Can Humans Evolve to Beat the Engines

[towforce]

The good news: uncovering new chess knowledge (deeper patterns in chess) would enable humans to easily beat today's chess computers.

Obviously this is false.
Chess is a concrete game with a search tree. Of course a human cannot match a machine which looks at millions of times more nodes, anymore than a runner can say "I'm sure in the future humans will run as fast as Ferraris".

As a human runner, I'm introducing a flight of stairs onto the path!

The point is easy to debunk. Go back to 1997, before the second Kasparov vs Deeper Blue match:

1. Engines looked at many orders of magnitude more nodes on the search tree than any human

2. At that time, they were hopelessly unable to compete with top humans

3. As previously stated in this thread, a way was found to enable a middling Go player to beat a program that is stronger than AlphaGo

[jefk]

chess is not Go (yep i repeat myself)

although it's true that (positional) knowledge can compensate for search (as i a second
person here, a philosopher(*) told/mailed me some weeks ago); nevertheless it's
not very efficient imo if build some neural net bigger than our cosmos
to compensate for a lack of tactical alfa beta search.

PS still waiting for the winning line for White; with 1. a4! maybe ?
well good luck


(*) yeah the person who asked for some comments about his book chapter
(about computer chess; i've seen his chapter quite good actually and mailed
some comments, for which he was grateful; some scientists can be polite)

[towforce]

chess is not Go (yep i repeat myself)

Obviously they're different games. The point is that, because it has a large branching factor, Go is less computer-friendly than chess, and therefore we are able to see computer game-playing issues more clearly.

..although it's true that (positional) knowledge can compensate for search (as i a second person here, a philosopher(*) told/mailed me some weeks ago); nevertheless it's not very efficient imo if build some neural net bigger than our cosmos to compensate for a lack of tactical alfa beta search.

Look around you and you'll see examples everywhere of animals displaying amazingly intelligent behaviours with small brains. My favourite example: the fruit fly. It has a large range of intelligent behaviours, but its brain only contains 125,000 neurons.

The key to cracking chess, and similar games, will be to find the deep underlying patterns that make it easy to play perfectly with only a small, cheap computer, or a human trained to use these patterns. Current chess NNs are not doing this: they're instead finding a large number of simple (shallow) patterns.

[Werewolf]

The good news: uncovering new chess knowledge (deeper patterns in chess) would enable humans to easily beat today's chess computers.

Obviously this is false.
Chess is a concrete game with a search tree. Of course a human cannot match a machine which looks at millions of times more nodes, anymore than a runner can say "I'm sure in the future humans will run as fast as Ferraris".

As a human runner, I'm introducing a flight of stairs onto the path!

The point is easy to debunk. Go back to 1997, before the second Kasparov vs Deeper Blue match:

1. Engines looked at many orders of magnitude more nodes on the search tree than any human

2. At that time, they were hopelessly unable to compete with top humans

3. As previously stated in this thread, a way was found to enable a middling Go player to beat a program that is stronger than AlphaGo

Yes but that's before engines had a decent EF.
Now with neural nets they know what they're doing, provided they have a decent search.

I can make the point even more strongly: not only will your future human genius never beat Stockfish (at sensible settings, decent hardware, opening book etc) - but your future human working with a computer won't be able to either!

Chess is nearly dead from a competitive point of view because although SF isn't perfect, it's good enough to hold the draw. Your future human just won't win.