It was only a matter of time. I believe it is a good thing as it should create a standard that is based on actual science and not opinions?
I'd think this would be almost a trivial exercise for some of the coin grading services that take pictures of the coins that they grade. They've got decent size databases of coin images and the grades to go with each coin. Set up your neural network and start training it. The one complication I can see is how far back their database of sufficient resolution images goes back as it would restrict their training dataset.
It's been tried multiple times including by PCGS. Hasn't worked yet. One major problem is that one image is not sufficient to grade a coin and the angle of different images required varies from coin to coin even if they are the same type. Also differentiating between wear and strike weakness is a non-trivial exercise even for humans.
