No One to Blame: A Framework of Constitutive AI Unaccountability

Recent research introduces the concept of 'constitutive AI unaccountability,' arguing that certain configurations of actors, systems, and institutions inherently prevent AI accountability, challenging the prevailing view that such gaps are solvable through improved standards or transparency. This framework suggests that some AI accountability issues are fundamentally unachievable, moving beyond the current understanding of AI accountability challenges.

Why it matters

This research is critical for understanding the inherent limitations of accountability in advanced AI systems, suggesting that some challenges may not be resolvable through conventional means. It necessitates a re-evaluation of how organizations and societies approach the governance and deployment of increasingly autonomous AI.

Key insights

• The deployment of autonomous, agentic AI systems challenges traditional accountability mechanisms.
• Existing research primarily views AI accountability gaps as barriers addressable by better standards, transparency, and institutional reform.
• This prevailing framing is deemed insufficient, as certain configurations of actors, systems, and institutions render AI accountability conceptually unachievable.
• The study introduces 'constitutive AI unaccountability' to describe these configurations, where accountability cannot be established regardless of effort.
• The findings are based on a three-stage qualitative study, including literature analysis and expert interviews with AI professionals.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.12104