Great response, I must say. As a conjecture, do you think splitting the logic up into several smaller neural networks with different objectives (i.e. for chess

Great response, I must say. As a conjecture, do you think splitting the logic up into several smaller neural networks with different objectives (i.e. for chess, one evaluates defensive maneuvers, one for attack, and so on) would alleviate the problem? Or at least improve the performance of the machine as a whole...