"Alignment as Jurisprudence" by Caputo

Nicholas Caputo (Harvard Law School) has published "Alignment as Jurisprudence" on SSRN. Here is the abstract: 

The fields of jurisprudence and the alignment of artificial intelligence have similar structures and confront similar problems. This paper puts them into conversation by illustrating those similarities across two important theories of jurisprudence, Dworkin's interpretivism and Sunstein's arguments for analogical reasoning, and two leading approaches to alignment, Constitutional AI and case-based reasoning. In particular, I argue that a conversation between these two fields will help each overcome core problems with representativeness or democratic accountability and scaling oversight, suggesting that progress on these issues might come from the development and articulation of meta-principles that make how new cases will be decided by judges and new outputs generated by AI models more predictable and controllable.