Computational Law as a Blueprint for AI Reasoning

Recent AI systems are increasingly described as “reasoning models,” capable of producing multi-step analyses and explanations. Yet growing research suggests that these claims may outpace reality: strong performance on standard evaluations does not necessarily demonstrate genuine reasoning. This raises fundamental questions. Do these systems possess correct knowledge that can be inspected? Can they associate formal guarantees with the conclusions they derive from that knowledge? The legal domain offers an instructive contrast. When the rules of law are available in explicit form — computational law has demonstrated that legal knowledge can be encoded and applied through deductive reasoning at scale. Systems based on these principles can perform precise calculations, enforce regulatory rules, and provide conclusions whose derivations are transparent and verifiable. Legal reasoning, however, extends beyond rule execution. It also involves interpreting statutes, applying precedents, and drawing analogies across cases. Even in these settings, the legal system demands standards of soundness, transparency, and accountability that go far beyond producing plausible text. This panel explores what modern AI can learn from the successes of computational law. Can computational law representations provide an anchor for AI reasoning that combines the scale of modern machine learning with the guarantees of formal logic? And what frameworks are needed to evaluate reasoning systems whose outputs increasingly influence legal and regulatory decisions? Panelists will address questions such as:

What genuinely new reasoning capabilities are emerging in recent AI models, and where do they fall short of the standards required in legal reasoning?

What has computational law already achieved in encoding legal rules and performing deductive reasoning at scale?

Can we catalogue the major forms of legal reasoning and the guarantees they require?

What formal tools from computer science can help provide guarantees for reasoning by induction, analogy, and prediction?

What architectures could combine the scale and flexibility of modern AI with the transparency and guarantees of computational law systems?

The goal is to move the discussion from vague claims of AI “reasoning” toward concrete foundations for systems that can responsibly augment human reasoning in law.