Showing posts from June, 2020
Instead of an Imitation Game, why not an Emulation Game? Instead of training AI to imitate human beings - which seems both wrongheaded and dangerous, because we are training them in the art of deception - we should be training them to figure out the truth. Give an AI a data set, let it train on that.  Now give it an incomplete piece of data, and let it predict what the missing piece is.  Then give it the missing piece, and evaluate how well it predicted.  Give it a score.  Now do the same thing over and over, lots and lots of times, so that it becomes good at figuring out the truth.