Vavilov and Generative AI

Abstract

This article considers whether a decision made by generative artificial intelligence can satisfy the standard of reasonableness set out in Canada (Minister of Citizenship and Immigration) v. Vavilov. Vavilov requires that administrative decisions be justified through reasons that are transparent and intelligible to the affected party. Earlier scholarship, law, and policy have assumed that AI cannot do this because it cannot provide reasons and its inner workings are opaque or uninterpretable. However, new capabilities of large language models challenge this view. Recent experiments show that when prompted with party submissions and relevant legal materials, generative AI can produce persuasive, legally grounded reasons for decisions. The article evaluates two responses: one argues that AI decisions remain unreasonable under Vavilov since their true basis lies in opaque technical processes; the other contends that Vavilov focuses on the cogency of stated reasons, not how they were generated. The article supports the latter position, suggesting that Vavilov leaves open the possibility that AI-generated decisions can be reasonable, provided their reasons meet the decision-making standard applied to human actors.