Large language models generate fluent, personalized text at scale, eroding the civic accountability traditionally anchored by human speakers. Artificial intelligence systems can participate in institutional processes like public dockets or surveys without bearing responsibility for their output, unlike human agents who face sanctions, shame, or demands for repair.
This proliferation of unanchored speech hollowing out public trust requires institutional redesign to preserve human responsibility alongside automated tools. Delegation succeeds when a human sets purpose and retains responsibility, but fails when synthetic text enters consequential processes without adoption. While technical alignment ensures systems behave as designed, it cannot confer democratic legitimacy or substitute for enduring social relationships built on verifiable accountability. Without chains of responsibility linking words to accountable persons, the architecture of civic trust collapses.
Generative models alter the integrity of public consultations by replacing verifiable human agency with automated text generation. When 18 million public comments on federal rules can be simulated effortlessly, administrative dockets lose the foundational feedback loop connecting policy decisions to affected populations. Unlike human respondents who risk reputational or legal consequences, deployed algorithms possess no continuity of identity and absorb zero liability for erroneous output.
No comments:
Post a Comment