The "smart contract" stuff from the cryptocoin crowd is more-or-less an attempt at actually doing the "impartial AI judge" thing. I suppose it'll be interesting to see if they get it sufficiently free of bugs and surprise edge cases that it's useful for more than enthusiasts or people doing sketchy things.

But it seems to be getting closer to the point where we can say AI judges are generally unwise based on real experiment rather than on thought experiment. (Assuming it turns out the way I expect.)