Implementing Computational Law in Wolfram Language for AI Governance
A new study on arXiv (2608.13958) presents a way to apply Reified Input/Output Logic using the Wolfram Language, aiming to create a framework for managing AI systems through defined legal standards. The author suggests that governing AI effectively doesn’t require understanding its reasoning but rather specifying rules regarding what is allowed, required, or forbidden while ensuring compliance. The study includes essential I/O principles and examines GPT-4’s ability to transform legal English into this formal format, highlighting problems like imagined functions and missing time context. It also features a case study of an AI guard dog operating under a computational contract, demonstrating how formal regulations can evolve into functional code. This work is significant for AI governance and computational law.
Key facts
- Paper on arXiv: 2608.13958
- Implements Reified Input/Output Logic in Wolfram Language
- Formalism behind DAPRECO knowledge base
- Tests GPT-4 for translating legal statements
- Reports failures: hallucinated functions, omitted temporal scope, deviation, silent wrong norms
- Case study: AI guard dog under computational contract
- Argues governance requires stating obligations, permissions, prohibitions, and checking compliance
- Includes core I/O axioms, obligations, permissions, constitutive norms, reified eventualities, temporal operators
Entities
Institutions
- arXiv
- DAPRECO