AlphaGo can play games against itself. You can have GANs, MCTS and more. General AI requires a feedback mechanism from the real world. Unless you have an accur

AlphaGo can play games against itself. You can have GANs, MCTS and more. General AI requires a feedback mechanism from the real world. Unless you have an accurate model of it in a computer, you can’t just test whether a joke will be funny without waiting for humans to laugh. You can’t check whether a tailored diet or workout regimen or gene therapy will have good results without humans trying them. So you’ve reduced your AI problem to a harder problem: modeling the world and all of its complexity in a computer, and somehow being able to run simulations faster than the stuff that happens in the actual real world