A look at the state of AI accountability research, which is focused not just on improving existing systems but also on whether such systems should ever be used
and, if so, who gets to govern them”, writes @FrankPasquale https://lpeblog.org/...
Law and Political EconomyFrank Pasquale
Context & Ripple Effects
When Frank Pasquale framed AI accountability research around two questions — improving systems, or whether they should run at all and who governs them — he was widening a debate that until then had centered on technical fixes like Explainable AI, the discipline aimed at making machine-learning decisions legible to humans.
That framing aged well: by 2021 researchers were warning that ethical review had quietly become the job of peer reviewers, and by 2024 US state bias bills were drawing fire for loopholes that let companies sidestep accountability — making the 'who decides' question Pasquale posed the operative fault line in AI policy.
First-order effects
Accountability research splits into two camps with different outputs: system-improvement work that vendors can adopt, and deployment-legitimacy work that questions whether a system should ship regardless of its accuracy.
Second-order effects
Legislators writing AI bias statutes inherit the governance question directly — and the loophole-riddled state bills show what happens when 'who gets to govern' is left to company-friendly drafting.
If the pattern holds, AI governance consolidates around who holds audit authority rather than who builds models, echoing the international contest over rule-setting already visible in nations and lobbyists vying over AI policy.
The trend: AI accountability is migrating from a technical transparency agenda toward a political fight over deployment legitimacy and regulatory authority.
Been thinking a lot on the current governance tensions we face, which are so well framed in this @FrankPasquale piece. Overriding thought I keep coming back to: the moderates/civility/enabler crew is poised to be dangerous as ever. No surprise but yeah. https://lpeblog.org/...
Must-read! Borrowing periodizing terminology from feminist theory, @FrankPasquale brilliantly distinguishes between waves 1 & 2 of algorithmic accountability. Example of wave 2 approach. Ban facial recognition. Don't aspire to make it more inclusive. https://lpeblog.org/...
This insightful essay by @FrankPasquale is in keeping with the spirit of my two forthcoming companion papers - we need to move beyond a surface preoccupation with fairness and ask about the fundamental assumptions of AI systems and who they serve. https://lpeblog.org/...
“Second wavers have helped slow down AI and robotics deployment enough so that first wavers have more time and space to deploy constructive reforms.” @FrankPasquale https://lpeblog.org/...
Great piece! Provocative q: wouldn't it be desirable for the second wave questions (i.e should this exist?) to be asked before the development, let alone the application of #AI ‘solutions’? Or would this stifle innovation entirely? #AIethics @FrankPasquale https://lpeblog.org/...
.@FrankPasquale lays out the current rifts in AI research: the 1st wave aims to improve data inputs for more equitable outputs; the 2nd wave qns why we need these systems, who gets to govern them & sees reform as another form of “predatory inclusion” https://lpeblog.org/...
“While the first wave of algorithmic accountability focuses on improving existing systems, a second wave of research has asked whether they should be used at all—and, if so, who gets to govern them”, writes @FrankPasquale https://lpeblog.org/...