> Because of only predicting moves in isolation. This actually may be the reason of higher ranking. What are the odds that a low ranked player will blunder a p

> Because of only predicting moves in isolation. This actually may be the reason of higher ranking. What are the odds that a low ranked player will blunder a piece in a particular position? Quite low. But what are the odds that a low ranked player will blunder a piece in a game? They are quite high. So while this engine may predict most likely move, it can’t fake a likely game because it is too consistent. I think GANs can be helpful to do something like this. One NN tries to make a man-like move and another one tries to guess whether the move was made by a human or the engine given the history of moves in a game.