A lot of the dumb laws where mercy is frequently allowed are that way because they were designed to be, as catch-alls for harder to define grievances. An AI 'judge' just needs to understand what it means to be a good person.
Of course then that could be expanded to an AI which simply tallies moral points and for which a given person committing a given crime gets a punishment designed to best rehabilitate them not to commit the crime again. But that would also be a system which is judge jury and executioner and not even human, and has some self-designed ultimate moral code...thats a lot to get right, and a lot to go wrong.
> I'm betting in 5 years we'll see lawyers use AI to help them do their job, specifically to help them create logical arguments for their cases.
So, actually, Lawyers are using AI right now - but not in that way. Specifically AI is used to analyze and bring up relevant case law, in the same way a law clerk might. It's not perfect, it still requires humans to clean up, but it saves some hours since it can analyze a lot of law quickly and vaguely run a likelihood analysis.
Law is also in English. English is not formally logical. Frequently laws have subtle wording problems.
What I'm suggesting is computers to take the logic of law code to poke around, tracing logic around and finding the obvious logical inconsistencies. A goal would be reducing law code as well as making it more laymen readable.
Obviously a huge benefit of being judged by humans and peers is a sense of mercy. I'm not saying we should mess with that. Also, I would say that anything the computer found would be a suggestion to congress as a way to refactor law, not trusting the computer to automatically do it.