A chess problem solvable by intuition but not by computers

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Re: A chess problem solvable by intuition but not by computers

    I don't know what kind of crappy chess engine was used to provide that 100 moves 'solution' (a helpmate, by the way), but engines like Houdini and Komodo correctly assess that it's a draw. Even an old version like Fritz 6 will play the correct moves (even though it seems to have trouble with the 50 moves rules).

    So my question is, what engine played the losing 98...Bh6???

    Comment


    • #17
      Re: A chess problem solvable by intuition but not by computers

      Also, a much more interesting game, in my opinion, is the one where Nakamura destroys Rybka by fooling it into avoiding the 50 moves rule because Rybka is two exchanges up.

      That one was a real embarassement for the engines and they corrected that kind of problem since.

      Comment


      • #18
        Re: A chess problem solvable by intuition but not by computers

        Also, Wayne, don't spend too much time of whatever this guy proposes. I've actually been to Oxford a couple of times. Some great research is done over there, but some profoundly dumb stuff, too.

        There's nothing about cracking the secrets of human ingenuity there. Just a simple, drawn position, which all modern engines will play correctly for both sides. Penrose saying that this position fools even supercomputers is more like Penrose trying to fool gullible people. Many false statement in that article on the Telegraph.
        Last edited by Mathieu Cloutier; Wednesday, 15th March, 2017, 11:51 PM.

        Comment


        • #19
          Re: A chess problem solvable by intuition but not by computers

          A chess problem solvable by intuition but not by computers

          March 15, 2017

          Alas Mathieu, I think you are right. In two days I haven’t done a line of my income tax return worrying about the position.

          I wonder if this is the Nakamura game you were thinking of?

          The Five Bishops Ending

          ICC blitz 3 0
          Internet Chess Club
          March 15, 2008
          Rybka Computer – Nakamura, Hikaru
          A00 Grob’s Attack

          1.g4 d5 2.h3 h5 3.g5 g6 4.d4 Bg7 5.Nc3 c6 6.Nf3 Bf5 7.Nh4 e6 8.Nxf5 exf5 9.Bf4 Ne7 10.e3 O-O 11.Qd2 Nd7 12.O-O-O Re8 13.Bg2 Nf8 14.Be5 Ne6 15.f4 Nc8 16.Kb1 Nd6 17.Bf1 Qd7 18.a4 b6 19.Ba6 Nc7 20.Be2 a6 21.Bd3 b5 22.h4 Ne6 23.Ne2 Re7 24.Ka1 Ree8 25.Qa5 Nb7 26.Qd2 Nd6 27.Rh2 Bf8 28.Rf2 Ng7 29.a5 Nb7 30.Ng1 Qd8 31.b4 Qd7 32.Nf3 Nd6 33.Bxd6 Bxd6 34.Ne5 Qb7 35.Kb2 Rac8 36.c3 Re7 37.Be2 Ree8 38.Bf3 Re7 39.Qc2 Rd8 40.Rfd2 Bb8 41.Be2 Re6 42.Bd3 Re7 43.Rf2 Re6 44.Qd2 Re7 45.Rg1 Re6 46.Rff1 Re7 47.Qd1 Re6 48.Rg2 Re7 49.Qc2 Re6 50.Rfg1 Re7 51.Rf2 Re6 52.Qe2 Re7 53.Qf3 Re6 54.Rd2 Re7 55.Rgd1 Re6 56.Qg3 Re7 57.Kb1 Re6 58.Bc2 Re7 59.Nd3 Re6 60.Nc5 Qc8 61.Nxe6 Qxe6 62.Kb2 Re8 63.Re1 Qd7 64.Qf3 Re7 65.Bd3 Re8 66.Rg2 Re7 67.Kb3 Re8 68.Ra2 Re7 69.Rc1 Re8 70.Rca1 Re7 71.Rg1 Re8 72.Rh2 Re7 73.Qg3 Re8 74.Ka3 Re7 75.Rb1 Re8 76.Rd2 Re7 77.Kb3 Re8 78.Ra1 Re7 79.Kb2 Re8 80.Re2 Re7 81.Ree1 Re6 82.Qf3 Qe7 83.Qf2 Qe8 84.Be2 Qe7 85.Bd1 Qe8 86.Bf3 Qe7 87.Qd2 Qe8 88.Kb3 Qe7 89.Kc2 Qe8 90.Kb2 Qe7 91.Rab1 Kf8 92.Qf2 Ke8 93.Be2 Kd8 94.Bd3 Qe8 95.Qf3 Ke7 96.Kb3 Kf8 97.Ra1 Qe7 98.Bc2 Kg8 99.Kb2 Qe8 100.Qf2 Qe7 101.Qd2 Qe8 102.Bd3 Qe7 103.Be2 Qe8 104.Rad1 Qe7 105.Rc1 Qe8 106.Bf3 Qe7 107.Qf2 Qe8 108.Rcd1 Qe7 109.Be2 Re4 110.Bd3 Qe6 111.Bxe4 fxe4 112.Ra1 Nf5 113.Kb3 Kf8 114.Rh1 Ke8 115.Rh2 Kd7 116.Qe1 Kd8 117.Rd1 Kd7 118.Rc2 Kd8 119.Qf2 Nd6 120.Ka2 Qf5 121.Rh1 Kd7 122.Qg3 Ke6 123.Qh3 Ke7 124.Qxf5 Nxf5 125.Rh3 Ke6 126.Kb3 Bc7 127.Re2 Bb8 128.Re1 Bc7 129.Ra1 Bb8 130.Rd1 Bc7 131.Re1 Bb8 132.Re2 Bc7 133.Rg2 Bb8 134.Rg1 Bc7 135.Rb1 Bb8 136.Rbh1 Bc7 137.Rg1 Bb8 138.Kb2 Bc7 139.Kc2 Bb8 140.Kd2 Bc7 141.Rh2 Bb8 142.Rc1 Bc7 143.Ra1 Bb8 144.Ke2 Bc7 145.Rg1 Bb8 146.Rhh1 Bc7 147.Kd2 Bb8 148.Rc1 Bc7 149.Rh3 Bb8 150.Ra1 Bc7 151.Rhh1 Bb8 152.Rae1 Bc7 153.Rh2 Bb8 154.Rh3 Bc7 155.Ra1 Bb8 156.Kc2 Bc7 157.Rf1 Bb8 158.Kb3 Bc7 159.Rf2 Bb8 160.Kc2 Bc7 161.Rg2 Bb8 162.Kd2 Bc7 163.Rh1 Bb8 164.Rf2 Bc7 165.Rfh2 Bb8 166.Rd1 Bc7 167.Rf1 Bb8 168.Re1 Bc7 169.Ke2 Bb8 170.Rh3 Bc7 171.Kd2 Bb8 172.Kc2 Bc7 173.Rb1 Bb8 174.c4 dxc4 175.Kb2 Bd6 176.Ka3 Ne7 177.Rb2 Nd5 178.Rg3 Kf5 179.Rb1 Be7 180.Rh3 Bd6 181.Kb2 Be7 182.Re1 Bxb4 183.Re2 Bxa5 184.Kc1 Bb6 185.Kc2 a5 186.Rh1 a4 187.Kc1 a3 188.Rc2 Ba5 189.Rh3 Bb4 190.Re2 Bd6 191.Re1 b4 192.Kb1 b3 193.Rh2 c3 194.Rc2 Bb4 195.Ka1 bxc2 196.Ka2 Nxe3 197.Kb3 Kxf4 198.Rc1 c5 199.dxc5 Bxc5 200.Rh1 Kg3 201.Ra1 Kxh4 202.Rc1 Kxg5 203.Rg1+ Kf4 204.Rh1 g5 205.Ra1 h4 206.Rc1 h3 207.Kxc3 g4 208.Kb3 g3 209.Ka4 g2 210.Kb5 Bd4 211.Ka6 Bb2 212.Ka7 Bxc1 213.Kb7 Bb2 214.Kb8 h2 215.Kb7 f5 216.Kb6 Ke5 217.Kc6 f4 218.Kb5 f3 219.Kb6 a2 220.Kb7 f2 221.Kc8 f1=B 222.Kb7 g1=B 223.Kb8 h1=B 224.Kb7 c1=N 225.Kc6 a1=B 226.Kd7 Nd5 227.Ke8 Ne7 228.Kxe7 Nd3 229.Kf7 Nc5 230.Ke7 Nd7 231.Kxd7 e3 232.Kd8 e2 233.Kc7 e1=B 234.Kd8 Ba5+ 235.Ke8 Bd8 236.Kf8 Be7+ 237.Kxe7 Bb5 238.Kf8 Bd5 239.Ke7 Bb6 240.Kf8 Bd8 241.Kg7 Ba3 242.Kh6 Bf8+ 243.Kh5 Bf7+ 244.Kg4 Bf1 245.Kg3 Bd4 246.Kg4 Bf2 247.Kf3 Be1 248.Kg4 Bg6 249.Kf3 Bh5+ 250.Ke3 Bh3 251.Kd3 Bg6+ 252.Kc4 Bd7 253.Kb3 Bb5 254.Ka2 Bda5 255.Kb2 Bfb4 256.Ka2 Bec3 257.Kb3 Bbd3 258.Ka4 Bc2+ 259.Kb5 Be8+ 260.Ka6 Bc6 261.Ka7 Bd3 262.Kb8 Bdb5 263.Kc8 Kd6 264.Kb8 Bf6 265.Kc8 Kd5 266.Kb8 Bd6+ 267.Kc8 Bd7+ 268.Kb7 Bd4 269.Ka8 Bac7 270.Kb7 Bdb6 271.Ka8 Bbc6# 0-1

          K7/2bb4/1bbb4/3k4/8/8/8/8 w - - 69 272



          See:
          http://www.chessgames.com/perl/chessgame?gid=1497429
          and
          https://www.chess.com/article/view/c...-evil-part-two

          Comment


          • #20
            Re: A chess problem solvable by intuition but not by computers

            Yes, that's the one. The trick happens around move 180, when Rybka is fooled into giving back a couple of pawns in order to avoid the 50 moves rule while up two exchanges. However, the resulting pawn march from black is both unstoppable and outside the search horizon of the engine.

            It has to be noted that more recent engines don't fall for that trick. But even then, that was a pretty nice trick. After that, Nakamura treated himself to quite a few dozen moves of absurd chess torture. Can't blame him, though. It was some payback for all these games he lost against the same engine.

            And finally, I must say I agree with you that analyzing that position is more interesting than doing income tax.
            Last edited by Mathieu Cloutier; Wednesday, 15th March, 2017, 11:53 PM.

            Comment


            • #21
              Re: A chess problem solvable by intuition but not by computers

              Originally posted by Mathieu Cloutier View Post
              .... There's nothing about cracking the secrets of human ingenuity there. Just a simple, drawn position, which all modern engines will play correctly for both sides. Penrose saying that this position fools even supercomputers is more like Penrose trying to fool gullible people. Many false statement in that article on the Telegraph.

              Before I respond to Mathieu, I have a question about this problem that no one has brought up yet:

              Why does it require 3 dark-square Bishops for Black? Why don't 2 Black dark-square Bishops suffice to demonstrate the problem? Perhaps Wayne Komer, in your analysis, you have found something that indicates Black does require all 3 dark-square Bishops?

              Now to Mathieu:

              You are misunderstanding the point. You say "all computer engines will play correctly for both sides." That isn't the issue here. The issue is whether ANY computer engine will ASSESS correctly this position. And what Penrose says about that is that for any engine to ASSESS the position correctly, would require more computer resources than exist on planet Earth.

              Penrose isn't "trying to fool gullible people". What a ridiculous assertion.

              Human beings, even far below GM strength, even below Master strength, can asses this position correctly in a very short time. The reason they can do this is because their brains act in a neural net fashion (and by the way, in everything I've seen on this problem including the Telegraph article, no one has yet mentioned the term "neural net" -- very surprising). This means that humans of reasonable chess strength have fully understood the rules of chess, including the 50 move rule. When a human chess player sees this position and understands that only the Black Bishops can move, and cannot cover any light squares, and the White King can simply meander among the light squares, they reach the conclusion: that means 50 move rule.

              But chess engines are not programmed to reach any such conclusion. All chess engines are brute force / minimax search engines (although perhaps there are some unknown neural net engines out there that are "in training"). Chess engines are programmed to find best move. Their assessment of a position is a result of doing a minimax search of the move tree, unless the engine is told to use endgame tablebases.

              For any chess engine to determine that 50 move rule comes into play: since it is told nothing about past moves, it means 50 moves per player have to transpire with no captures / pawn advances for the engine to asses 50 move rule applies. 50 moves per player means 100 plies, and a 100 ply search where just the 3 dark-square Bishops can move for Black would most likely require, if using pure minimax algorithm, as Pensrose said, more computational resources than exist on Earth.

              But wait... somewhere in this thread, Egidijus Zeromskis made this point:

              "Thus we left with White king and three Black bishops to dance around for a 50 moves rule. That's in a principle 4 piece tablebase. Should be solve-able for a decent computer in a blink (with proper programming)"

              Very good point, Egidijus. But there's a remaining problem: normally a chess engine only uses tablebases in the endgame phase! This position does NOT represent an endgame phase... although it could, if you had a heuristic for that which I mentioned in a previous post. But somehow, you the author of the chess engine code have to tell the engine when to use tablebases. And guess what: the possible times when that can occur is totally unpredictable. The number of heuristics you'd have to code into that engine to tell it when to switch from a minimax search to using tablebases is beyond comprehension.

              The only type of chess engine that can possibly emulate the workings of the human brain is a neural net chess engine. I'm surprised that Penrose (at least in the Telegraph article) doesn't mention that. And such a neural net engine would't just automatically know what to do. It has to be trained, just as the human brain has to be trained since childhood to play chess.

              So despite what Mathieu says, there is a difference here. This problem does demonstrate human ingenuity versus the rigid inflexible workings of all typical computer chess engines. And let's not forget: the same human who recognizes in 5 minutes or less that this position is a draw.... can in the same day go out and win a tennis match using totally different thinking and totally different functionality.. Try that, computer chess engine!
              Only the rushing is heard...
              Onward flies the bird.

              Comment


              • #22
                Re: A chess problem solvable by intuition but not by computers

                Paul,

                First of all, the problem is titled 'A chess position to defeat computers'. Sorry, but I have to disagree. This position cannot be used to defeat modern engines even running on modest hardware. Penrose claims that even supercomputers don't understand the position. I say bollocks.

                Try playing moves from the position with a modern engine running on infinite analysis. I did it with Komodo and Houdini and it's not long before the evaluation falls to 0.00. i.e. as soon as the 50 moves rules comes within the search horizon of the engine.

                This whole thing boils down to a very simple fact: in chess, there is the 50 moves rule, which is in fact 100 plys for an engine and this is beyond their search depth. Change the 50 moves rule for a 15 or 20 moves rule and any engine will correctly assess the initial position. Nothing esoterical going on here, as Penrose claims.

                Comment


                • #23
                  Re: A chess problem solvable by intuition but not by computers

                  Mathieu, I am in agreement with your position. If a computer just moves its King around, then it understands the position. It realizes that any pawn move would lose. Penrose has no point to make at all.

                  Fake news. (Whether or not that mysterious chess tournament on another thread was also fake news I do not know.)

                  Comment


                  • #24
                    Re: A chess problem solvable by intuition but not by computers

                    Originally posted by Brad Thomson View Post
                    Mathieu, I am in agreement with your position. If a computer just moves its King around, then it understands the position. It realizes that any pawn move would lose. Penrose has no point to make at all.

                    Fake news. (Whether or not that mysterious chess tournament on another thread was also fake news I do not know.)
                    Well, is it fake news when the guy is using government money to conduct research on the pretense that this chess position might be a proof that quantum stuff is happening in our brains and that's why we match well against standard computers for that kind of problems?

                    I mean, it's not my money because it's in another country, but I still have a problem with that kind of 'research' being conducted in such places as Oxford. As a society, we need to invest in research, but when the money is wasted on such stupid projects, it's not helping anyone.

                    Any GM or anyone in the chess community called Penrose on his absurd claims? Because the guys at the funding agency clearly won't be able to do so.

                    Comment


                    • #25
                      Re: A chess problem solvable by intuition but not by computers

                      Article now up on chessbase.com and Friedel brings some sense to that drivel by Penrose. And yes, Friedel has the perfect example, dating back to 1912, of a position that can fool chess engines for real.

                      Penrose is just being the typical old physicist tackling into other fields, thinking he's a universal genius because he worked on black holes. Move along folks, nothing to see here.

                      http://www.smbc-comics.com/?id=2556
                      Last edited by Mathieu Cloutier; Thursday, 16th March, 2017, 12:58 PM.

                      Comment


                      • #26
                        Re: A chess problem solvable by intuition but not by computers

                        I wonder to what extent an engine could "understand" those ones or similar to http://www.chessgames.com/perl/chessgame?gid=1281614 type of endings as well.

                        https://en.wikipedia.org/wiki/Softwa...chess_problems , http://chessexplorer.republika.pl and en.chessbase.com/post/computer-generated-chess-problems-for-everyone talk about specific <composing/solving> engines of puzzles, but perhaps not from the standard way of having a middlegame engine switch at some point in its middlegame to run alone against an ending tablebase look-up ending-engine (usually to improve depth of ply to encourage speed-up during look-up table accesses).

                        I'm not sure whether or not such an engine would benefit much or slow down too much by having a pointer to a solving/composition ending engine, rather than simply a tablebase look-up table engine.

                        In the case of a composing engine, does it use its own ending engine composing engine to check the validity of a study which it has constructed, or some other middlegame or tablebase or mate-in-n solver?

                        Comment


                        • #27
                          Re: A chess problem solvable by intuition but not by computers

                          Originally posted by Paul Bonham View Post
                          Before I respond to Mathieu, I have a question about this problem that no one has brought up yet:

                          Why does it require 3 dark-square Bishops for Black? Why don't 2 Black dark-square Bishops suffice to demonstrate the problem? Perhaps Wayne Komer, in your analysis, you have found something that indicates Black does require all 3 dark-square Bishops?
                          I'm not sure if it was mentioned in Mathieu's reply or elsewhere: the purpose of the 3 bishops is to ensure that the search tree for "all possible moves" will be exponentially larger and thus limit the search depth. Three bishops clog the search tree better than two or one.

                          Steve

                          Comment


                          • #28
                            Re: A chess problem solvable by intuition but not by computers

                            Originally posted by Egidijus Zeromskis View Post
                            Though, which computer will generate a 50-moves rule so quick as humans?

                            I don't know how these days a blockade is programmed but it's quite obvious that locked pieces have zero moves. If White does not take any of rooks and no move with a pawn on c6, those locked pieces will stay forever in the zero-move state. Thus we left with White king and three Black bishops to dance around for a 50 moves rule. That's in a principle 4 piece tablebase. Should be solve-able for a decent computer in a blink (with proper programming) :)

                            Egidijus, I thought of a reason why this can't be reduced to a 4 piece tablebase. All 4 piece tablebase positions require the Black King be included. But the Black King is locked in, so you would have to include all the pieces that are locking it in. You end up having to include all the pieces!
                            Only the rushing is heard...
                            Onward flies the bird.

                            Comment


                            • #29
                              Re: A chess problem solvable by intuition but not by computers

                              Originally posted by Paul Bonham View Post
                              Egidijus, I thought of a reason why this can't be reduced to a 4 piece tablebase. All 4 piece tablebase positions require the Black King be included. But the Black King is locked in, so you would have to include all the pieces that are locking it in. You end up having to include all the pieces!
                              Well, Egid *did* say "with proper progamming" ... the tablebases are only usable (in normal circumstances) when the number of pieces on the board drops to less than 4,5, or 6 - whatever level of tablebases are employed. The fact that some pieces are pylons is not something most/any chess programs would specifically have code to deal with. I haven't tried myself, but I would have to believe that the tablebase includes checks for the 50 move rule of course.

                              I think Mathieu pointed out elsewhere that specific cases *like* this could be programmed but the payoff for that would obviously be so low that it isn't worth the effort. Including blockades, there must be many such positions...
                              ...Mike Pence: the Lord of the fly.

                              Comment


                              • #30
                                Re: A chess problem solvable by intuition but not by computers

                                Originally posted by Kerry Liles View Post
                                Well, Egid *did* say "with proper progamming" ... the tablebases are only usable (in normal circumstances) when the number of pieces on the board drops to less than 4,5, or 6 - whatever level of tablebases are employed. The fact that some pieces are pylons is not something most/any chess programs would specifically have code to deal with. I haven't tried myself, but I would have to believe that the tablebase includes checks for the 50 move rule of course.

                                I think Mathieu pointed out elsewhere that specific cases *like* this could be programmed but the payoff for that would obviously be so low that it isn't worth the effort. Including blockades, there must be many such positions...

                                Yes, Kerry, I wasn't being critical of Egid, just wanted to point that out for this particular case. Sorry if it came across the wrong way.

                                The one sentence you wrote, "The fact that some pieces are pylons is not something most/any chess programs would specifically have code to deal with" is (I believe) the whole point Penrose is making. We humans can adjust our thinking to the fact that so many pieces, and most of them Black, are not movable. We can very quickly draw the conclusion since Black can only move dark square Bishops that 50 move rule will apply.

                                The chess engine would know that 12 of the pieces on the board cannot move, but there would be no code to treat those pieces as a group and then add in the fact that Black can only move dark square Bishops and then apply a heuristic that lets it know 50 move rule will apply.

                                By the way, it was I who mentioned this heuristic and the fact that while it could be added, there would be almost no payoff ever. Perhaps Mathieu mentioned it also.....

                                Anyway, I think Penrose has a valid point even if as Mathieu says his research might be a little over the top in terms of potential payoff. But there are companies trying to create robotic systems, and the ultimate vision of that would be a robot that could do everything a human could do, something on the path to the Data character on Star Trek NG. Or let's even consider self-driving cars. An engine that drives your car for you would need to adjust to weird situations that the designers might never that thought of. I personally don't think that can happen without a neural net, some fantastic hardware, and a LOT of training. Any attempt to just write such an engine from scratch and cover all scenarios is doomed to failure, imo....

                                ....although "failure" in this case might mean only that some scenarios are going to come up that the engine can't handle, and the same can be said of humans. I guess the question is, can they make from scratch a system (no neural net, no training) that can outperform humans in avoiding accidents. That might be possible.... remains to be seen.

                                Overall I think Mathieu's criticism of Penrose here is unwarranted. If the research could lead somehow to better robotic systems, the payoff would be huge.
                                Only the rushing is heard...
                                Onward flies the bird.

                                Comment

                                Working...
                                X