even at the trial level, judges are very often asked to make decisions that while theoreticaly are based solely on "law" in fact require synthesis of many points of view in past cases with the judge's own person interpretation of justice. In fact, I would say that countries like the US, UK, and Canada that use a common law system could never have computers serve as judges because common law is by definite law that is accreted through generations of opinions in past cases rather than being specifically codified into statutes.

At the appelate level judges are asked to interpret past rulings as well as staturory laws passed by legislative bodies and those enacted by government agencies. On top of all that, judges are given the leeway to decide as a group whether a set of laws either blatantly violates the supreme law of the land (in the US this is the Constitution) or if the past precedent of laws has led us down a path that is now outmoded or ridiculous. Because this process is so subjective and leaves room for personal opinion and error, appelate judges always serve on panels of three or more that together come to opinions. I think it would be rather difficult to create scores of AI judges that would each have its own perspectives and opinions such that would allow for a chorus of voices. A simple collection of like judges that always ruled the same way would be useless unless a team of humans followed behind to reprogram the AI as necessary to reflect changing laws or to correct outlying opinions.

I don't really see how all of the effort needed would save us the time and be worth the risk and sacrifice needed to give up human adjudication.