You're reading the public-facing archive of the Category Theory Zulip server.
To join the server you need an invite. Anybody can get an invite by contacting Matteo Capucci at name dot surname at gmail dot com.
For all things related to this archive refer to the same person.
Ryan Schwiebert (Mar 06 2026 at 20:12):When I think of knowledge representation, which I understand to mean "representing information in a disciplined way, with as little ambiguity as possible, so that a computer can understand exactly what it means," it sounds like legislation is (an admittedly ambitious but possibly very helpful) a possible point of application.
I mean, if someday AI systems need to conform to laws, and we can specify them in such a way as they can be uploaded into the system so that they can understand the letter of the law and make decisions based on that, it seems highly practical. It also might have the side-effect of making better legislation (with less loopholes and mistakes, with a sort of compiler to check them for consistency and completeness).
I realize that creating laws that are unambiguous seems pretty unrealistic considering the entire art we have around creating and interpreting laws at the moment. I don't need to hear about how hard that problem is :) I'd rather focus on the idea of representing laws that are amenable to representation with something like OLOGs or something similar.
I searched around a bit, but didn't find anything in the intersection of knowledge representation and legislation. Has anyone run across anything like that before?
Kevin Carlson (Mar 06 2026 at 21:19):I would think you’d do much better to think about representing legislation together with jurisprudence. Written laws don’t really mean much of anything on their own, especially in a common law country. But mix in the history of judicial opinions on issues touching on the law’s matter, and you might have a chance.
Ryan Schwiebert (Mar 06 2026 at 21:23):@Kevin Carlson that’s probably a component of it, yes. Whatever an agent would need to know to comply with the law.
Maybe jurisprudence is something overarchingnover all laws. Or maybe it’s a sidecar of information that can be updated without changing the text of the law.
Moby-Dick (Mar 06 2026 at 22:54):https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1467-9337.1990.tb00060.x
Ryan Wisnesky (Mar 07 2026 at 00:01):Things like the French tax code have been formalized to the point where correctness proofs have succeeded: https://inria.hal.science/hal-03002266v3/document . I think though you would need more than just the limits and colimits an olog provides to axiomatize such systems. Years ago David and I met a professor who tried to start a company formalizing the law, called "lawgorithms". The logic he needed was multi-valued, because he claimed in the law statements aren't true or false, but always have degrees of possibility attached to them. Moreover, it seems like one would really need to be able to formalize a big chunk of English to formalize any law.
Matteo Capucci (he/him) (Mar 07 2026 at 10:57):Ryan Wisnesky said:
Things like the French tax code have been formalized to the point where correctness proofs have succeeded: https://inria.hal.science/hal-03002266v3/document .
Excellent quote from the abstract:
This algorithm relies on a legacy custom language and compiler originally designed in 1990, which unlike French wine, did not age well with time.
Matteo Capucci (he/him) (Mar 07 2026 at 11:00):Ryan Schwiebert said:
I mean, if someday AI systems need to conform to laws, and we can specify them in such a way as they can be uploaded into the system so that they can understand the letter of the law and make decisions based on that, it seems highly practical.
Well, at the moment it seems that LLMs allow exactly to bypass this problem, since they 'understand' natural language.
Matteo Capucci (he/him) (Mar 07 2026 at 11:01):In any case, Catala is the formal language you are looking for. They also list alternatives.
Morgan Rogers (he/him) (Mar 07 2026 at 11:44):Worth mentioning work of Luc Pellissier in this context regarding how e.g. tax codes are necessarily interpreted by algorithms (for automatic calculation of taxes) and how the process for that transformation lacks accountability.
Ryan Schwiebert (Mar 07 2026 at 13:47):Matteo Capucci (he/him) said:
Well, at the moment it seems that LLMs allow exactly to bypass this problem, since they 'understand' natural language.
Whoof, I dunno about you but that grade of "understanding" does not put me at ease at all.
Ryan Schwiebert (Mar 07 2026 at 13:48):Matteo Capucci (he/him) said:
In any case, Catala is the formal language you are looking for. They also list alternatives.
Nice! Thanks for the reading...