facct-topic-selection
GitHub用于判断负责任AI项目应投稿至ACM FAccT还是其他ML、HCI、法律或伦理会议。核心依据是公平性、问责制或透明度是否为核心贡献，且社会技术框架是否原生融合，而非事后附加。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill facct-topic-selection -g -y
SKILL.md
Frontmatter
{
"name": "facct-topic-selection",
"description": "Use when deciding whether a responsible-AI project belongs at ACM FAccT or should route to a pure-ML venue (NeurIPS\/ICML\/ICLR), an HCI venue (CHI\/CSCW), a law\/policy venue, or an AI-ethics venue (AIES), by testing whether fairness, accountability, or transparency is a first-class contribution and whether the interdisciplinary framing is native rather than bolted on."
}
FAccT Topic Selection
Decide the venue before drafting. ACM FAccT — the Conference on Fairness, Accountability, and Transparency — is the flagship interdisciplinary responsible-AI venue. Its reviewer pool spans computer science, law, the social sciences, the humanities, and policy, and its defining demand is that fairness, accountability, or transparency (FAccT) is a first-class contribution, not a fairness paragraph appended to a systems result. A technically strong paper whose real center is a new model, a new interaction technique, or a doctrinal legal argument — with FAccT concerns merely gestured at — is respected and then rejected as out of scope.
The routing question that matters most
The decisive question is rarely "does this touch fairness/AI?" but "is a fairness, accountability, or transparency question the actual contribution, and is the sociotechnical framing native?" FAccT uniquely rewards work that takes the social and the technical as inseparable. A paper that would lose nothing if you deleted the equity framing belongs elsewhere; a paper whose whole point is who is harmed, who is accountable, or what can be made legible belongs here.
Sibling-venue routing table
| Signal in your project | Better home | Why |
|---|---|---|
| Fairness/accountability/transparency is the contribution; social + technical are entangled | ACM FAccT | The interdisciplinary responsible-AI flagship; FAccT concerns are first-class |
| A new model/algorithm whose fairness angle is a secondary evaluation | NeurIPS / ICML / ICLR | ML flagships; a fairness metric alone does not make it FAccT |
| A new interaction technique or system; the study is about use more than justice/power | CHI / CSCW | HCI flagships; FAccT wants the accountability/harm question central |
| Primarily doctrinal legal analysis or regulatory design for a legal readership | Law reviews / policy venues | FAccT welcomes law, but wants cross-disciplinary reach, not doctrine alone |
| AI-ethics argument aimed at a philosophy/ethics readership | AIES and adjacent | Overlapping sibling; FAccT leans empirical + sociotechnical + policy-facing |
| Critical/qualitative/participatory engagement better as a session than a paper | FAccT CRAFT track | Participatory and world-building formats live in CRAFT, not the paper track |
Contribution shapes FAccT rewards
FAccT is genuinely pluralistic — the following are all native, and a good program mixes them:
- Algorithmic fairness / interpretability method — a new measure, algorithm, or auditing technique for bias, recourse, explainability, or transparency, evaluated on real data (the fairness-metrics lineage).
- Empirical audit — measuring disparate performance or harm in a deployed or commercial system across subgroups (the Gender Shades lineage).
- Documentation / accountability infrastructure — datasheets, model cards, data statements, audit frameworks, and impact-assessment tooling that change how the field builds and reports (the Model Cards / Datasheets lineage).
- Critical / conceptual / position work — an argument that reframes what the field takes for granted about harm, power, or measurement (the Stochastic Parrots lineage).
- Qualitative / sociotechnical study — interviews, ethnography, or a case study of how a system affects an affected community or institution, with sound method.
- Law & policy — legal, regulatory, or governance analysis that engages the technical substrate and reaches a mixed audience.
The two sharpening tests
- Delete-the-equity test: remove every sentence about fairness, accountability, transparency, harm, or power. If a complete, publishable contribution remains, the FAccT framing is decoration — route to the ML/HCI/legal home of the surviving core. If nothing coherent remains without it, FAccT is right.
- Mixed-reviewer test: imagine your paper read by a computer scientist, a lawyer, and a qualitative social scientist at once. FAccT-shaped work gives each of them something to hold and survives all three; a paper that only one of them can evaluate is usually a sibling-venue paper wearing a FAccT title.
Interdisciplinary rigor, not interdisciplinary gesture
Fit is necessary but not sufficient. FAccT reviewers penalize thin interdisciplinarity: a CS paper that cites one sociology book without method, or a critical paper that name-drops an algorithm it never engages. Whichever lane you sit in, meet that lane's standard of rigor — statistical care and honest baselines for a method/audit paper; coding schemes, saturation, and reflexivity for a qualitative paper; doctrinal precision for a legal paper — and then connect it across the divide.
Cheap reconnaissance before committing
[Scope] scan the last two FAccT programs (facctconference.org, ACM DL, dblp db/conf/fat) for
your topic -> several recent papers = a reviewer pool exists; none = opening or mismatch
[Focus areas] can you name a primary + secondary FAccT focus area (algorithm development; data &
algorithm evaluation; applications; human factors; privacy & security; law; policy;
critical/humanistic/social-scientific) that genuinely fit? -> if not, reconsider the venue
[Audience] would a lawyer AND a computer scientist both find a contribution? -> that dual pull is
the FAccT signature; a single-discipline pull points to a sibling venue
Decision procedure
[Who is affected] whose fairness/accountability/transparency changes if the claim holds?
[Contribution type] method / audit / documentation-infra / critical-conceptual / qualitative / law-policy
[First-class check] delete-the-equity test -> does a contribution survive without the FAccT framing?
[Interdisciplinary check] mixed-reviewer test -> do a CS + a law + a social-science reader each hold something?
[Format check] is it a paper, or a participatory session? -> paper track vs CRAFT
[Verdict] FAccT paper track / FAccT CRAFT / sibling venue (NeurIPS/ICML/CHI/AIES/law), one-line reason
Run this before the writing skills; a wrong venue decision wastes every later step. When the verdict
is FAccT, continue with facct-workflow for the calendar and facct-writing-style for the paper
shape.
Version History
- 9f86f09 Current 2026-07-19 15:44
