Roger Penrose

Originally posted by BigDoggProblem
That's more a tedious task than a difficult one...

Edit: but since I'm stuck at work anyway...
[pgn][Event "Chess for Android Game"]
[Site "United States"]
[Date "2017.03.14"]
[Round "?"]
[White "Self"]
[Black "Self"]
[Result "*"]
[PlyCount "89"]
[TimeControl "-"]
1. d4 c5 2. Qd2 Qa5 3. Qb4 cxb4 4. d5 Kd8 5. Nc3 Kc7 6. Ne4 Kb6 7. Nd6 exd6 ...[text shortened]... 8. dxc6 Rh8 39. Kf3 Rh2 40. Ke4 Rd2 41. Kf3 Rd5 42. Ke2 Rb5 43. Nc5+ dxc5 44. c4 b6 45. b3[/pgn]

In my mind, this post is more interesting- and amazing- than Penrose's claim. It is indeed the journey, not the destination. Nice work!

The article is so unclear!
"The Penrose Institute scientists are inviting readers to workout how white can win, or force a stalemate and then share their reasoning."
White can't win, without black deliberately trying to lose.

"The main goal is to force a draw, although it is even possible to trick black into a blunder that might allow white to win."
Can you really 'trick' a computer into a blunder?

"The first person who can demonstrate the solution legally will receive a bonus prize.
Both humans, computers and even quantum computers are invited to play the game and solutions should be emailed to puzzles@penroseinstitute.com."
What does "demonstrate the solution legally" mean?

"The three bishops forces the computer to perform a massive search of possible positions that will rapidly expand to something that exceeds all the computational power on planet earth."
I think this is perhaps clearer. If you want to brute-force it then yes it will be massive on the computation front. Doing some rough calculations: there are about 25 different squares that black can move to in each move, followed by about 4 possible squares for the white King. (These are approximate, because it depends on the previous moves. Many times, the white King won't be able to move to black squares because the black bishops are covering them. If the King is in a corner then there are only 3 (or fewer) legal moves.)
For 50 moves this would give 100^50 possibilities, which is 10^100. Therefore there are a googol possible ways to reach then end of move 50 and the draw.
Surely chess boxes don't work like this, though...

Originally posted by Diapason
The article is so unclear!
"The Penrose Institute scientists are inviting readers to workout how white can win, or force a stalemate and then share their reasoning."
White can't win, without black deliberately trying to lose.

"The main goal is to force a draw, although it is even possible to trick black into a blunder that might allow white to win." ...[text shortened]... to reach then end of move 50 and the draw.
Surely chess boxes don't work like this, though...

In other words he is saying...

To win simply write down every digit for pi.

Originally posted by byedidia
I have many issues with this article. First is the idea that a computer will botch this up. The computers very clearly make the right moves even if their evaluation of the position is wrong. My computer, HIARCS, has harmless king moves for every move in all the lines. Second is the use of the phrase, "force stalemate." Stalemate and draw are not synony ...[text shortened]... rapidly expand to something that exceeds all the computational power on planet earth." Really?

Well, assuming the machine works by an alpha-beta search then the engine won't be able to prune the tree since all the positions will be identical, so the machine will never twig that it's a draw. The number of positions the machine has to evaluate diverges, but that will happen if you give an engine the starting position in chess and don't let it use its opening database.

Penrose has a theory that there is some linkage between consciousness and quantum physics, which means that humans can solve some classes of problems faster than machines can. The problem is that AIs based around neural networks can now defeat strong Go players, so it's not clear to me that computers aren't better than humans at solving any problem. The difference between humans and machines is liable to be at the level of qualia, so that no matter how fast they are or good at solving problems they'll never be anything other than philosophical zombies. Anyway I think that Penrose regards this position as some sort of practical demonstration of his theory, I don't think he's right.

But how fast can a top notvch Rybka, Komodo, Fritz (what ever mark it is these days)
Hiarcs chess computer come up with draw eval.

Right away, 10 seconds....30 seconds.? Anybody know?

Move the king to g1 and play g2 stalemate

Originally posted by Marinkatomb
Move the king to g1 and play g2 stalemate

How does white play g2 there is nothing on the g-file. Do you mean b1? I'm not seeing a stalemate.

Originally posted by Eladar
[b]“If you put this puzzle into a chess computer it just assumes a black win because of the number of pieces and positions, but a human will look at this and know quickly that is not the case,” said Sir Roger.


Looks like Sir Roger believes ahead means a win. My computer on deep analysis never came up with a win, just a consistant advantage which I assume is on material.

I think the point is that a computer will say that black is ahead. We all know this is wrong, you included, how can black be ahead when any of us humans playing white can guarantee at least a draw without breaking into a sweat?

Originally posted by DeepThought
Well, assuming the machine works by an alpha-beta search then the engine won't be able to prune the tree since all the positions will be identical, so the machine will never twig that it's a draw. The number of positions the machine has to evaluate diverges, but that will happen if you give an engine the starting position in chess and don't let it use i ...[text shortened]... s this position as some sort of practical demonstration of his theory, I don't think he's right.

I think he has come up with a fine example of the current limitations of silicon thinking and AI in general. Personally I hope he is right and I think he is but time is a limitation on all us mortals.

Originally posted by LittleDonkey
I think the point is that a computer will say that black is ahead. We all know this is wrong, you included, how can black be ahead when any of us humans playing white can guarantee at least a draw without breaking into a sweat?

Poor programming which does not take into account mobility.

Combine that with the faulty belief that ahead in material is winning.

Originally posted by Eladar
Poor programming which does not take into account mobility.

Combine that with the faulty belief that ahead in material is winning.

Fair points but give a computer a mate in one to white vs overwhelming black material advantage, the computer will always show the mate in one and score the position as a decisive win for white. Here it is a clear draw but the computer won't score it as 0.00

Originally posted by LittleDonkey
Fair points but give a computer a mate in one to white vs overwhelming black material advantage, the computer will always show the mate in one and score the position as a decisive win for white. Here it is a clear draw but the computer won't score it as 0.00

I guess the question is what is the purpose of the chess engine, or what was it designed to do.

Was it designed to give absolutes, other than forced mates?

Originally posted by Eladar
I guess the question is what is the purpose of the chess engine, or what was it designed to do.

Was it designed to give absolutes, other than forced mates?

A really good question. Chess has often been held up as an example of human intelligence something that could never be matched by another animal or machine. Proof that we are the superior beings on this planet. Then the machine came along and relatively quickly knocked us off our perch. We hung in there very briefly by saying that Go! was a much tougher nut to crack but we have been beaten again.

It was all originally a challenge, invent a machine that can beat a human at their own intellectual pastime. If a machine can win at chess then surely it can make better business decisions, play the stock market better, find cures for diseases, do everything better, what need is there for humans?

So Penrose is trying to demonstrate that engines are just crunching big data very quickly and are still (and potentially always will be) inferior to humans because we can all see (with a little relatively shallow analysis) that his position is a draw. Something else is going on inside us that (as yet?) is not occurring in the machine. If he is right we are back at number one! If he is right...

Originally posted by LittleDonkey
A really good question. Chess has often been held up as an example of human intelligence something that could never be matched by another animal or machine. Proof that we are the superior beings on this planet. Then the machine came along and relatively quickly knocked us off our perch. We hung in there very briefly by saying that Go! was a much toughe ...[text shortened]... t?) is not occurring in the machine. If he is right we are back at number one! If he is right...

Look at it this way: who built the machine?

Originally posted by LittleDonkey
A really good question. Chess has often been held up as an example of human intelligence something that could never be matched by another animal or machine. Proof that we are the superior beings on this planet. Then the machine came along and relatively quickly knocked us off our perch. We hung in there very briefly by saying that Go! was a much toughe ...[text shortened]... t?) is not occurring in the machine. If he is right we are back at number one! If he is right...

I'm a little more of a pragmatist. What does it mean for a computer to be intelligent? I think it's pretty clear by now that while chess can be addressed by "intelligence" it can also be tackled by brute force and some elegant programming. We completely understand what engines are doing when they play chess, even though we cannot do it nearly as well. What was interesting about the go effort was that we did NOT understand what the computer was thinking when it beat Lee Sedol. It didn't solve it by brute force calculation. They are using a completely different approach for go where the computer learns from millions of games played by people and by itself.

I don't buy Penrose's idea that because the computer engines don't see this position as 0.00, AI still doesn't match what humans can do. a) Computers will play this to a draw. and b) So what? Just because the engines misevaluate the outlook of a position (but still play correctly) doesn't mean that AI is not yet viable. I am sure someone could create an engine that learned chess the way deepmind learned go.