State of the art and possible development of judging AIs in laws field

Question

I'd like a general explanation of that in AIs that were to mimic judges, prosecutors or lawyers, on very general terms they would act on this way for each case:

A judge AI would give a verdict, having the following input:

All sources of law that have some relationship with the act being judged, knowing the relative importance of each of that sources.

All the facts presented to it, being theorically able to distinguish if some fact were false if there are enough evidence for that.

Prosecutors AI would act into searching the facts that would most help for accusing, matching them with known sources of law so they can make the best accusation.

Lawyers AI would act exactly as prosecutors but for defense of the one being accused.

Obviously AIs are way too far nowadays from something like this, but I find strange that it doesn't even look that this has been done as a proof of concept, the most near thing I find by searching on the Internet is this. I'm not sure of the reach of everything that it's explained there but it looks more of assisting judges, prosecutors and lawyers rather than making what they do. I find this a bit strange, although I'm very far from being an expert on AIs, I think that for very simple cases an AI could even do those things.

In the answer about possible development, I'd like to know about which would be the facts that would make developing an AI like the one I'm mentioning to be possible or not.

PD: I don't find any suitable tag, I ask an admin to put a better one if needed.

Requested edit: My question is about the general state of the art of this field (of AIs being able to somehow mimic a judge, a prosecutor or a lawyer), and how pausible is a development in this way and how it would develop (if this is with current knowledge impossible to even try until a far future, or if we are at least near for stablishing AIs understanding law, which would be the logical way for a type of AIs like them evolving (understing laws, judging, being able to see videos..). I know it's broad so I'm happy with a general explanation.

Answer 1

There has definitely been work relating to judicial sentencing:

In a new study from the National Bureau of Economic Research, economists and computer scientists trained an algorithm to predict whether defendants were a flight risk from their rap sheet and court records using data from hundreds of thousands of cases in New York City. When tested on over a hundred thousand more cases that it hadn’t seen before, the algorithm proved better at predicting what defendants will do after release than judges.
_{Source: M.I.T. Technology Review: How to Upgrade Judges with Machine Learning}

In terms of actually trying cases to deliver a verdict of guilty or not guilty, I think that would be difficult because, although Machine Learning has been effective at outperforming humans in problems where parameters can be clearly defined, we haven't figured out how to tackle the symbol grounding problem. This means that making decisions based on language, which is less precise than mathematics, would be highly problematic. Guilty/Not Guilty is a basic True/False condition, but the rest of the language utilized in law cannot be so precisely translated.