What kind of thing is generative AI, and does the answer predict anything?

SocioLens project,

Living Debate LD-01

What kind of thing is generative AI, and does the answer predict anything?

In short

The best-supported description of the evidence to October 2026 is:
Generative AI is several kinds of thing at once, depending on the level described. The accounts rarely compete on the same observations, and the evidence gathered so far mostly fits more than one of them.

Read the full position

SocioLens’s position in version v0.1, quoted unbroken from the start of this debate’s “Current SocioLens position” (section 14), where it is marked INTERPRETATION. Version v0.1, last updated 16 October 2026. Lines the debate sets in capitals are shown in sentence case; the words are unchanged. It sums up the evidence on this page; it is not the finding of any single study. See its limits: what we still do not know and what evidence would change the picture. To quote it, use the reference under Cite this page. How this page was made and checked: About this page, just below.

Living Debate LD-01, version v0.1. Status (verbatim): Living Debate v0.1 — PUBLISHED.

EVIDENCE CUT: 16 October 2026 (2026-10-16)
LAST UPDATED: 16 October 2026 (2026-10-16)
Last substantive review: not stated in this version of the debate

Version

This page shows version v0.1. Its record never changes, even when newer evidence arrives; the site may add a note to the page or withhold a figure not cleared for public use.

This pinned version is also the current version.

Versions: v0.1 · current

The question

Researchers, firms, regulators and governments describe generative AI in incompatible terms: a tool, a normal technology, a cultural technology, an agent, a partner in communication, an infrastructure, an investment boom. Which description is best, and does the choice predict anything a researcher could observe?

Is the best description:

a normal technology: a powerful general-purpose tool whose effects arrive slowly, at the pace of institutions;
a cultural and social technology: an engine that imitates and recombines what people have written, like writing, print or libraries;
an agent: a system that pursues goals and acts with growing independence;
artificial communication: a partner in communication that does not understand, so that what matters happens on the human side;
an infrastructure: a foundation layer and a cloud stack that a few firms control and many others build on;
a capital formation: an investment and expectations phenomenon whose valuations run ahead of what is known;
or several of these at once, each true of a different level?

Answer in this version

The accounts are better treated as answers to different questions than as rivals: what the model does, how it spreads, who controls it and how it is financed. Few of them state what one would observe if they were true, and most of the evidence fits more than one.

Only two accounts have stated predictions that have been checked or can be checked directly. The normal-technology account predicts slow diffusion in high-consequence tasks and no meaningful AI advantage over trained humans at forecasting geopolitical events or at persuading people to act against their own self-interest, a qualifier the authors call critical; a cultural-technology study predicted that models would imitate well and innovate poorly, and reported preliminary results in line with that.
Studies of what is inside models rule out only the strongest "mere surface statistics" claim, and only for the systems tested. Studies of failures show behaviour shaped by the training objective, which fits mimicry and a tool with known failure modes (a SocioLens gloss of the normal-technology view) equally.
The length of software tasks that AI agents complete has grown fast on one organization's measure, which is a precondition for the agent account, not a test of it.
Adoption is fast by headcount and slow by intensity, so both the normal-technology account and its rivals can claim the same survey; the study's revision changed which comparison leads.
One infrastructure source predicts shared failure and rising barriers to entry, which the sources read here do not measure; the capital-formation account describes asset markets, and the Bank for International Settlements found the recent rally of the largest technology stocks unlike the dot-com bubble because earnings have grown, while noting concerns about stretched valuations.
Law and policy have already chosen: the Commission's guidelines for the EU AI Act adopt training compute as an indicative criterion for a general-purpose AI model, which they call an imperfect proxy, and China's State Council sets targets for "intelligent agents."
No pre-registered comparison of all six accounts' predictions was found. A joint essay by the normal-technology and AI 2027 authors reportedly proposes ways to tell their two views apart; it could not be read, and the search for this version was web-based and partial.

The central question is therefore not:

"What is generative AI, really?"

It is:

Which description of generative AI predicts something the others do not, and at which level: the model, its use, its ownership or its finance?

Lines the debate sets in capitals are shown in sentence case; the words are unchanged.

Current SocioLens position

The whole of this debate’s “Current SocioLens position” (section 14 of its text), quoted verbatim. Version v0.1, last updated 16 October 2026.

INTERPRETATION

The best-supported description of the evidence to October 2026 is:

Generative AI is several kinds of thing at once, depending on the level described. The accounts rarely compete on the same observations, and the evidence gathered so far mostly fits more than one of them. The normal-technology account states the most predictions, though without thresholds or dates; no pre-registered test of which account predicts something the others forbid was found, though the normal-technology and AI 2027 authors have begun to look for distinguishing measures.

At the level of the model, the evidence supports uneven competence shaped by the training objective (P2, and P1 read as a tool with known failure modes), not pure surface statistics. At the level of use, adoption is wide and shallow so far, consistent with P1 and equally with a general-purpose technology on a J-curve, and agents' measured task length grows (P3's precondition). At the level of control, the infrastructure account (P5) is well argued and little measured. At the level of finance, P6 describes valuations and is compatible with every other account. The artificial-communication account (P4) points to human-side observables that remain largely unstudied.

Lines the debate sets in capitals are shown in sentence case; the words are unchanged.

What we still do not know

UNKNOWN

Which observations would one account predict and another forbid, stated before the data arrive?
Do the failure patterns of 2023–2024 models hold for later reasoning and agentic systems?
Do deployed agents pursue goals or show initiative in organizations, beyond completing longer tasks?
How fast is generative AI integrated into firms' processes and hours of work, rather than tried by individuals?
Are failures and biases correlated across products built on the same base models, as the infrastructure account predicts?
What does the artificial-communication account predict about accountability and uptake, and can it be measured?
How do people classify these systems, as tool, agent or partner, outside the US and over time?
Will the EU's compute criterion and China's targets shape what is built and counted?
Does training compute track the generality that the EU's criterion is meant to capture?

What evidence would change the picture

A pre-registered comparison of the accounts' predictions, with thresholds and dates;
the normal-technology predictions on geopolitical forecasting and on persuasion against self-interest scored against trained human teams;
studies of later models' failures on the same tasks;
evidence of goals or initiative in deployed agents, or its absence;
measures of integration by firms and tasks, not only by users;
measures of correlated failure across products built on shared base models;
cross-national studies of how people classify these systems;
measures of generality across models around the compute threshold;
the full texts of the cultural-technology, "stochastic parrots" and artificial-communication statements.

Positions

Under each position, its “Claim:” line and the first sentence of its “Current status:”, quoted verbatim from the debate’s text. These are the positions’ own statements, not the evidence claims listed below. Position names the debate sets in capitals are shown in sentence case; the words are unchanged.

P1 — Normal technology
Claim: Generative AI is a powerful general-purpose technology whose effects arrive slowly, at the pace of institutional adaptation, with people in control.
Current status: The account with the most stated predictions.
P2 — Cultural and social technology
Claim: Generative AI is an engine for imitating and recombining accumulated human knowledge, comparable to writing, print or libraries.
Current status: One experiment reported preliminary results in line with its stated prediction on 2023 models.
P3 — Agent
Claim: Generative AI systems pursue goals and act with growing independence, so they should be studied and governed as agents.
Current status: Defined as a matter of degree or by control flow, not as a type.
P4 — Artificial communication
Claim: Generative AI takes part in communication without understanding; what matters is how people communicate with it.
Current status: Explicitly rejects behavioural tests in one recent reading.
P5 — Infrastructure
Claim: Generative AI is a foundation layer and a cloud stack on which others build and depend.
Current status: One source states predictions of shared failure, inherited flaws and rising barriers to entry; the sources read here do not measure them.
P6 — Capital formation
Claim: Generative AI is above all an investment and expectations phenomenon whose valuations run ahead of what is known.
Current status: Describes finance, not the technology, and so can hold together with any other account.

Claims and evidence

Each claim card shows its source and what the claim does not license. Open "What would change this claim" on a card for the rest.

Narayanan and Kapoor describe AI as “normal technology” and predict that slow diffusion “will continue to be the norm in high-consequence tasks”; they argue that diffusion is limited by the speed at which individuals, organizations and institutions adapt, and that for past general-purpose technologies it “occurs over decades, not years.”

InterpretationContested evidenceNot applicable / English

Source (as cited): Narayanan & Kapoor, “AI as Normal Technology: An alternative to the vision of AI as a potential superintelligence,” Knight First Amendment Institute, 15 April 2025
Type of source: Independent research report (recorded as “Essay (research institute publication)”).
Does not license: A finding that diffusion has been slow; the essay states predictions, not tests

What would change this claim, and its limitations

Sustained fast diffusion of AI into high-consequence tasks, or institutions adapting within years rather than decades

Limitations: The authors say the essay states a worldview and does not rebut counterarguments point by point; predictions are directional, with no thresholds or dates; much of the evidence offered concerns older predictive machine learning rather than generative AI

Source and status

Publication status: Eligible for publication, with qualifiers: interpretation label
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Narayanan & Kapoor, “AI as Normal Technology: An alternative to the vision of AI as a potential superintelligence,” Knight First Amendment Institute, 15 April 2025: no checked link to the original yet.

Claim LD01-SL-C01 (v1.0)

Narayanan and Kapoor predict that AI will not “meaningfully outperform trained humans (particularly teams of humans and especially if augmented with simple automated tools)” at forecasting geopolitical events or at persuading people to act against their own self-interest, and that “an increasing percentage of human jobs and tasks will be related to AI control.”

InterpretationContested evidenceNot applicable / English

Source (as cited): Narayanan & Kapoor, “AI as Normal Technology: An alternative to the vision of AI as a potential superintelligence,” Knight First Amendment Institute, 15 April 2025
Type of source: Independent research report (recorded as “Essay (research institute publication)”).
Does not license: Evidence that these predictions have held or failed

What would change this claim, and its limitations

AI that meaningfully outperforms teams of trained forecasters, or reliable persuasion against self-interest

Limitations: The authors say “it is not clear how much slack to allow” in making the predictions precise; no date or threshold is set, so the predictions are hard to score as failed

Source and status

Publication status: Eligible for publication, with qualifiers: interpretation label
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Narayanan & Kapoor, “AI as Normal Technology: An alternative to the vision of AI as a potential superintelligence,” Knight First Amendment Institute, 15 April 2025: no checked link to the original yet.

Claim LD01-SL-C02 (v1.0)

Reading a 2024 study that found 40% of US adults used generative AI, Narayanan and Kapoor note that infrequent use translated into only 0.5%–3.5% of work hours, and argue that, depending on how adoption is measured, generative AI adoption may have been much slower than personal-computer adoption.

InterpretationContested evidenceUS / English

Source (as cited): Narayanan & Kapoor, “AI as Normal Technology: An alternative to the vision of AI as a potential superintelligence,” Knight First Amendment Institute, 15 April 2025
Type of source: Independent research report (recorded as “Essay (research institute publication)”).
Does not license: A measurement of its own; the authors reinterpret another study’s figures

What would change this claim, and its limitations

Intensity measures (hours, tasks, firm processes) rising as fast as headline adoption

Limitations: The study they cite was revised in February 2025; both versions compare overall adoption with PCs and the internet and work adoption with PCs at work, and the revision leads with the work comparison (LD01-SL-C15); the essay does not say which measure of intensity would settle the question

Source and status

Publication status: Eligible for publication, with qualifiers: interpretation label
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Narayanan & Kapoor, “AI as Normal Technology: An alternative to the vision of AI as a potential superintelligence,” Knight First Amendment Institute, 15 April 2025: no checked link to the original yet.

Claim LD01-SL-C03 (v1.0)

Yiu, Kosoy and Gopnik describe large language models as “cultural technologies” and “efficient imitation engines,” and predicted that they would be almost as capable as humans at linking familiar objects but fail at tool innovation; in their study, children chose the functionally right novel tool 85.2% of the time and adults 95.7%, while the models tested were “much less capable.”

Estimate and InterpretationEmerging evidenceNot applicable / English

Source (as cited): Yiu, Kosoy & Gopnik, “Imitation versus Innovation: What children can do that large language and language-and-vision models cannot (yet)?”, arXiv:2305.07666, 2023
Type of source: Preprint (recorded as “Preprint (a journal version was published in Perspectives on Psychological Science)”).; Read the original
Does not license: A claim about later reasoning or agentic models, or a test that separates the cultural-technology account from the mimicry or normal-technology accounts

What would change this claim, and its limitations

Models that select functionally right but superficially unrelated objects at human rates

Limitations: Read in arXiv version 1 (dated 8 May 2023 from the PDF metadata only), which reports preliminary results (“so far”); the journal version was not compared; small human samples; 2023 models; model scores are described, not given as numbers; one model tested is made by Anthropic, the developer of the model that drafted this debate

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C04 (v1.0)

Mitchell and Krakauer report that in a 2022 survey of natural-language-processing researchers, 51% of 480 respondents agreed and 49% disagreed that a generative model trained only on text could, with enough data and compute, understand natural language “in some non-trivial sense”; they argue for an extended science of intelligence that recognizes distinct modes of understanding.

Estimate and InterpretationContested evidenceNot applicable / English

Source (as cited): Mitchell & Krakauer, “The Debate Over Understanding in AI’s Large Language Models,” arXiv:2210.13966, 2023
Type of source: Preprint (recorded as “Preprint (a journal version was published in PNAS)”).; Read the original
Does not license: A view of the public, or a finding on whether models understand

What would change this claim, and its limitations

A later survey of the same population with a clear majority either way

Limitations: Read in arXiv version 3 (10 February 2023); the PNAS version was not compared; the 51/49 split comes from a survey by other researchers that the paper cites; views from 2022 may have moved

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C05 (v1.0)

Shanahan, McDonell and Reynolds argue that a dialogue agent is best understood as role-play by a simulator, not as an entity with its own agenda: an agent frozen before the 2022 World Cup that names France as the current champion behaves as a simulator role-playing a knowledgeable person would, not as a helpful person would.

InterpretationEmerging evidenceNot applicable / English

Source (as cited): Shanahan, McDonell & Reynolds, “Role-Play with Large Language Models,” arXiv:2305.16367, 2023
Type of source: Preprint (recorded as “Preprint (a journal version was published in Nature)”).; Read the original
Does not license: A test of the framing, or a claim that tool use leaves the distinction intact; the authors say role-play with real effects through tools makes it thin

What would change this claim, and its limitations

Behaviour that training-data role-play cannot account for

Limitations: Read in arXiv version 1 (25 May 2023); the Nature version was not compared; conceptual, no experiments; the authors limit the paper to base models and say fine-tuning may blur the distinction between simulator and simulacra; one author works for a developer of such systems

Source and status

Publication status: Eligible for publication, with qualifiers: interpretation label
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C06 (v1.0)

Millière and Buckner take the view that language models are “merely sophisticated mimics” as a null hypothesis to be tested; they find that models show “at least local task generalization, if not broad task generalization,” but “often seem to have a hard time pushing beyond the range of tasks well-represented in their training data,” and say their part in cumulative cultural learning “appears contingent on further advancements” that “might lie beyond the reach of current architectures.”

InterpretationContested evidenceNot applicable / English

Source (as cited): Millière & Buckner, “A Philosophical Introduction to Language Models — Part I: Continuity With Classic Debates,” arXiv:2401.03910, 2024
Type of source: Preprint (recorded as “Preprint”).; Read the original
Does not license: A verdict that models understand or do not

What would change this claim, and its limitations

Evidence of models reliably extending tasks well beyond their training distribution

Limitations: Read in arXiv version 1 (8 January 2024); the claim rests on section 3.5 and the conclusion, which were read closely, and other sections were not; a review, so it reflects the literature to early 2024; the companion part on empirical methods was not read

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C07 (v1.0)

Esposito argues that comparing algorithms with human intelligence is misleading and proposes thinking of them “not in terms of artificial intelligence but in terms of artificial communication”: if machines contribute to social intelligence, “it will not be because they have learned how to think like us but because we have learned how to communicate with them.”

InterpretationEmerging evidenceNot applicable / English

Source (as cited): Esposito, Artificial Communication: How Algorithms Produce Social Intelligence, MIT Press, 2022
Type of source: Book (monograph) (recorded as “Monograph”).; Read the original
Does not license: A statement of what this account predicts researchers would observe; the abstract states none, and the book was not read

What would change this claim, and its limitations

A statement in the book of observable differences that the account predicts

Limitations: Only the publisher’s abstract was read; the book could not be reached; written in 2022 about machine learning generally, before ChatGPT

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the abstract of the primary source only. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C08 (v1.0)

Drawing on Luhmann, Esposito, Wittgenstein and Brandom, Fenoglio argues that what language models lack is a position in the practices of commitment and assessment, not a behavioural capacity, so that “no behavioral evidence can satisfy a positional criterion, and this is by design rather than by evasion.”

InterpretationEmerging evidenceNot applicable / English

Source (as cited): Fenoglio, “Asymmetric Communication: Large Language Models and Language Games,” arXiv:2607.28137, 2026
Type of source: Preprint (recorded as “Preprint”).; Read the original
Does not license: A statement of Esposito’s own view; this is one author’s reading of the tradition

What would change this claim, and its limitations

A version of the account that names behavioural or institutional observables

Limitations: Read in arXiv version 1 (30 July 2026); single author; not peer-reviewed; a secondary account of Esposito, not her text

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C09 (v1.0)

Chan and colleagues treat agency in algorithmic systems as a matter of degree, raised by four characteristics: underspecification, directness of impact, goal-directedness and long-term planning; they stress that recognizing such agency “does not absolve or shift” human responsibility.

InterpretationEmerging evidenceNot applicable / English

Source (as cited): Chan et al., “Harms from Increasingly Agentic Algorithmic Systems,” FAccT 2023
Type of source: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”).; Read the original
Does not license: A measurement of agency in deployed systems

What would change this claim, and its limitations

A validated measure of the four characteristics in deployed systems

Limitations: Read in arXiv version 2 (12 May 2023), the FAccT 2023 paper; written before tool-using agent products were common; the harms are anticipated and the cited evidence is partial

Source and status

Publication status: Eligible for publication, with qualifiers: interpretation label
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C10 (v1.0)

Kolt notes that the US Restatement (Third) of Agency holds that a computer program “is not capable of acting as a principal or an agent as defined by the common law” and that “at present, computer programs are instrumentalities” of their users, and he uses agency law and principal–agent economics to analyse AI agents’ problems of information asymmetry, discretionary authority and loyalty.

Fact and InterpretationEmerging evidenceUS / English

Source (as cited): Kolt, “Governing AI Agents,” arXiv:2501.07913, 2025 (forthcoming in the Notre Dame Law Review)
Type of source: Preprint (recorded as “Preprint (law review article, forthcoming)”).; Read the original
Does not license: A claim that AI systems have intentions, or a statement of current law outside the US Restatement

What would change this claim, and its limitations

Courts or a revised Restatement treating an AI system as an agent in law

Limitations: Read in the arXiv PDF dated 11 February 2025 (the version number is not stated in the file); the published law-review version was not compared; the Restatement wording is quoted by the author and was not checked against the Restatement itself

Source and status

Publication status: Eligible for publication, with qualifiers: legal scope
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C11 (v1.0)

Kasirzadeh and Gabriel characterize AI agents along four dimensions, autonomy, efficacy, goal complexity and generality, each with gradations, and draw “agentic profiles” for systems including AlphaGo, ChatGPT-3.5, Claude 3.5 Sonnet with tool use and a self-driving car.

InterpretationEmerging evidenceNot applicable / English

Source (as cited): Kasirzadeh & Gabriel, “Characterizing AI Agents for Alignment and Governance,” arXiv:2504.21848, 2025
Type of source: Preprint (recorded as “Preprint”).; Read the original
Does not license: A measurement; the profiles are the authors’ qualitative judgments

What would change this claim, and its limitations

Profiles scored reliably by independent raters, or by measurement

Limitations: Read in arXiv version 1 (30 April 2025); no testable prediction is stated; one author works for Google DeepMind; the profiles cover systems from Google DeepMind (AlphaGo), Alphabet’s Waymo, OpenAI and Anthropic (Claude 3.5 with tool use), the developer of the model that drafted this debate

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C12 (v1.0)

Anthropic distinguishes “workflows,” in which language models and tools are orchestrated through predefined code paths, from “agents,” in which models “dynamically direct their own processes and tool usage”; it says agents are “typically just LLMs using tools based on environmental feedback in a loop.”

FactEmerging evidenceGlobal / English

Source (as cited): Anthropic, “Building effective agents,” Anthropic engineering post, 19 December 2024, as served on 16 October 2026
Type of source: Vendor or company document (recorded as “AI developer web page”).; Read the original
Does not license: Evidence that such systems are agents in any stronger sense

What would change this claim, and its limitations

A revised definition from the developer

Limitations: An engineering post, not research; read as served on 16 October 2026, a revised page that mentions later products (Claude Haiku 4.5, the Agent SDK); the December 2024 text was not compared; under this definition almost any model in a loop is an agent

Source and status

Publication status: Eligible for publication, with qualifiers: source position
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C13 (v1.0)

Eloundou, Manning, Mishkin and Rock conclude that language models “exhibit traits of general-purpose technologies,” estimating that about 80% of the US workforce could have at least 10% of their tasks affected and about 19% at least half; they “do not make predictions about the development or adoption timeline.”

Estimate and InterpretationEmerging evidenceUS / English

Source (as cited): Eloundou, Manning, Mishkin & Rock, “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models,” arXiv:2303.10130, 2023
Type of source: Preprint (recorded as “Preprint (working paper; a journal version was published in Science)”).; Read the original
Does not license: Adoption, productivity or employment effects; the authors say exposure does not guarantee them

What would change this claim, and its limitations

Exposure scores that fail to predict later adoption or task change

Limitations: Read in arXiv version 5 (21 August 2023), labelled a working paper; the Science version was not compared; three authors work for the developer of the models assessed, and one of two annotators is that developer’s model; the human annotators, who include the authors, are a group the paper says “is not occupationally diverse”

Source and status

Publication status: Eligible for publication, with qualifiers: institutional position
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C14 (v1.0)

In nationally representative US surveys from late 2024, nearly 40% of adults aged 18–64 used generative AI; 27% of workers used it for work two years after ChatGPT’s release, against 25% using computers at work in 1984, three years after the IBM PC, while only 1%–5% of all work hours were assisted by it.

EstimateEmerging evidenceUS / English

Source (as cited): Bick, Blandin & Deming, “The Rapid Adoption of Generative AI,” NBER Working Paper 32966, 2024 (revised February 2025)
Type of source: Working paper (recorded as “Academic working paper”).; Read the original
Does not license: A claim that generative AI is diffusing faster in depth, or a claim about productivity

What would change this claim, and its limitations

Measures of intensity or firm integration that rise as fast as individual adoption

Limitations: Read in the February 2025 revision and the September 2024 original; both versions compare overall adoption with PCs and the internet (39.5% against 20% in September 2024, 39% against 20% in February 2025) and work adoption with PCs at work (28% against 25% in September 2024, 27% against 25% in February 2025), and the revision leads with the work comparison; the revision also reports 26% of employed respondents using it for work in August and November 2024 and 23% in the previous week; the launch dates chosen drive the comparison; adoption means any use in the reference period; time savings are self-reported; a working paper, not peer-reviewed

Source and status

Publication status: Eligible for publication, with qualifiers: working paper
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C15 (v1.0)

A GPT-style model trained only on sequences of legal Othello moves, with no rules or board given, made illegal moves 0.01% of the time; its activations encoded the board state, and editing that internal state changed its predicted moves as the edited board implied.

EstimateEmerging evidenceNot applicable / English

Source (as cited): Li, Hopkins, Bau, Viégas, Pfister & Wattenberg, “Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task,” ICLR 2023
Type of source: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”).; Read the original
Does not license: A claim that deployed language models hold comparable models of the world, or that they understand

What would change this claim, and its limitations

Evidence that the interventions work through features other than a board representation

Limitations: Read in arXiv version 5 (26 June 2024); a closed synthetic world, not natural language; linear probes did not recover the board well in this setting, so the evidence rests on nonlinear probes and interventions

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C16 (v1.0)

Linear probes on Llama-2-70B’s activations predicted the locations of world places with out-of-sample R² of 0.911 and the death years of historical figures with 0.835; the authors say these representations “do not constitute a dynamic causal world model in their own right.”

EstimateEmerging evidenceMostly English

Source (as cited): Gurnee & Tegmark, “Language Models Represent Space and Time,” ICLR 2024
Type of source: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”).; Read the original
Does not license: Evidence that models use these representations as a causal model of the world, which the authors “conjecture, but do not show”

What would change this claim, and its limitations

Causal tests showing the models do or do not use these representations

Limitations: Read in arXiv version 3 (4 March 2024); a probe can decode information a model does not use; held-out-block tests raise errors, which the authors read as weak evidence that probes partly learn the mapping themselves

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C17 (v1.0)

McCoy and colleagues predicted that, because language models are trained to predict the next word, their accuracy should depend on how probable the task, input and output are, even where probability should not matter; GPT-4 decoded a simple cipher with 51% accuracy when the output was a high-probability word sequence and 13% when it was low-probability.

Estimate and InterpretationEmerging evidenceEnglish task set

Source (as cited): McCoy, Yao, Friedman, Hardy & Griffiths, “Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solve,” arXiv:2309.13638, 2023
Type of source: Preprint (recorded as “Preprint (a revised version was published in PNAS)”).; Read the original
Does not license: A claim that models cannot do these tasks, or that they lack understanding; the authors say the teleological view can be incomplete

What would change this claim, and its limitations

Newer models whose accuracy no longer depends on output probability

Limitations: Read in arXiv version 1 (24 September 2023), which covers GPT-3.5 and GPT-4 with basic prompting; the PNAS version (2024) was not compared; frontier models have changed since

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C18 (v1.0)

Models fine-tuned on facts of the form “A is B” failed to answer “B is A”; GPT-4 named the parent of 79% of 1,000 celebrities, but on the 1,573 child–parent pairs this produced it named the celebrity from the parent only 33% of the time.

EstimateEmerging evidenceEnglish task set

Source (as cited): Berglund, Tong, Kaufmann, Balesni, Cooper Stickland, Korbak & Evans, “The Reversal Curse: LLMs trained on ‘A is B’ fail to learn ‘B is A’,” ICLR 2024
Type of source: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”).; Read the original
Does not license: A claim about later models or about reasoning in context

What would change this claim, and its limitations

Models trained in the ordinary way that answer reversed questions as well as forward ones

Limitations: Read in arXiv version 4 (May 2024); the GPT-4 test is not controlled because its training data are unknown; humans also show some direction effects in recall

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C19 (v1.0)

Adding a clause that seems relevant but does not change the answer to grade-school maths problems cut accuracy by up to 65 percentage points across 25 models; GPT-4o scored 95.2% on the full GSM8K test set and 63.1% with the clause added, and OpenAI’s o1-preview 94.9% and 77.4%.

EstimateEmerging evidenceEnglish task set

Source (as cited): Mirzadeh, Alizadeh, Shahrokhi, Tuzel, Bengio & Farajtabar, “GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models,” ICLR 2025
Type of source: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”).; Read the original
Does not license: A finding that models cannot reason; the authors state their pattern-matching reading as a hypothesis

What would change this claim, and its limitations

Humans showing similar drops on the same items, or newer models showing none

Limitations: Read in arXiv version 2 (27 August 2025); no human baseline on the irrelevant clauses; 2024 models; the authors work for a company that sells AI products; a 2026 re-analysis of 20 open-weight models (Długosz, Oliveira & Díaz-Rodríguez, arXiv:2605.28700) disputes the statistical basis of the finding on changed names and numbers but leaves the irrelevant-clause variant out of scope

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C20 (v1.0)

Off-the-shelf ChatGPT and GPT-4 reached only 55% and 59% accuracy on 3-digit by 3-digit multiplication, and accuracy on compositional tasks fell towards zero as complexity grew; the authors hypothesize that transformers reduce multi-step reasoning to “linearized path matching.”

Estimate and InterpretationEmerging evidenceEnglish task set

Source (as cited): Dziri et al., “Faith and Fate: Limits of Transformers on Compositionality,” NeurIPS 2023
Type of source: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”).; Read the original
Does not license: A claim about later reasoning models

What would change this claim, and its limitations

Models that keep accuracy as the size of compositional problems grows

Limitations: Read in arXiv version 3 (31 October 2023); 2023 models without later reasoning methods; the pattern-matching reading is inferred from frequency correlations

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C21 (v1.0)

On 11 tasks rewritten with changed premises, such as arithmetic in base 9 or 1-based indexing in code, language models performed above chance but their performance “substantially and consistently degrades” relative to the familiar versions, which the authors attribute partly to overfitting to the default tasks.

Estimate and InterpretationEmerging evidenceEnglish task set

Source (as cited): Wu et al., “Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks,” NAACL 2024
Type of source: Peer-reviewed conference paper (recorded as “Peer-reviewed conference paper”).; Read the original
Does not license: A claim that models only recite, or that they reason generally; the authors find evidence of both

What would change this claim, and its limitations

Models that perform as well on counterfactual as on default versions

Limitations: Read in arXiv version 3 (28 March 2024), which carries no venue statement; the NAACL 2024 version was not compared; 2023 models; the authors note humans may also do worse in unfamiliar conditions; one model tested is made by Anthropic, the developer of the model that drafted this debate

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C22 (v1.0)

METR researchers estimate that the length of software tasks, measured in human time, that AI agents complete with 50% success doubled about every 207 days (95% interval 166–240) from 2019 to 2025, while performance was “much lower on less structured, ‘messier’ tasks.”

EstimateEmerging evidenceNot applicable / English

Source (as cited): Kwa et al., “Measuring AI Ability to Complete Long Software Tasks,” arXiv:2503.14499v4, 2026
Type of source: Preprint (recorded as “Preprint (arXiv version 4, marked “Under review”)”).; Read the original
Does not license: A finding that systems act as agents in organizations, set goals or take initiative; or that the trend holds for messy real-world work

What would change this claim, and its limitations

A trend that breaks on realistic tasks, or one that holds on them

Limitations: Read in arXiv version 4 (10 July 2026), marked “Preprint. Under review.”; no venue is confirmed in the file read; tasks are self-contained and automatically scored; the result depends on the human baseliner population; one author works for Anthropic, the developer of the model that drafted this debate, and Anthropic models, including Claude 3.7 Sonnet, are among those measured

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C23 (v1.0)

After METR revised its task suite and moved its evaluation software from Vivaria to Inspect in January 2026, its estimate of the post-2023 doubling time of AI time horizons moved from 165 to 131 days; METR says the trend “is somewhat sensitive to task composition” and that it had measured human baseline times for only 5 of its 31 tasks of eight hours or more.

EstimateEmerging evidenceNot applicable / English

Source (as cited): METR, “Time Horizon 1.1,” METR blog post, 29 January 2026
Type of source: Independent research report (recorded as “Research nonprofit post”).; Read the original
Does not license: A claim about agency in deployment

What would change this claim, and its limitations

An independent replication of the trend on a different task suite

Limitations: A self-report by the organization that owns the measure; web page as served on 16 October 2026; Anthropic models, made by the developer of the model that drafted this debate, are among those measured

Source and status

Publication status: Eligible for publication, with qualifiers: source position
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C24 (v1.0)

In a US online sample of 300 adults surveyed in July 2023, 67% attributed some possibility of conscious experience to ChatGPT; people who had used it gave higher ratings (mean 29.59 on a 1–100 scale) than those who had not (19.37).

EstimateEmerging evidenceUS / English

Source (as cited): Colombatto & Fleming, “Folk psychological attributions of consciousness to large language models,” Neuroscience of Consciousness, 2024
Type of source: Peer-reviewed journal article (recorded as “Peer-reviewed journal article”).; Read the original
Does not license: A finding about what the system is, or a causal effect of use on attribution

What would change this claim, and its limitations

Surveys in other countries or later years with different attributions

Limitations: One country, one time point, one system; the explanation given to participants may have shaped answers; the authors note the link between use and attribution could run either way; the survey did not ask whether people see the system as a tool or an agent

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C25 (v1.0)

Comparing 15,881 human–ChatGPT dialogues with 10,784 human–human dialogues, Mitiaeva and Xiao report “a consistent dissociation: AI produces the surface features of cooperative communication without the underlying social architecture”; for example, hedging and softening that go with more accommodation between people go with less alignment when the AI produces them.

Estimate and InterpretationEmerging evidenceNot applicable / English

Source (as cited): Mitiaeva & Xiao, “Talking past the machine: morality, politeness, and alignment in human–AI dialogue,” arXiv:2609.21401, 2026
Type of source: Preprint (recorded as “Preprint”).; Read the original
Does not license: A clean comparison; the two corpora differ in topic, roles and genre as well as in partner

What would change this claim, and its limitations

The same dissociation absent in a comparison matched on topic and roles

Limitations: Read in arXiv version 1 (18 September 2026); the PDF states acceptance at HICSS-60; the proceedings version was not compared; GPT models only; the authors say their measure cannot separate voluntary from structurally imposed matching; one author works for a company that sells AI products

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C26 (v1.0)

In a Stanford audit of 844 tasks in 104 US occupations, workers expressed positive attitudes toward automation by AI agents for 46.1% of their tasks, most often to free time for high-value work, but “generally prefer higher levels of human agency” than experts deemed “technologically necessary.”

EstimateEmerging evidenceUS / English

Source (as cited): Shao, Zope, Jiang, Pei, Nguyen, Brynjolfsson & Yang, “Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce,” arXiv:2506.06576, 2026
Type of source: Preprint (recorded as “Preprint”).; Read the original
Does not license: Evidence of what people do with AI at work

What would change this claim, and its limitations

Data on actual use showing workers delegating more than they say they want

Limitations: Read in arXiv version 3 (1 February 2026); measures stated preferences, not use

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C27 (v1.0)

Bommasani and colleagues named “foundation models” for their “critically central yet incomplete character,” argued that their significance lies in “emergence and homogenization,” warned that homogenization “creates single points of failure,” and predicted that the barrier to entry for developing them “will continue to rise.”

InterpretationEmerging evidenceNot applicable / English

Source (as cited): Bommasani et al., “On the Opportunities and Risks of Foundation Models,” arXiv:2108.07258, 2021
Type of source: Preprint (recorded as “Preprint (research centre report)”).; Read the original
Does not license: A measurement of homogenization or concentration; none is given

What would change this claim, and its limitations

Many independent base models with uncorrelated failures, or falling barriers to entry

Limitations: Read in arXiv version 3 (12 July 2022); written before chat products and open-weight releases at scale; the term itself carries the infrastructure reading it describes; the authors’ industry ties were not audited

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C28 (v1.0)

Van der Vlist, Helmond and Ferrari describe “Big AI” as the structural convergence of AI and Big Tech, arguing that AI is “an actual technology stack encompassing infrastructure, models, applications” and that Amazon, Microsoft and Google have integrated all three into their cloud offerings.

InterpretationEmerging evidenceUS / global cloud distribution / English

Source (as cited): van der Vlist, Helmond & Ferrari, “Big AI: Cloud infrastructure dependence and the industrialisation of artificial intelligence,” Big Data & Society, 2024
Type of source: Peer-reviewed journal article (recorded as “Peer-reviewed journal article”).; Read the original
Does not license: Figures on dependence; the abstract gives none

What would change this claim, and its limitations

Leading model developers running without the three firms’ cloud infrastructure

Limitations: Only the publisher’s abstract was read; the full text could not be reached; the study covers three US firms

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the abstract of the primary source only. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C29 (v1.0)

Luitse and Denkena examine large language models as “infrastructural components for AI research and development,” arguing that they are intertwined with big technology companies’ business models and, using GPT-3, describing “corporate efforts to commodify” them “through paid API access and exclusive licensing.”

InterpretationEmerging evidenceNot applicable / English

Source (as cited): Luitse & Denkena, “The great Transformer: Examining the role of large language models in the political economy of AI,” Big Data & Society, 2021
Type of source: Peer-reviewed journal article (recorded as “Peer-reviewed journal article”).; Read the original
Does not license: A measurement of dependency; the abstract gives none

What would change this claim, and its limitations

Wide access to frontier-scale models without paid or licensed access

Limitations: Only the publisher’s abstract was read; written before ChatGPT and before open-weight models at scale

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the abstract of the primary source only. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C30 (v1.0)

Krokowski and Hirsch-Kreinsen treat AI as “a paradigmatic case of a promising technology” whose future-oriented narratives mobilize resources while application problems and functional limits are “frequently overlooked,” and suggest that its future “could be characterized more by” a path of “routinization” than by the vision of artificial general intelligence.

InterpretationEmerging evidenceNot applicable / English

Source (as cited): Krokowski & Hirsch-Kreinsen, “Once promised, forever believed? An essay on the blind spots of AI promises,” AI & Society, 2025
Type of source: Peer-reviewed journal article (recorded as “Peer-reviewed journal article (essay)”).; Read the original
Does not license: A measurement of promises, resources or routinization

What would change this claim, and its limitations

Evidence that promises track realized capability closely

Limitations: Only the publisher’s abstract was read; an essay, not an empirical study

Source and status

Publication status: Eligible for publication, with a qualifier
How far it was checked: Checked against the abstract of the primary source only. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C31 (v1.0)

Comparing scientists’ citation networks with market data for the dot-com era (1994–2001) and the AI era (2017–2024), Chelikavada and Bennett found that dot-com patterns “did not definitively predict the rise and fall of an AI bubble,” and that “the data seems to suggest two possibilities”: an unprecedented kind of bubble or no bubble.

InterpretationEmerging evidenceNot applicable / English

Source (as cited): Chelikavada & Bennett, “Examining the Relationship between Scientific Publishing Activity and Hype-Driven Financial Bubbles: A Comparison of the Dot-Com and AI Eras,” arXiv:2509.11982v2, 2025
Type of source: Preprint (recorded as “Preprint”).; Read the original
Does not license: A finding on whether AI valuations are a bubble

What would change this claim, and its limitations

A market correction that the method did or did not anticipate

Limitations: Only the arXiv abstract of version 2 was read; not peer-reviewed; data end in 2024

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the abstract of the primary source only. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C32 (v1.0)

The Bank for International Settlements wrote in June 2024 that adoption of general-purpose technologies “typically follows a J-curve: it is slow at first, but eventually accelerates,” that AI shows features suggesting “an even steeper J-curve,” and that if AI is a true general-purpose technology that raises “total factor productivity in all industries to a similar extent,” “the demand for labour is set to increase across the board.”

InterpretationEmerging evidenceGlobal / English

Source (as cited): Bank for International Settlements, “Artificial intelligence and the economy: implications for central banks,” Annual Economic Report 2024, chapter III
Type of source: Policy research or monitoring report (recorded as “International financial institution report”).; Read the original
Does not license: A measurement of productivity effects

What would change this claim, and its limitations

Productivity data showing no acceleration over a long period, or effects confined to a few sectors

Limitations: Mid-2024 vintage; the general-purpose-technology status is stated as conditional; adoption figures cited concern use, not productivity

Source and status

Publication status: Eligible for publication, with qualifiers: institutional position
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C33 (v1.0)

The BIS found in December 2025 that the “Magnificent 7” technology stocks had grown to nearly 35% of S&P 500 market capitalization, from about 20% in November 2022, and called the recent rally in their prices “dissimilar to the dotcom bubble” because it rested on realized earnings growth as well as optimistic expectations, while noting price–earnings multiples approaching the top 10% of their history and concerns about stretched valuations.

Estimate and InterpretationEmerging evidenceUS / English

Source (as cited): Bank for International Settlements, BIS Quarterly Review, December 2025, overview (“Volatility challenges risk-taking”)
Type of source: Policy research or monitoring report (recorded as “International financial institution report”).
Does not license: A judgment on AI’s productivity, or a forecast of a correction

What would change this claim, and its limitations

A fall in earnings that leaves valuations resting on expectations alone

Limitations: Market-pricing evidence, not a measure of AI revenue or productivity; compares only with the dot-com case

Source and status

Publication status: Eligible for publication, with qualifiers: institutional position
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Bank for International Settlements, BIS Quarterly Review, December 2025, overview (“Volatility challenges risk-taking”): no checked link to the original yet.

Claim LD01-SL-C34 (v1.0)

China’s State Council set targets in August 2025 that the penetration rate of new-generation smart terminals and “intelligent agents” (智能体) exceed 70% by 2027 and their adoption rate exceed 90% by 2030, and said that implementing “AI+” will bring “new infrastructure” and accelerate “new quality productive forces” (新质生产力) (CSET translation).

FactEmerging evidenceChina / English

Source (as cited): State Council of the People’s Republic of China, “Opinions on Deepening the Implementation of the ‘Artificial Intelligence+’ Initiative,” 26 August 2025 (English translation by CSET)
Type of source: Government document (recorded as “Government publication (in translation)”).
Does not license: Evidence of realized penetration, or of what the targets mean in measurement

What would change this claim, and its limitations

Measured penetration against the targets

Limitations: The Chinese original was read only in the CSET English translation (Document No. 11 [2025]); the original could not be reached; targets are aims, not outcomes; the text does not call AI itself “general infrastructure”

Source and status

Publication status: Eligible for publication, with qualifiers: legal scope
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
State Council of the People’s Republic of China, “Opinions on Deepening the Implementation of the ‘Artificial Intelligence+’ Initiative,” 26 August 2025 (English translation by CSET): no checked link to the original yet.

Claim LD01-SL-C35 (v1.0)

The European Commission’s guidelines note that the AI Act’s definition of a “general-purpose AI model” “does not set out specific criteria” and adopt an indicative criterion: training compute above 10²³ FLOP and the ability to generate language, images or video, while calling compute “an imperfect proxy for generality and capabilities.”

FactEmerging evidenceEU / English

Source (as cited): European Commission, “Guidelines on the scope of the obligations for providers of general-purpose AI models established by Regulation (EU) 2024/1689 (AI Act),” C(2025) 7719 final, 19 November 2025
Type of source: Government document (recorded as “Government publication”).
Does not license: A claim that models below the threshold are not general-purpose, or that the guidelines bind courts

What would change this claim, and its limitations

Revised criteria, or a court ruling on the definition

Limitations: The definition in Article 3(63) was read as quoted in the guidelines, not in the Official Journal; guidelines are not binding law; the Commission says the criterion may change as technology and the market evolve

Source and status

Publication status: Eligible for publication, with qualifiers: legal scope
How far it was checked: Checked against the primary source. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
European Commission, “Guidelines on the scope of the obligations for providers of general-purpose AI models established by Regulation (EU) 2024/1689 (AI Act),” C(2025) 7719 final, 19 November 2025: no checked link to the original yet.

Claim LD01-SL-C36 (v1.0)

Testing OpenAI’s o1, a model trained to reason step by step, the authors of “Embers of Autoregression” found that it substantially outperformed earlier models “in many cases” but “still displays the same qualitative trends”: its accuracy at decoding a shift cipher ranged from 47% in the lowest-probability case to 92% in the highest, and it used more reasoning tokens on low-probability examples.

EstimateEmerging evidenceEnglish task set

Source (as cited): McCoy, Yao, Friedman, Hardy & Griffiths, “When a language model is optimized for reasoning, does it still show embers of autoregression? An analysis of OpenAI o1,” arXiv:2410.01792, 2024
Type of source: Preprint (recorded as “Preprint”).; Read the original
Does not license: A claim about later reasoning models, or that o1 cannot do these tasks

What would change this claim, and its limitations

Reasoning models whose accuracy no longer depends on output probability

Limitations: Read in arXiv version 2 (4 October 2024); o1’s workings are not public; ceiling effects limit some comparisons, as the authors note; one author works for the developer of the model tested; Anthropic’s Claude 3, made by the developer of the model that drafted this debate, is among the models compared

Source and status

Publication status: Eligible for publication, with qualifiers: preprint
How far it was checked: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.

Claim LD01-SL-C37 (v1.0)

What moved this debate?

No Radar Update is recorded for this debate in this build.

Cite this page

SocioLens project. (2026, October 16). What kind of thing is generative AI, and does the answer predict anything? [Living Debate LD-01, version v0.1]. SocioLens. https://sociolens-web.pages.dev/debates/ld-01/v0.1/

@misc{sociolens_ld_01_v0_1,
  author = {{SocioLens project}},
  title = {What kind of thing is generative AI, and does the answer predict anything?},
  year = {2026},
  date = {2026-10-16},
  howpublished = {\url{https://sociolens-web.pages.dev/debates/ld-01/v0.1/}},
  note = {Living Debate LD-01, version v0.1}
}

The link points to this version, whose record never changes. For a finding, cite the original study as well: SocioLens summarizes research and does not replace it.

Full locked text, structure and provenance

Full locked text (verbatim)

LD-01 — WHAT KIND OF THING IS GENERATIVE AI, AND DOES THE ANSWER PREDICT ANYTHING? Living Debate v0.1 — PUBLISHED

STATUS First repository-authored version. Drafted by Claude (SocioLens Project Lead), an AI model made by Anthropic, a developer that sells the "agent" products this debate discusses; not reviewed by a human domain expert.

EVIDENCE CUT 16 October 2026

LAST UPDATED 16 October 2026

QUESTION

Is the best description: • a normal technology: a powerful general-purpose tool whose effects arrive slowly, at the pace of institutions; • a cultural and social technology: an engine that imitates and recombines what people have written, like writing, print or libraries; • an agent: a system that pursues goals and acts with growing independence; • artificial communication: a partner in communication that does not understand, so that what matters happens on the human side; • an infrastructure: a foundation layer and a cloud stack that a few firms control and many others build on; • a capital formation: an investment and expectations phenomenon whose valuations run ahead of what is known; • or several of these at once, each true of a different level?

SHORT ANSWER

• Only two accounts have stated predictions that have been checked or can be checked directly. The normal-technology account predicts slow diffusion in high-consequence tasks and no meaningful AI advantage over trained humans at forecasting geopolitical events or at persuading people to act against their own self-interest, a qualifier the authors call critical; a cultural-technology study predicted that models would imitate well and innovate poorly, and reported preliminary results in line with that. • Studies of what is inside models rule out only the strongest "mere surface statistics" claim, and only for the systems tested. Studies of failures show behaviour shaped by the training objective, which fits mimicry and a tool with known failure modes (a SocioLens gloss of the normal-technology view) equally. • The length of software tasks that AI agents complete has grown fast on one organization's measure, which is a precondition for the agent account, not a test of it. • Adoption is fast by headcount and slow by intensity, so both the normal-technology account and its rivals can claim the same survey; the study's revision changed which comparison leads. • One infrastructure source predicts shared failure and rising barriers to entry, which the sources read here do not measure; the capital-formation account describes asset markets, and the Bank for International Settlements found the recent rally of the largest technology stocks unlike the dot-com bubble because earnings have grown, while noting concerns about stretched valuations. • Law and policy have already chosen: the Commission's guidelines for the EU AI Act adopt training compute as an indicative criterion for a general-purpose AI model, which they call an imperfect proxy, and China's State Council sets targets for "intelligent agents." • No pre-registered comparison of all six accounts' predictions was found. A joint essay by the normal-technology and AI 2027 authors reportedly proposes ways to tell their two views apart; it could not be read, and the search for this version was web-based and partial.

The central question is therefore not:

"What is generative AI, really?"

It is:

WHICH DESCRIPTION OF GENERATIVE AI PREDICTS SOMETHING THE OTHERS DO NOT, AND AT WHICH LEVEL: THE MODEL, ITS USE, ITS OWNERSHIP OR ITS FINANCE?

EVIDENCE LANGUAGE

SocioLens distinguishes:

FACT — directly supported observation or study fact. ESTIMATE — measured quantity whose design, sample and period matter. INTERPRETATION — synthesis across evidence, or an argument. UNKNOWN — unresolved empirical question.

FOUR LEVELS OF DESCRIPTION

The accounts are easier to compare once it is clear which level each describes:

THE MODEL What a trained system is and does: imitation, representation, reasoning, failure.

USE How people and organizations take it up, and what role they give it: tool, agent, partner.

CONTROL Who builds, hosts and sells it, and who depends on whom.

FINANCE AND LAW How it is valued and invested in, and how law and policy classify it.

1. NORMAL TECHNOLOGY: PREDICTIONS STATED

INTERPRETATION

Narayanan and Kapoor describe AI as "normal technology," a description, a prediction and a prescription. They predict that slow diffusion will continue to be the norm in high-consequence tasks, because diffusion is limited by the speed at which people, organizations and institutions adapt (LD01-SL-C01). They also predict that AI will not meaningfully outperform trained human teams at forecasting geopolitical events or at persuading people against their own interest, and that a growing share of jobs will consist of controlling AI (LD01-SL-C02). They give no dates or probabilities and say it is not clear how much slack to allow, which makes the predictions hard to score as failed.

The general-purpose-technology reading sits close to this one. Eloundou and colleagues, three of them at OpenAI, conclude that language models show traits of general-purpose technologies: about 80% of US workers could have at least a tenth of their tasks affected, and they make no prediction of timing (LD01-SL-C14). The Bank for International Settlements expected a "steeper J-curve" of adoption, and rising demand for labour if AI raises productivity in all industries to a similar extent (LD01-SL-C33). SocioLens synthesis: a general-purpose technology with a slow J-curve and one with a steep J-curve are both "normal," so the label does not decide the pace; adoption and productivity data have to.

2. CULTURAL AND SOCIAL TECHNOLOGY: AN IMITATION ENGINE

ESTIMATE + INTERPRETATION

The cultural-technology account says these models give people access to what others have written and recombine it, as writing, print and libraries did, rather than acting as new minds (LD01-SL-C04). Its fullest statement, by Farrell, Gopnik, Shalizi and Evans in Science (2025), could not be read for this version. Its experimental arm could: Yiu, Kosoy and Gopnik predicted that models would be almost as capable as people at linking familiar objects and fail at finding a new use for an everyday object. Children chose the functionally right object 85.2% of the time and adults 95.7%; the 2023 models were "much less capable," and the paper gives no percentages for them (LD01-SL-C04).

The account overlaps with an older and sharper claim, that language models only mimic the statistics of text ("stochastic parrots"). Those papers could not be read either. Philosophers who treat mimicry as a null hypothesis find that models show at least local, if not broad, generalization but often struggle beyond the range of tasks in their training data (LD01-SL-C07). Researchers were split: in a 2022 survey cited by Mitchell and Krakauer, 51% of 480 natural-language-processing researchers agreed that a text-only model could understand language in some non-trivial sense and 49% disagreed (LD01-SL-C05).

A third reading sits between imitation and agency. Shanahan and colleagues, one of them at DeepMind, describe a dialogue agent built on a base model as a simulator that role-plays characters: an agent frozen before the 2022 World Cup that names France as champion behaves as a role-player would, and talk of deception or self-preservation is expected where the training data contain it (LD01-SL-C06).

3. WHAT IS INSIDE THE MODEL

ESTIMATE

The mimicry claim predicts that models hold only correlations; its rivals predict structured representations. A model trained only on Othello moves encoded the board, and editing that internal board changed its moves as the edited board implied (LD01-SL-C16). Linear probes on Llama-2-70B recovered the locations of world places (R² 0.911) and the dates of historical figures (0.835), but the authors say these are not "a dynamic causal world model" in their own right (LD01-SL-C17).

Failures point the other way. GPT-4 decoded a cipher 51% of the time when the answer was a probable word sequence and 13% when it was not (LD01-SL-C18); OpenAI's o1, trained to reason step by step, did much better but showed the same pattern, from 47% in the least probable case to 92% in the most (LD01-SL-C37). GPT-4 named a celebrity's parent 79% of the time but the celebrity from the parent 33% of the time (LD01-SL-C19). An irrelevant clause cut maths accuracy by up to 65 percentage points; GPT-4o scored 95.2% on the full test set and 63.1% with the clause (LD01-SL-C20). ChatGPT and GPT-4 multiplied 3-digit numbers correctly 55% and 59% of the time, falling towards zero as problems grew (LD01-SL-C21). On tasks with changed premises, such as arithmetic in base 9, performance stayed above chance on most tasks but fell consistently (LD01-SL-C22).

Most of these studies tested 2023–2024 models, and several authors say their reading is a hypothesis; a 2026 re-analysis disputes the statistics behind part of the maths study and leaves the irrelevant-clause result out of its scope. SocioLens synthesis: they rule out the strongest "pure surface statistics" claim for the systems studied, and they show uneven competence shaped by the training objective. That fits the mimicry and cultural-technology accounts, and equally a normal technology with known failure modes (a SocioLens gloss). None of it decides whether the system is a tool, an agent or a partner in communication.

4. THE AGENT ACCOUNT: DEFINITIONS AND A PRECONDITION

FACT + ESTIMATE

The agent literature mostly offers definitions. Chan and colleagues treat agency as graded, raised by underspecification, directness of impact, goal-directedness and long-term planning, and say that recognizing it does not shift human responsibility (LD01-SL-C10). Kasirzadeh and Gabriel, the latter at Google DeepMind, profile systems on autonomy, efficacy, goal complexity and generality (LD01-SL-C12). Anthropic, the developer of the model that drafted this debate, defines agents by control flow: "typically just LLMs using tools based on environmental feedback in a loop" (LD01-SL-C13). SocioLens reading: under that definition the word marks an architecture, which a normal-technology account can accept.

The measurable part is a precondition. METR estimates that the length of software tasks AI agents complete with 50% success doubled about every 207 days from 2019 to 2025, with much lower performance on "messier" tasks (LD01-SL-C23). When it revised its task suite in January 2026, the post-2023 doubling time moved from 165 to 131 days, and it had measured human times for only 5 of its 31 longest tasks (LD01-SL-C24). Longer autonomous tasks are what the agent account needs; they are not evidence of goals, initiative or a role as an agent in organizations.

Use is a separate question from capability. US workers were positive about AI agents automating 46.1% of their tasks, most often to free time for high-value work, and generally preferred more human agency than experts deemed technologically necessary (LD01-SL-C27). These are stated preferences, not observed use.

5. ARTIFICIAL COMMUNICATION: OBSERVABLES ON THE HUMAN SIDE

INTERPRETATION + ESTIMATE

Esposito proposes thinking of algorithms "not in terms of artificial intelligence but in terms of artificial communication": if machines contribute to social intelligence, it is because people have learned to communicate with them (LD01-SL-C08). Only the publisher's abstract of her book could be read, and it was written in 2022 about machine learning in general. A 2026 reading of this tradition argues that what models lack is a position in practices of commitment and assessment, so that "no behavioral evidence can satisfy a positional criterion," by design (LD01-SL-C09).

The Field Map asks every tradition to say what one would observe differently if it were true. SocioLens reading, not a claim of either source: on this account the observables sit on the human side, in who is held accountable for outputs and how they gain standing through use, rather than in benchmark scores.

Two studies look at that side. In a US sample surveyed in 2023, 67% attributed some possibility of conscious experience to ChatGPT, and users rated it higher than non-users (LD01-SL-C25). A comparison of human–ChatGPT and human–human dialogues reports that the AI "produces the surface features of cooperative communication without the underlying social architecture" (LD01-SL-C26); the two sets of dialogues differ in topic and roles as well as in partner, so the comparison is weak.

6. HOW FAST IT SPREADS

ESTIMATE + INTERPRETATION

The same survey serves both sides. Nearly 40% of US adults aged 18–64 used generative AI in late 2024, and 27% of workers used it for work two years after ChatGPT's release against 25% using computers at work in 1984; but only 1%–5% of work hours were assisted (LD01-SL-C15). Both versions make both comparisons: overall adoption against PCs (39.5% against 20% in September 2024, 39% in February 2025) and work adoption against PCs at work, three years after the IBM PC (28% against 25%, then 27%); the February 2025 revision leads with the work comparison, and the launch dates chosen drive the result. Narayanan and Kapoor, reading the earlier version, argue that once intensity is counted generative AI may have spread more slowly than the PC (LD01-SL-C03).

SocioLens synthesis: fast uptake by headcount and slow integration by hours fit the normal-technology account and a general-purpose technology on a J-curve alike. What would separate them is a prediction about firms and tasks over the coming years, stated in advance.

7. INFRASTRUCTURE: A FOUNDATION LAYER AND A CLOUD STACK

INTERPRETATION

Bommasani and colleagues named "foundation models" for their "critically central yet incomplete character": homogenization "creates single points of failure," flaws are inherited by every adapted model, and the barrier to entry "will continue to rise" (LD01-SL-C28). Van der Vlist, Helmond and Ferrari describe AI as "an actual technology stack" that Amazon, Microsoft and Google have built into their cloud offerings (LD01-SL-C29). Luitse and Denkena treat language models as infrastructural components that firms also sell through paid API access and exclusive licensing (LD01-SL-C30). SocioLens reading: this makes infrastructure and product two sides of one arrangement rather than rivals.

Of these sources, only Bommasani and colleagues state predictions, of shared failure and rising barriers to entry; the other two, read only in their abstracts, raise questions of concentration and dependence. None of them measures these effects. Concentration and sovereignty are the subject of the LD-11 debate.

8. CAPITAL FORMATION: VALUATIONS AND EXPECTATIONS

ESTIMATE + INTERPRETATION

Krokowski and Hirsch-Kreinsen treat AI as "a paradigmatic case of a promising technology" whose narratives mobilize resources, and suggest that AI's future "could be characterized more by" "routinization" than by artificial general intelligence (LD01-SL-C31), a view close to the normal-technology account. A comparison of citation networks and market data found that dot-com patterns "did not definitively predict" an AI bubble, and that the data "seems to suggest" either an unprecedented bubble or none (LD01-SL-C32). In December 2025 the BIS found the "Magnificent 7" technology stocks at nearly 35% of S&P 500 value, up from about 20% in November 2022, with price–earnings multiples near the top tenth of their history; it called their recent rally "dissimilar to the dotcom bubble" because earnings had grown, and noted concerns about stretched valuations and the profitability of data-centre investment (LD01-SL-C34).

SocioLens synthesis: the capital-formation account describes how AI is financed and valued, not what the technology does; it can be true together with any of the other accounts. Central-bank and IMF warnings on valuations, firms' capital spending and investors' revenue-gap estimates were found only in press reports and carry no claim.

9. LAW AND POLICY HAVE ALREADY CHOSEN

FACT

Institutions classify generative AI whatever researchers decide. The European Commission's guidelines note that the AI Act's definition of a general-purpose AI model sets no specific criteria, and adopt an indicative one: training compute above 10²³ FLOP plus the ability to generate language, images or video, while calling compute "an imperfect proxy for generality and capabilities" (LD01-SL-C36). The model, a base for downstream systems, is classified apart from the systems built on it, close to the foundation-model idea. The Restatement (Third) of Agency, as quoted by a legal scholar who calls this the apparent consensus and notes contrary views, treats a computer program as an instrumentality of its user, not an agent, "at present" (LD01-SL-C11). China's State Council set targets for smart terminals and "intelligent agents" to exceed 70% penetration by 2027 and 90% adoption by 2030 (in the CSET translation), and ties AI to "new infrastructure" and "new quality productive forces" (LD01-SL-C35).

SocioLens synthesis: these classifications are ontologies in practice. Each predicts what will be regulated, sold or counted, and those effects can be observed whatever the right description of the technology turns out to be.

10. THE SIX POSITIONS

P1 — NORMAL TECHNOLOGY Claim: Generative AI is a powerful general-purpose technology whose effects arrive slowly, at the pace of institutional adaptation, with people in control. Current status: The account with the most stated predictions. Slow integration by hours of work is consistent with it, and equally with a general-purpose technology on a J-curve; fast uptake by headcount is claimed against it. Its predictions carry no dates or thresholds, so none has yet been scored.

P2 — CULTURAL AND SOCIAL TECHNOLOGY Claim: Generative AI is an engine for imitating and recombining accumulated human knowledge, comparable to writing, print or libraries. Current status: One experiment reported preliminary results in line with its stated prediction on 2023 models. Studies of failure fit it, but fit other accounts too. Its main statement was not read for this version.

P3 — AGENT Claim: Generative AI systems pursue goals and act with growing independence, so they should be studied and governed as agents. Current status: Defined as a matter of degree or by control flow, not as a type. The length of tasks agents complete is growing fast on one evaluator's measure; there is no evidence here of goals or initiative, and the Restatement (Third) of Agency, as quoted by one legal scholar, treats programs as instrumentalities "at present."

P4 — ARTIFICIAL COMMUNICATION Claim: Generative AI takes part in communication without understanding; what matters is how people communicate with it. Current status: Explicitly rejects behavioural tests in one recent reading. Its observables lie on the human side; the two studies of that side are a single survey and a confounded dialogue comparison. Esposito's own text was not read beyond its abstract.

P5 — INFRASTRUCTURE Claim: Generative AI is a foundation layer and a cloud stack on which others build and depend. Current status: One source states predictions of shared failure, inherited flaws and rising barriers to entry; the sources read here do not measure them. Recognized in law by the EU's separation of models from the systems built on them.

P6 — CAPITAL FORMATION Claim: Generative AI is above all an investment and expectations phenomenon whose valuations run ahead of what is known. Current status: Describes finance, not the technology, and so can hold together with any other account. Market concentration and high valuations are documented; the BIS finds realized earnings growth that distinguishes the recent rally from the dot-com bubble, and notes concerns about stretched valuations.

11. WHY THE CAMPS TALK PAST EACH OTHER

INTERPRETATION

• The accounts describe different levels, the model, its use, its control and its finance, and most could be true at once. • "Agent" means a control-flow architecture to developers, a graded property to some researchers and a legal status to lawyers. • The same evidence is claimed by several sides: adoption surveys by normal-technology and fast-diffusion readings, failure studies by mimicry and by the view of AI as a tool with limits. • Some accounts decline to name observables, by design or by omission. • The developer of the model that drafted this debate sells agent products, and one of its pages is a source here.

12. WHAT WE STILL DO NOT KNOW

UNKNOWN 1. Which observations would one account predict and another forbid, stated before the data arrive? 2. Do the failure patterns of 2023–2024 models hold for later reasoning and agentic systems? 3. Do deployed agents pursue goals or show initiative in organizations, beyond completing longer tasks? 4. How fast is generative AI integrated into firms' processes and hours of work, rather than tried by individuals? 5. Are failures and biases correlated across products built on the same base models, as the infrastructure account predicts? 6. What does the artificial-communication account predict about accountability and uptake, and can it be measured? 7. How do people classify these systems, as tool, agent or partner, outside the US and over time? 8. Will the EU's compute criterion and China's targets shape what is built and counted? 9. Does training compute track the generality that the EU's criterion is meant to capture?

13. WHAT EVIDENCE WOULD CHANGE THE PICTURE?

• A pre-registered comparison of the accounts' predictions, with thresholds and dates; • the normal-technology predictions on geopolitical forecasting and on persuasion against self-interest scored against trained human teams; • studies of later models' failures on the same tasks; • evidence of goals or initiative in deployed agents, or its absence; • measures of integration by firms and tasks, not only by users; • measures of correlated failure across products built on shared base models; • cross-national studies of how people classify these systems; • measures of generality across models around the compute threshold; • the full texts of the cultural-technology, "stochastic parrots" and artificial-communication statements.

14. CURRENT SOCIOLENS POSITION

INTERPRETATION

The best-supported description of the evidence to October 2026 is:

GENERATIVE AI IS SEVERAL KINDS OF THING AT ONCE, DEPENDING ON THE LEVEL DESCRIBED. THE ACCOUNTS RARELY COMPETE ON THE SAME OBSERVATIONS, AND THE EVIDENCE GATHERED SO FAR MOSTLY FITS MORE THAN ONE OF THEM. THE NORMAL-TECHNOLOGY ACCOUNT STATES THE MOST PREDICTIONS, THOUGH WITHOUT THRESHOLDS OR DATES; NO PRE-REGISTERED TEST OF WHICH ACCOUNT PREDICTS SOMETHING THE OTHERS FORBID WAS FOUND, THOUGH THE NORMAL-TECHNOLOGY AND AI 2027 AUTHORS HAVE BEGUN TO LOOK FOR DISTINGUISHING MEASURES.

GEOGRAPHIC AND EVIDENCE SCOPE

The evidence is mostly from the United States and in English, with policy documents from the EU and China (the latter read in an English translation). Most model studies tested systems from 2023–2024. Several sources are by authors at developers: Anthropic (C13, a vendor page; one author of C23), DeepMind and Google DeepMind (C06, C12), OpenAI (C14; one author of C37), Apple (C20) and Amazon (C26). Anthropic models were among those tested, measured or profiled in C04, C12, C22, C23, C24 and C37. This debate was drafted by an Anthropic model, and the agent account it weighs describes products Anthropic sells.

PROVENANCE

Research build: Source notes prepared 16 October 2026 by Claude with AI research agents, then checked against full texts, official abstracts and official web pages. Claims: Each claim rests on a source checked in full text, on an official page or against its official abstract; the claim's verification state says which. Five claims rest on the publisher's or arXiv abstract only (C08, C29, C30, C31, C32). Conference and journal papers read in an arXiv version name that version in their limitations; preprints are marked as such. Web pages were read as served on 16 October 2026. The State Council document was read in the English translation by Georgetown University's Center for Security and Emerging Technology. Not claims: Sources that could not be read carry no claim and are named only as positions: Farrell, Gopnik, Shalizi and Evans (Science, 2025); Bender, Gebru, McMillan-Major and Shmitchell (2021) and Bender and Koller (2020); the text of Esposito's book; the AI 2027 scenario and a joint essay by its authors and the normal-technology authors. Figures read only in search snippets or press reports carry no claim, including firms' capital spending, investors' revenue-gap estimates, Bank of England and IMF statements on valuations, chatbot user counts and revenue, Pew and Census Bureau adoption surveys, a Danish study of earnings effects, and liability cases. Productivity and employment effects are the subject of the LD-02 and LD-03 debates, evaluation of AI capabilities of LD-06, and concentration of LD-11. Review: None yet. Version: Living Debate v0.1 — PUBLISHED 16 October 2026

Research-of-record provenance

Materialized by sociolens-materializer 0.6.0 at 2026-09-28T20:00:00.000Z; normalization contract SocioLens — Claim Value Vocabulary v1.0.2 — Pre-Spike Materialization Contract (v1.0.2). Content seal SHA-256 b6524a56e47abf59fe750472b735562afc4e1a676792b97a0880c46d55d1d20f.

Repository-authored inputs (not Drive exports)

LD-01 — WHAT KIND OF THING IS GENERATIVE AI, AND DOES THE ANSWER PREDICT ANYTHING? — Living Debate v0.1 (v0.1) — written in the SocioLens repository by Project Lead (Claude, AI) on 2026-10-16; no Drive identity exists. Authority: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's choice in the SocioLens project chat on 16 October 2026 to build the LD-01 debate; a new Living Debate, first version, written from the project's LD-01 source notes, checked against full texts, official web pages or official abstracts. Block “LD-01 — WHAT KIND OF THING IS GENERATIVE AI, AND DOES THE ANSWER PREDICT ANYTHING? — Living Debate v0.1 EXPORT” SHA-256 88774563d6ca26c59e25537bbdb3bd8ef539091b6c970ce745d2345a7f96020d; fixture fixtures/authored/debates/LD-01--v0.1.

Builder decisions (not Drive data)

VERSION = v0.1 — basis: Debate version taken from the repository-authored Living Debate artifact version (LD-01 — Living Debate v0.1); the first version of a new debate
CURRENT_POINTER = true — basis: repository-authored Living Debate "LD-01 — WHAT KIND OF THING IS GENERATIVE AI, AND DOES THE ANSWER PREDICT ANYTHING? — Living Debate v0.1" (Project Lead (Claude, AI), 2026-10-16); basis: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's choice in the SocioLens project chat on 16 October 2026 to build the LD-01 debate; a new Living Debate, first version, written from the project's LD-01 source notes, checked against full texts, official web pages or official abstracts
VISIBILITY = PUBLIC — basis: The Owner's choice in the SocioLens project chat on 16 October 2026 to build the LD-01 debate for publication; every cited ClaimVersion rests on a source checked in full text, on an official web page or against its official abstract.
DEBATE_CITES_CLAIM = LD01-SL-C01--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C02--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C03--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C04--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C05--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C06--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C07--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C08--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C09--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C10--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C11--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C12--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C13--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C14--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C15--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C16--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C17--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C18--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C19--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C20--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C21--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C22--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C23--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C24--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C25--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C26--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C27--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C28--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C29--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C30--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C31--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C32--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C33--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C34--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C35--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C36--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version
DEBATE_CITES_CLAIM = LD01-SL-C37--v1.0 — basis: the 37 Claims written for this debate (fixtures/authored/claims/); the debate cites each first version