Generative AI and Cognitive Delegation in Legal Practice

The video examines an academic paper I wrote focus on the ethical and procedural challenges of cognitive delegation, where lawyers outsource critical mental functions to generative AI. Rather than focusing on technical glitches, I argue that the primary danger is the appearance of authority, where probabilistic systems produce plausible but fabricated legal citations that evade traditional detection. Through an analysis of international jurisprudence—including cases from the United States, Argentina, and England—the text demonstrates that even prestigious firms fail to catch these "hallucinations" due to automation bias and systemic oversight lapses. It critiques the "human in the loop" strategy as insufficient, suggesting that nominal supervision often fails to prevent the submission of non-existent authorities to the court. Ultimately, the main paper asserts that a lawyer’s signature is a non-delegable certification of professional judgment and institutional responsibility that cannot be replaced by algorithms. Emerging global standards are thus converging on a duty of verification, framing unverified AI reliance not just as a technical error, but as a potential breach of procedural bad faith.

See the paper in SSRN

Legal Theory Blog (Lawrence Solum)