Moral Perception and Uncertainty Expression in LLM-Augmented Judicial Practice

Abstract

This paper challenges the dominant approach to uncertainty in LLM-augmented judicial practices. Current research primarily focuses on measuring and quantifying uncertainty to prevent overconfidence in AI systems. This narrow framing overlooks how certain forms of uncertainty actively support moral perception in judicial contexts—the ability to recognize ethically significant features in cases that might otherwise go unnoticed. I examine the extent to which institutional frameworks' capacity to surface different types of uncertainty plays an enabling role in shaping this perceptual intelligence, particularly in how these frameworks render certain uncertainties visible while obscuring others.

Drawing on legal theory and phenomenology of moral perception, I demonstrate that technical approaches to uncertainty expression in language models risk privileging measurable forms of uncertainty while suppressing the ethical ambiguities vital to judicial reasoning. I propose design principles for ‘attention-conscious’ interfaces that support both individual judges' perception and collective interpretive practices. These designs would distinguish between different uncertainty types based on purpose and preserve the productive tensions inherent in adjudication. The stakes extend beyond technical optimization to how technological infrastructure shapes our capacity for moral attention—our ability to see past the routinely visible in legal cases. This paper offers concrete recommendations for developing LLM interfaces that enhance rather than diminish the moral perception essential to judicial legitimacy.

Keywords: large language models, uncertainty, judicial practices, normativity, moral perception, moral attention