An AI judge won't have any empathy or compassion, no emotions at all. There can be a false positive in the judging process because there is no common sense as well.
You'll end up with a ML program that examines behavior to determine guilt or innocence. You end up with a 24/7 monitoring system with cameras and facial recognition everywhere that looks for certain behavior that might lead to crime based on past behavior of criminals.
You want to end corruption but you replace it with a system that gives you no privacy and spies on you without a warrant. Changes in your behavior could get you arrested on a false positive, like say you have a bad day at work and get stressed out. It detects you are angry about something, and are a danger to yourself or others.
We shouldn't joke about this sort of stuff because one day someone might try to make it happen. After 9/11 there has been a lot of domestic spying by the federal government in the Prism system. https://prism-break.org/en/ other governments have done the same thing. They collect a lot of data on people and try to figure out who is going to do what. But so far they have information overload and can't predict stuff yet. All they need is a better ML program to detect behavior based on past behavior and words.
It kind of is like a sci fi movie with the domestic spying done on the Internet and phones. All they need is a ML program that judges behavior based on the data they collect to arrest people and hold them for trial.
Edit: In the James Bond movie "Specter" they replaced the 00 Program with the Nine Eyes program of drones and cameras monitoring things in 9 different nations that would judge human behavior. The drones would then decide who to kill based on an AI/ML program. The movie established that it was better using human beings to make judgments on matters like that.