How to use SocioLens

SocioLens maps the evidence on contested questions about AI and society. It is built for researchers who need to know what is established, what is not, and where a useful study could go next.

Prefer to be shown around? The guided tour walks one research task through the site, step by step.

What is on the site

Debates (11): The main entry point. Each debate asks one question, gives the current answer, and says what is still unknown.
Claims (256): Single, sourced statements that a debate rests on. Each says what kind of statement it is, how mature the evidence is, and what it does not license.
Sources (293): The papers, reports and official documents behind the claims, with a link to the original where one has been checked.
Research opportunities (70): Research questions that follow from a debate's unknowns, each screened against the literature before publication.
Radar updates, Field Map and concepts: New evidence that moved a debate (listed on the home page and on the debate it moved), a map of the field's domains, and the theoretical concepts the debates use.

Reading a debate page

Read from the top down. Each part answers one question:

The question: what the debate is about.
Current answer: SocioLens's position, given the evidence at the stated cut-off date.
What we still do not know: the open gaps. This is where research opportunities come from.
What evidence would change the picture: the results that would move the answer.
Claims and evidence: the individual claims, each linked to its own page and sources.
What moved this debate?: radar updates since the last review.

The full original text of the debate, unchanged, sits in the collapsed section at the bottom. The version line at the top says which version you are reading and whether a newer one exists.

Paragraphs in the answer may carry a label:

FACT: Directly supported by a cited source.
ESTIMATE: A number or range inferred from sources, with uncertainty.
INTERPRETATION: SocioLens's reading of the evidence. Reasonable people may disagree.
UNKNOWN: Not established by any source found.

Reading a claim

A claim card shows the statement itself, then:

Claim type and maturity. Maturity runs from emerging (few studies) through rapidly maturing to mature (a settled body of work). Contested means credible studies disagree.
Does not license: conclusions the evidence does not support, even though they are often drawn from it. Read this before citing the claim.
What would change this claim: the finding that would overturn or narrow it.
Source and status: where the claim comes from and whether it was checked against the primary source.

Follow the claim link for the full record: every field, every source, and the claim's version history.

Reading a research opportunity

Each opportunity gives a question, why it matters, the claims that motivate it, a possible design, the data needed, a rough scale (from a master's thesis to a funded team project) and a cost band. The list can be filtered by size, cost and whether related work exists, and each claim page links to the opportunities it motivates.

The "Has this been done?" panel records the literature screen: when it was run, the search queries, the nearest existing work with links, and what remains open. A verdict of no study found means the screen found none, not that none exists. Its "How strong this evidence is" list says who searched and checked (so far only AI), how many of the nearest works were judged from an abstract or summary only, and where the search could not reach. Opportunities searched again in more depth after publication show that check's outcome, where it searched and its queries; those marked first-pass screen only have not had that check. Each page links its earlier versions, kept as they were published. Check the queries and run your own search before committing to a topic.

Opportunities are drafted with AI assistance and are research hypotheses, not findings. Each page says who drafted and checked it.

Three common tasks

Find out where a debate stands

Open a debate and read the question and current answer. Then read the unknowns. Five minutes gives you the state of the field and its limits.

Check a claim before you cite it

On the claim card, read "Does not license". Open the claim page and follow the source link to the original. If the source page has no link, its full reference is given so you can find it.

Find a research question

Start at Research opportunities (for example, Why do Nordic registers and US data disagree about young workers in AI-exposed jobs?). Pick one, read its motivating claims, then the "Has this been done?" panel. Use the queries as a starting point for your own search.

What you can and cannot rely on

Every claim traces to named sources, and claims are never silently rewritten. A change is a new version, and old versions stay readable.
Evidence has a cut-off date, shown on each debate. Work published later is not reflected until a review or radar update.
Gaps are reported as gaps. Where the evidence is missing, the site says so rather than filling it in.
SocioLens is a map, not a substitute for reading the sources. Cite the original work, not this site.