AI successes with board games still rely on "unfair machine advantages" such as impossibly deep tree search or training on millions of games
Aren't humans still vastly better at doing pattern recognition on a game board, given a human-size training dataset?
-
-
Agree re: human size training dataset & computation restrictions. Depending on underlying task, that's either v important or not at all.
-
Transfer learning is also important re: humans. Many tasks aren't "do well at Go", it's the equivalent of "you've seen these games in the...
- Show replies
New conversation -
-
-
We do get several decades of general purpose training before we look at the board, though.
-
I agree with
@BenedictEvans. We are just good at transferring knowledge. That’s why there are classes on how to play chess, etc
End of conversation
New conversation -
-
-
We start with a pretty powerful (evolved) Cognitive Architecture. An insect, a bird, or even a horse are ready to perform nearly from birth.pic.twitter.com/ZIxmjDVGYu
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Humans, as a perpetual learning machine, also have the advantage of training in real world for 10s of years before they played the game!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
But humans can "cheat" by learnin from other's experiences (discussion, tutorials...), extracting distilled knowledge from other datasets.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Humans can learn to decently play _any_ game in a matter of hours while machines need a proper unique design and a lot of time for each.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Well, humans have datasets (decades of experience) orders of magnitude larger than anything else.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.