Grant Shanklin (Yale University – Digital Ethics Center), Claudio Novelli (Yale University – Digital Ethics Center), Giovanni Sartor (European University Institute Law Department), and Luciano Floridi (Yale University – Digital Ethics Center) have posted How Legal AI Architectures Encode Assumptions About Legal Reasoning on SSRN. Here is the abstract:
Artificial intelligence (“AI”) systems are increasingly being purpose-built for legal applications. Yet the question of what current architectural approaches, including neuro-symbolic and agentic systems, each assume about legal reasoning has received little attention. We argue that design choices about representing legal knowledge, organizing inference, and constraining inputs and outputs shape how a system handles central features of legal reasoning in ways that correspond, with differing strength, to positions in contested jurisprudential debates. We propose a methodology for identifying these architectural features and apply it to four families of legal AI. We then use debates in legal theory to formulate those features as jurisprudentially significant assumptions about legal reasoning that, taken together, form each architecture’s implicit picture of legal reasoning. From this, we conclude (1) that the jurisprudential character of legal AI cannot be fully inferred from functional behavior alone; (2) that the fit between an architecture’s jurisprudential assumptions and its legal context of use is a design criterion; and (3) that the large-scale deployment of a single architectural approach across heterogeneous legal contexts may create a structural tension with the jurisprudential pluralism of legal practice.
Highly Recommended!
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum