Skills Soft Skills Positioning Interdisciplinary AI Research

Positioning Interdisciplinary AI Research

v20260724
facct-related-work
This guide provides best practices for authors writing the "Related Work" section for interdisciplinary AI conferences. It teaches how to precisely position novelty across multiple domains—including ML fairness, law, HCI, and critical theory—by writing delta-first contrasts and properly attributing borrowed concepts to their original academic source.
Get Skill
386 downloads
Overview

FAccT Related Work

Use this to audit novelty and disciplinary reach. FAccT reviewers come from different fields, and each expects to see the nearest work in their lane engaged. A fairness-metrics reviewer wants the ML fairness literature; a legal reviewer wants the relevant law and governance work; an STS/critical reviewer wants the theory you are (often implicitly) drawing on. The fastest way to lose a mixed panel is a bibliography that is deep in one field and blank in the others. Reopen the current CFP for anonymity, dual-submission, and prior-publication rules before advising.

Positioning checks

  • Name the FAccT novelty precisely. What is new: a fairness/transparency method, an empirical harm nobody had measured, an accountability framework, a reframing of a taken-for-granted construct, a qualitative account of an affected community, or a legal-technical synthesis?
  • Cover the disciplinary lanes (see table). A paper that cites only its home field reads as unaware of the interdisciplinary conversation FAccT exists to host.
  • Write delta-first. Each closely related work gets one sentence naming what it did and one naming what you do differently — across the divide where relevant ("the ML work optimized the metric; the legal work named the right; we connect them by...").
  • Cite borrowed constructs to their real origin. If you use "disparate impact," "contestability," "situated knowledge," or "the right to explanation," cite the field that coined it, not a second-hand ML paper — mixed reviewers notice mis-attribution instantly.
  • Preserve mutual anonymity. Cite your own prior work in the third person; never link reviewers to an identity-revealing preprint, repository, project page, or the arXiv version of this paper.
  • Declare overlap with a prior workshop/CRAFT version or concurrent submission; do not re-submit archival work as new.

FAccT literature lanes

Lane Typical venues / bodies What FAccT reviewers check
Algorithmic fairness & ML FAccT, NeurIPS/ICML/ICLR, JMLR Whether the nearest fairness measure/method is compared or distinguished
HCI & human factors CHI, CSCW Whether prior work on how people use/contest the system is credited
Law, policy & governance Law reviews, policy journals, regulation Whether the relevant legal doctrine or regulatory instrument is engaged correctly
STS & critical theory STS venues, critical data/algorithm studies Whether the theoretical lineage of your critique is named, not just gestured at
Documentation & accountability infra Prior FAccT (datasheets, model cards, audits) Whether existing documentation/audit frameworks are built on rather than reinvented
Domain literature (health, credit, hiring...) The applied field Whether you understand the real decision context you study

A bibliography that reaches across at least the lanes your claim touches signals command of the interdisciplinary field; one confined to a single lane invites the "unaware of the neighbor discipline" critique that a mixed panel is unusually well-positioned to make.

Delta-first positioning vignette

Suppose the paper proposes a contestability mechanism for automated benefit decisions. Its neighbors span lanes: an ML paper on algorithmic recourse (technique, no institutional grounding), an HCI study of how claimants experience appeals (experience, no mechanism), and legal scholarship on due-process rights in automated administration (the right, no system). The novelty sentence names all three contrasts — a mechanism where recourse gave only a technique, grounded in the appeal experience HCI documented, realizing the due-process right the law names — which is exactly the cross-lane synthesis FAccT rewards.

Concurrent and prior-version judgment calls

[Concurrent arXiv work]   cite neutrally, state the difference, avoid unverifiable priority claims;
                          keep the citation mutually anonymous
[Your workshop/CRAFT version]  usually non-archival and citable, but confirm against the current CFP
                          and phrase so anonymity survives
[Prior short/position version] declare the overlap and state what the full paper adds beyond it
[Archival status unclear]  declare the overlap in the submission form rather than guessing

Eligibility red flags

  • Substantial text overlap with a published paper by the same authors (self-plagiarism risk).
  • A "new" audit that re-reports a prior dataset's disparities without a new question or population.
  • Citations confined to one discipline while the paper claims interdisciplinary contribution — the clearest signal that the interdisciplinarity is a label, not a method.

Output format

[Eligibility] clear / needs declaration / risky
[Lanes covered] <ML-fairness / HCI / law-policy / STS-critical / documentation / domain>
[Nearest 3 works] <work -> one-line cross-lane delta>
[Construct attribution] <borrowed term -> cited to its real origin? yes/no>
[Archival-overlap risk] <none / declare: what>
[Novelty sentence] <FAccT-ready contribution contrast across the relevant lanes>
Info
Category Soft Skills
Name facct-related-work
Version v20260724
Size 5.56KB
Updated At 2026-07-28
Language