Paper by Aileen Nielsen, Chelse Swoopes and Elena Glassman: “As large language models (LLMs) enter judicial workflows, courts face mounting risks of uncritical reliance, conceptual brittleness, and procedural opacity in the unguided use of these tools. Jurists’ early ventures have attracted both praise and scrutiny, yet they have unfolded without critical attention to the role of interface design. This Essay argues that interface design is not a neutral conduit but rather a critical variable in shaping how judges can and will interact with LLM-generated content. Using Judge Newsom’s recent concurrences in Snell and Deleon as case studies, we show how more thoughtfully designed, AI-resilient interfaces could have mitigated problems of opacity, reproducibility, and conceptual brittleness identified in his explorative LLM-informed adjudication.
We offer a course correction on the legal community’s uncritical acceptance of the chat interface for LLM-assisted work. Proprietary consumer-facing chat interfaces are deeply problematic when used for adjudication. Such interfaces obscure the underlying stochasticity of model outputs and fail to support critical engagement with such outputs. In contrast, we describe existing, open-source interfaces designed to support reproducible workflows, enhance user awareness of LLM limitations, and preserve interpretive agency. Such tools could encourage judges to scrutinize LLM outputs, in part by offering affordances for scaling, archiving, and visualizing LLM outputs that are lacking in proprietary chat interfaces. We particularly caution against the uncritical use of LLMs in “hard cases,” where human uncertainty may perversely increase reliance on AI tools just when those tools may be more likely to fail.
Beyond critique, we chart a path forward by articulating a broader vision for AI-resilient law: a system of incorporating law that would support judicial transparency, improve efficiency without compromising legitimacy, and open new possibilities for LLM-augmented legal reading and writing. Interface design is essential to legal AI governance. By foregrounding the design of human-AI interactions, this work proposes to reorient the legal community toward a more principled and truly generative approach to integrating LLMs into legal practice…(More)“.
