Will this happen in chess? Imagine Alphazero becoming much stronger! A scary/fascinating thought.https://twitter.com/GaryMarcus/status/1168596718922813440 …
-
2) Agreed. But lets say we add tablebases. That is "facts", not just unverified human knowledge. Surely that can only help? Alphazero is by far the most amazing I have ever seen, but experimenting with improving it sounds like a cool next step.
in the long run i believe that knowledge will help a lot, though our current technology is not that good at incorporating such knowledge. monte carlo tree search is itself a powerful prior that does help a lot, already incorporated into alpha*
-
-
I mentioned incorporating "tablebases". They are actually "facts" and not just knowledge. Anyway interesting times and thanks a lot for the link to your paper!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
The thing about chess (that does not carry over to the real world) is that the rules of chess encode complete knowledge of the game. The AlphaGo family of methods is just applying computation to make that knowledge explicit in the form of training data compressed into a network
-
So despite
@GaryMarcus valid points about prior knowledge in real world tasks, in chess, the only prior knowledge you need is that this is a two-player zero-sum game of perfect information played according to the given rules. - 2 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.