Do you think it might be possible to eventually use a Deep Learning model like GPT-3 to do the formalisation for us? It seems like a herculean task to do it by hand.

GPT-3 is a good step but not enough. We had some good result trying to use it to translate legaleese to more parasable pseudo code. However there is still to much approximation for it to be useful today.

Short/mid term I believe that we will have tool allowing lawyer / regulator to draft in a formal form from which we will derive the legaleese. The stanford LSP is pushing in this direction and there is early tentative tools already published such as blawx or Oracle Policy Automation. I m aware of at least 3 other tool under development that push for this paradigm.

Fun fact one of the earliest effort in computational law involve lawyer writing contract in a formal english in ~1957 (original paper still full of good ideas on legal formalism https://digitalcommons.law.yale.edu/cgi/viewcont )

I was wandering the same thing, here in my country a lot of reforms are stoped by the constitutional tribunal, basically they are a bunch of lawyers that implement a function,

constitutional :: Constitution -> (Law | Reform) -> Boolean

I really would like to see if one can train a model like that. seeing what gpt-3 can do, it seems that we are not that far away to achieve something like an AI lawyer.

It's already been tried : http://ceur-ws.org/Vol-2143/paper9.pdf. In a nutshell : NLP techniques don't work so well on the text of law, because it's too messy.

> If you want to work on this topic DM me :)

hnchat seems dead from a quick google :(

Deeply interested in this, thought a lot about this before, but can't work on it right now (at my startup). Do you have a blog or something? If private, pls shoot me a mail (address in bio)