About SocioLens
SocioLens maps the evidence on the social science of AI. For each debate it shows what the evidence establishes, what it does not license, what is still unknown, and which research questions follow. It now holds 12 debates, 293 claims, 328 sources and 79 research opportunities.
Who publishes it
Publisher and editor: SocioLens project. SocioLens is a project run by its founder. It is a private preview: reachable by link, not announced, and not indexed by search engines.
Funding: not yet stated. A funding statement will be published before launch.
How the content is made
- Debates and claims. Drafted with AI models under the direction of the project owner. The debates of September 2026 were produced with several AI models; since 12 October 2026 the drafting model is Claude, made by Anthropic. Each claim names its sources, says what kind of statement it is (fact, estimate, interpretation or unknown), how mature the evidence is, what it does not license and what would change it.
- Checking. Each claim records how far it was checked against its source, and some sources were read from their abstract only; claim pages say so. Who checked each claim is not recorded claim by claim. Claim checks so far were AI-led: the project owner confirmed on 17 October 2026 that no person has checked claims against their sources. No independent human domain expert has reviewed it yet.
- Research opportunities. Drafted by AI from a debate's claims and unknowns, then screened for existing work by AI web search. Each is a hypothesis about a gap, not a finding, and each page says how far it was checked.
- What is never done. Model output is not presented as research evidence. Gaps in the evidence are reported as gaps, not filled with plausible guesses.
The box "About this page" at the top of each debate, claim and opportunity repeats this for that page.
Conflict of interest
The model that drafts most of SocioLens, Claude, is made by Anthropic, an AI company. Anthropic, its models or its documents appear in several debates: who can know about AI (LD-08) discusses Anthropic's data, funding and access programmes; the sources on how AI systems are evaluated (LD-06) evaluate Anthropic's models among others, though no Anthropic document carries a claim there; and concentration in the AI industry (LD-11) cites market-share estimates that include Anthropic. An AI company's model writing about AI companies is a conflict of interest. Read what these debates say about AI firms with that in mind, and check the cited sources, which every claim names.
Versions and corrections
Published claims and debates are never silently rewritten. A change is a new version that says what changed and why, and the old version stays available at its own address. To cite a page, use "Cite this page"; the citation points to the version you read.
Reporting an error: a contact address for corrections will be published before launch.
Limits
- The evidence base is narrow: 16 to 37 claims per debate. Coverage by period, region, language and type of publication has not been audited, so a debate can miss relevant work.
- Each debate covers evidence up to a cut-off date. Where a debate states its evidence cut, the page shows it; where it does not, the page says the cut is not stated. Work published after the cut is not covered until the debate is reviewed.
- SocioLens is a map of the research, not a substitute for reading it.
The founding documents
The Start Here statement and the Field Map were written while the project was planned. They are kept unchanged as part of the record, so some of their details describe plans rather than the site today.