Research opportunity · What kind of thing is generative AI, and does the answer predict anything?
Which observations does each account of generative AI forbid?
In short: has this been done?
Related work: No study found. A first-pass search found no study that does the core design. It has not been checked in depth, so a fuller search could still find one.
How far checked: First-pass screen only. A first-pass AI web search on 16 October 2026. No deeper check yet. Every search and check so far was done by AI; no person or domain expert has checked them yet.
The searches, the nearest work found and what the search could not reach
Follow this opportunity: Atom feed of RO-LD01-01 updates (new versions, and revisions of the claims it rests on).
Question
If proponents of the normal-technology, cultural-technology, agent, artificial-communication, infrastructure and capital-formation accounts each state in advance, with thresholds and dates, observations their account predicts and observations it rules out, which predictions differ between accounts, and which hold over the following years?
Gap type: Wrong or missing validation targetFunded team projectLow to medium costRelated work: No study found
Why it matters
The debate finds that most evidence fits more than one account and that few accounts state what they predict. The normal-technology essay gives predictions without thresholds or dates, and one reading of the artificial-communication tradition rejects behavioural tests by design. Until rival accounts commit to observations that would count against them, the question "what kind of thing is it?" cannot be settled by evidence, only argued.
The debate lists this among what we still do not know: Which observations would one account predict and another forbid, stated before the data arrive?
Evidence that motivates it
- 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.”
How this study uses it: The normal-technology account states predictions about slow diffusion, without thresholds or dates. - 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.”
How this study uses it: Its predictions on forecasting and persuasion are specific but, by the authors' account, hard to make precise. - 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.”
How this study uses it: The cultural-technology experiment shows the form such a test can take, a prediction stated and then checked against preliminary results. - 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.”
How this study uses it: One reading of the artificial-communication tradition holds that no behavioural evidence can satisfy its criterion.
A possible design
Run an adversarial collaboration. Recruit authors or close readers of each account; for each, elicit five to ten observations the account predicts and five it rules out, each with a measurable threshold, a data source and a resolution date between 2027 and 2030. Code the overlap between accounts to find discriminating predictions, pre-register the set, and score it as data arrive. Treat accounts that decline to name observations as a finding and record their reasons.
Data needed
Structured elicitation with proponents; public data sources named for each prediction (adoption surveys, evaluation results, market data, court records); a pre-registration platform.
Feasibility
Funded team project. Low to medium; mostly researcher time for elicitation and later scoring. Proponents may decline or hedge; the design should record refusals rather than fill them in. Resolution takes years, but the elicitation and the overlap map are publishable on their own.
Proposal pack: this page and its references on one printable page, with BibTeX.
Has this been done?
The search found no study that does the core design (first-pass screen of 16 October 2026; not yet checked in depth).
No pre-registered set of discriminating predictions across accounts of what generative AI is was found. The nearest work is an adversarial collaboration on existential risk from AI, which sought cruxes between forecasters, and a joint essay by the normal-technology and AI 2027 authors, which could not be read and is reported to propose measures that would distinguish those two views. What stays open is a pre-registered comparison across all six accounts; the search was web-based and partial.
How strong this evidence is
- Who searched and checked: no checker is recorded beyond the first-pass screen, which was part of the AI drafting (Claude (SocioLens Project Lead), AI-drafted from the LD-01 debate). Every check so far was done by AI; no person or domain expert has checked it yet.
- What was read: 2 of the 2 nearest works below were judged from an abstract, summary or search listing only; each is marked. Not every entry says how much of the work was read.
- Where the search could not reach: Screened by web searches and the LD-01 source scan on the date shown; the joint essay and forecasting-institute reports were not read. Absence of a hit is not proof that the work does not exist.
- Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.
Nearest work found
- Forecasting Research Institute, "Roots of Disagreement on AI Risk: Exploring the Potential and Pitfalls of Adversarial Collaboration" (2024) Judged from an abstract, summary or search listing only
Seen in search results only. Elicited cruxes between forecasters who disagreed about catastrophic risk by 2100; not about what kind of thing generative AI is. - Kapoor, Narayanan, Kokotajlo and others, joint essay comparing "AI as normal technology" and "AI 2027", Asterisk Magazine (November 2025) Judged from an abstract, summary or search listing only
Seen in search results only; the essay was not read. Reported to list agreements and disagreements between two of the accounts.
Search queries used
adversarial collaboration pre-registered predictions normal technology versus AGI forecasts scoring indicatorsCommon Ground between AI 2027 and AI as Normal TechnologyarXiv and Crossref searches for comparisons of AI ontologies by testable predictions (LD-01 source scan, 16 October 2026)
Cite this page
SocioLens project. (2026, October 16). Which observations does each account of generative AI forbid? [Research opportunity RO-LD01-01, version v1]. SocioLens. https://sociolens-web.pages.dev/opportunities/ro-ld01-01/v1/
@misc{sociolens_ro_ld01_01_v1,
author = {{SocioLens project}},
title = {Which observations does each account of generative AI forbid?},
year = {2026},
date = {2026-10-16},
howpublished = {\url{https://sociolens-web.pages.dev/opportunities/ro-ld01-01/v1/}},
note = {Research opportunity RO-LD01-01, version v1}
}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.
Provenance
RO-LD01-01 v1. Drafted by Claude (SocioLens Project Lead), AI-drafted from the LD-01 debate on 2026-10-16. Not yet reviewed. Not yet reviewed by a domain expert. This is a research hypothesis, not a finding.