Correct and one of the key causes is an understaffed and under trained USPTO staff. Even if you could train them to understand the (constantly changing) state o

Correct and one of the key causes is an understaffed and under trained USPTO staff. Even if you could train them to understand the (constantly changing) state of the art in ML learning, it could take years to truly determine if there was prior art or not. I still think the way to solve the patent issue for software is that you make it's time limited to 5 years and make them non transferrable. Unlike patents on physical things, it is much easier to make something similiar that just works around a single one of the claims. Another thing we probably need is that any patent dispute that goes to court requires a prior art check and "novel" reassessment and if it is found that the patent was never novel or prior art existed, then the patent is thrown out. This could be done as part of the USPTO office instead of the courts as a precursor to actual trial.