Eric Alston (University of Wyoming – College of Law) and William Lehr (Massachusetts Institute of Technology (MIT) – Computer Science and Artificial Intelligence Laboratory (CSAIL)) have posted On (Human) Adjudication on SSRN. Here is the abstract:
The judicial role exemplifies a deeper feature of human decision-making under uncertainty: the act of judgment cannot be reduced to algorithmic computation when the rules and standards being applied rest on a substrate of human values that is itself contested and dynamically renegotiated. We identify three canonical classes of legal cases that outstrip the extant information set available at the time of judgment and force adjudicators to engage this contested substrate directly: novelty (where social or technological change generates fact patterns the law was not designed to anticipate), imprecision (where the mapping of even settled facts onto the governing legal standard underdetermines which side of the line of liability the case falls on), and incommensurability (where the case requires the adjudicator to weigh competing social interests that cannot be reduced to a common metric). Drawing on famous U.S. cases, we argue that LLMs face structural limits in each of the three classes that reflect a single underlying problem: algorithmic adjudication can only operate when the value substrate against which evidence is weighed has been fixed in advance, and in hard cases that fixation is exactly what the case is being asked to perform. The argument is robust to advances in AI capability, and we develop it explicitly against current rejoinders, drawing on direct experimental evidence and addressing the prospective objection that AI may develop something value-like through iterative training. Rather than urging AI’s exclusion from judicial applications, we develop a complementary framing: AI can productively augment judicial decision-making in routine matters and constrain bias and noise more broadly, while the dispositive judgment in hard cases remains essentially human. AI does not depoliticize adjudication; at the hard margins, it relocates political and constitutional choice from the adjudicative stage to the design and governance of the system that produces the decision. This carries broader implications for AI integration in public administration generally, and reveals why judicial selection is inherently political.
Highly Recommended!
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum