Reactions from ranked influencers
14 posts
But the reality is that this is probably the CS conference that still has had the most material impact on what policymakers think & understand about what to do about AI. That's just the historical reality.
Saying that FAccT had "little to no policy impact" is just plainly incorrect - the actions taken by the community led to successful FTC investigations, the AI Bill of Rights, the Biden EO, etc. DHS, DoJ, NTIA, GAO, etc. regularly cite the papers; NIST/CAISI pay attention as well!
@RishiBommasani I agree that many papers don't have policy impact but it's a remarkable feat of a conference of this size and (recent) history. Much larger conferences have attempted to have that kind of policy impact (eg CHI, Neurips/ICML via the "position papers" and they just haven't yet).
Understand the critique but I actually believe this is largely revisionist history. OSTP under Nelson & FTC under Khan seconded and hired *several* senior advisors directly from the @FAccTConference exec committee; its still the CS conference attended by most policymakers I know
I agree on the ignorance of many in the frontier AI governance space about this body of work. (Eg a simple heuristic is papers that talk about "auditing" that cite no papers of Deb Raji are doing something wrong, but I see this often.) With that said, I see it as a stronger indictment of a lot of the research from FAccT in 2017 - 2026. For a conference and community premised on societal impact more so than any other academic CS venue, my blunt take is most of the research has failed to actually take the next step and affect policy/public discourse/other forms of consequential decision-making. There are some counterexamples, but if one sees the FAccT community as having emerged out of the impact of work like Gender Shades or the like in the mid-to-late 2010s, in my view the 10 year retrospective is the research from this community has largely failed. It has failed to figure out how to have the broader impact it often calls for, and notably falls considerably short of having impact on the scale of Gender Shades.
@RishiBommasani If you look at the NIST RMF (https://airc.nist.gov/airmf-resources/playbook/), GAO's AI accountability framework (https://www.gao.gov/products/gao-21-519sp), NTIA accountability report (https://www.ntia.gov/issues/artificial-intelligence/ai-accountability-policy-report), FTC memos, DoJ settlements, etc. FAccT papers are often cited more than papers from any other conf/journal.
@RishiBommasani Fwiw, I'm not here to accuse you of anything! You are far from the only person raising several valid critiques! But the reality is that the conference and it's impact remains strangely under-estimated. There's a lot of tangible evidence to what the community has already achieved.