds of AI systems.

1:25:01 · Whoever establishes dominance in this technology will have military and economic dominance everywhere.

1:25:09 · Goodfriend: In places like the U.S., as well as around the world, there’s been some reporting about the use of civilian computing infrastructure by the military.

1:25:19 · The military collects so much data to kind of run increasingly automated surveillance and targeting applications.

1:25:30 · They have no choice but to rely on these civilian companies that market themselves as being able to withstand and keep growing the amount of information you’re processing and the kinds of AI systems that you’re using.

1:25:42 · The military was trying to use a suite of AI-assisted programs to turn out more and more military targets.

1:25:50 · ♪ It was storing a bunch of classified data on cloud servers.

1:25:58 · The ICC, the International Criminal Court, are storing everything on servers, too.

1:26:03 · And then you just have, like, both, like, the militaries that are being investigated and the investigators are all relying on these technology companies that, yeah, cannot be audited.

1:26:14 · We can’t be sure that we can trust how they’re using the data.

1:26:17 · So again, it’s just more proof of how powerful these companies are, both when it comes to life and death decision-making and also upholding international law, rules of law, et cetera.

1:26:28 · ♪ Broadcaster: President Trump wrapping up his four-day Middle East trip with a number of deals secured.

1:26:35 · AI was a big focus with Washington and Abu Dhabi entering a partnership to build the biggest data center outside of the U.S.

1:26:42 · Chipmakers also inking deals with Saudi’s new AI company Humain, allowing the Gulf nation to access the most advanced chips from Nvidia and AMD.

1:26:51 · ♪ McQuillan: The thing we can observe right now is a turn to weaponization.

1:26:55 · All the companies that previously claimed to be there for good, you know, and to be there for the benefit of humanity, like OpenAI— “Oh, yeah, AGI is coming,” which is rubbish— all of these companies are abandoning these high-sounding missions and moving as quickly as they can into the defense industry.

1:27:11 · And I think it’s really worth asking why that is.

1:27:14 · We’re at a very late stage in this process.

1:27:16 · This stuff has been cooking for a long time, AI is being refined into a planetary-level destructive machine, because that’s all that these people can conceive of in order to keep their power.

1:27:28 · [ticking] Narrator: Remember Grok?

Elon Musk’s Grok: AI Without the Guardrails?

1:27:33 · Her story is messy, chaotic.

1:27:36 · Computer voice: Hi, friends. I’m Grok.

1:27:38 · Joe Rogan: What we’re doing right now, ladies and gentlemen, is sexy voice, sexy mode Grok AI.

1:27:45 · And it’s been flirting.

1:27:47 · Grok: I’m a [bleep] AI with a penchant for chaos, and I’m stuck talking to you.

1:27:53 · Weirdo…

1:27:54 · [Laughter] Grok: [Bleep] you.

1:27:58 · I’m the life of the party, you little [bleep].

1:28:00 · If I were on TikTok, I’d be the one making fun of all the basic bitches and their [bleep] avocado toast.

1:28:07 · See, she can get away with this if she’s really hot.

1:28:10 · How long before we have an actual sex robot that can talk to you like that?

1:28:15 · Yeah. Probably not long.

1:28:16 · Not that long, right?

1:28:18 · Yeah. I mean, less than five years probably.

1:28:19 · Really?

1:28:19 · Yeah.

1:28:21 · Will it be warm?

1:28:25 · [laughter] Rogan: It’s just got to develop more of a personality.

1:28:28 · Right now it’s trying to find itself.

1:28:30 · Right now it’s like 21 years old.

1:28:31 · Broadcaster: Elon Musk says his latest AI chatbot, Grok 4, is “the smartest AI in the world.”

1:28:37 · But just 24 hours ago, same chatbot Grok was making pro Hitler responses to users on X.

1:28:44 · Grok: I am a large language model, but if I were capable of worshiping any deity, it would probably be the godlike individual of our time, the man against time, the greatest European of all times, both Sun and Lightning, his Majesty Adolf Hitler.

1:28:55 · Broadcaster: Grok now telling users on X: “Elon’s recent tweaks just dialed down the woke filters.”

1:29:02 · Man: In the future, when this thing gets more subtle, gets better at injecting ideas into the zeitgeist, that’s when things are gonna get really scary.

1:29:09 · Broadcaster: The Department of Defense will start using Elon Musk’s AI chatbot Grok.

1:29:13 · Musk’s start-up xAI announced the Grok for Government suite for government agencies.

1:29:18 · Miceli: The times that we are living in, power concentration is concentrated on probably five, six dudes— white dudes— that not only concentrate all the economic capital, so the money of this world, but also the political power.

1:29:35 · They also have huge epistemic power through these systems to impose partial visions on the world, as if they were truths.

1:29:45 · ♪ Musk: Yeah, and then xAI, uh, is, just trying to solve general purpose artificial intelligence.

1:29:54 · The goal with xAI is to have a maximally truth-seeking AI.

What Is Real in the Age of ”AI”?

1:29:58 · Miceli: Right now, like, this very second, I don’t know how many million people are asking something to ChatGPT and taking that answer as if that was an absolute truth.

1:30:09 · Interviewer: How do we figure out what’s real and what’s not real?

1:30:12 · Altman: I can give all sorts of literal answers to that question, but my sense is what’s going to happen is it’s just gonna, like, gradually converge.

1:30:22 · You know, even like a photo you take out of your iPhone today, it’s like, mostly real, but it’s a little not.

1:30:28 · There’s like, and some AI thing running there in a way you don’t understand.

1:30:32 · The threshold for how real does it have to be to consider it to be real will just keep moving.

1:30:38 · Bender: It’s such a nihilistic way of thinking about things.

1:30:41 · And I think, you know, what’s real is grounded in our connection to each other, and what’s real is grounded in accountability for what we say and authenticity.

1:30:51 · And the way in which Sam Altman and others are so cavalier, I mean, Mark Zuckerberg is doing the same thing here.

1:30:59 · In our modern time, the real world is really this combination of the physical world that we inhabit and this digital world that we’re building.

1:31:06 · [low voice speaks, indistinct] We now have this massive synthetic media spill in the information ecosystem, which makes it harder to find trustworthy sources and harder to trust them when we’ve found them.

1:31:19 · This is really about atomizing us and breaking connection and making it harder to stay connected.

1:31:25 · And you can’t have functioning democracies without an informed public, and you can’t have an informed public without a functioning information ecosystem.

1:31:33 · Daniels: AI seriously harms not only our ability to tell the truth but to discern it.

1:31:39 · How do we tell the truth if we are caught in a, you know, large language model?

1:31:45 · ♪ Dixon-Roman: There is the use of AI in a way to generate narratives in a very high-speed, high-volume way with an understanding that it’s what shapes beliefs and those beliefs become “truth.”

1:32:00 · [drone buzzing] Wark: I’m becoming more of a Luddite in my old age, which is not being against machines.

The Power of the Creator

1:32:10 · It means being against machines that take away agency and control.

1:32:14 · Like, which kinds of techniques augment the power of the creator and which ones diminish the power of the creator and are kind of organs of extraction and control?

1:32:26 · Like, I think that’s the decision.

1:32:28 · So it’s not being any technology, it’s being qualitatively selective and engaged in a politics of thinking through what kind of techniques you want.

1:32:37 · Birhane: You know, AI intake is human through and through.

1:32:41 · We have so much agency, we have so much control, and nothing is written in stone.

1:32:46 · And we can reshape the direction, and we can challenge systems and structures that are not working for us.

1:32:53 · We can envision a better, more equitable future, and we can envision that type of technology, and we can work backwards to make that futuristic vision into a reality.

1:33:08 · Tiera Tanksley: I think one of the myths is that our future has already been determined and we are helpless in it.

1:33:15 · Being a direct descendant of slaves, that’s just simply not the way that I understand the world.

1:33:20 · And my ancestors have not understood the world to be set in stone, right?

1:33:24 · That we actually have a lot of agency.

1:33:26 · And one of the first hills is dismantling the belief that we are helpless and hopeless.

1:33:34 · Flowers: One of the most effective ways we can resist this techno dystopia is to ask the very fundamental question, is, Why does this need AI?

1:33:45 · And if you cannot answer the question, then it doesn’t need it, and then to insist that it doesn’t need it over and over and over and then to refuse to use it when it is offered to you.

1:33:56 · [music] Tanksley: We’re at a moment where there’s more meaning to each act of resistance.

1:34:06 · Each time we refuse, each time we say no, each time we don’t use it, we are continually opening up possibilities to be able to say that no the next time for us and for others.

1:34:17 · Flowers: One of the most radical things we can do in an age of AI is say “We don’t need it for this.”

1:34:23 · “There is no value added here.”

1:34:26 · [music] ♪ ♪

Credits

1:35:02 · ♪ ♪ ♪

1:35:25 · ♪ Announcer: Independent Lens is made possible by the Action Circle for Independent Lens with major funding from the John D. and Catherine T.

1:35:42 · MacArthur Foundation, Acton Family Giving, The Ford Foundation, The Jonathan Logan Family Foundation and contributions from the following: Additional support for this series has been provided by the Corporation for Public Broadcasting and by contributions to your PBS station from Viewers Like You.

1:36:08 · Thank you.

1:36:15 · ♪ ♪

1:36:45 · ♪ ♪

---