Can Houdini 3 be beaten?

Bro... I don't know if this is too much info all at once, but Houdini has an estimated FIDE rating well over 3000.

I doubt yours is over 1700...

Are all the experts wrong and you're right? I don't know if you really think you're winning or not, but there's just no way you could do this.

You said "Iv'e drawn Houdini SEVERAL times, do you know how insane that sounds?

Look, try beating the computer here on chess.com http://www.chess.com/play/computer on it's most difficult setting.

I doubt it's better than Houdini, (sarcasm) if you lose to it, you're doing something wrong...

On my computer Houdini takes only 35 sec to find Bf7 (and my PC is not a very powerful machine...)

Bizzaro thread. 

I contacted a GM who is an expert on computers and he agreed that my computer is  pretty good, but called it, "Baby Houdini" compared to his industrial grade computer.

For humans to defeat computers we must rethink development, especially as it applies to Knights. Knights can gain time by retreating, thus in the move sequence Nf3/Ne1/Nd3 all are developing moves. You will note in my Game against Houdini 3 in the Wilkes Barre presented below, each of these is a developing move.

As applied to Knights---Development Rule #1. Any tempo required to  permit a Knight to access a given square is a developing move. In the game below where I outplayed Houdini in a  complex middle game, each of these moves is a developing move---Nc3/Na4/b3/Nb2/Nd1/f3/Nf2. At the end of this  complex maneuver White was better because Black did not make a corresponding developing move.

Moody-Houdini 3 Game 6

5.Bxf7ch Ke7 6.Bc4! (There is almost no analysis of this variation and no master games where this was played except for one crush by White reported by Estrin in Wilkes-Barre Variation Two Knights Defense. Williams, The Real American Wilkes-Barre, offers almost no analysis; MCO 13, 14, or 15, offer no analysis, nor does BCO 2 ---White anticipates an exchange sacrifice on f3 which can be neutralized if White can play Be2) Na5?! (One way to meet the threat of Nf7, the point being if Nf7 Qe8 8.Nxh8 Nxc4 and the Knight is dead---Qe8 is equivalent, +/- [I did get an advantage later in the game of +-]; Rf8 is +/=)7.Bd3!! (When I first studied this variation, Houdini found this; I exploit this here. White defends e4 in anticipation of h6 8.Nf3 Nxe4; true, White has an advantage after Nxe4 but it isn't as great) h6 8.Nf3 d6 9.Nc3 Be6 10.Be2 (I have been able to draw with Na4, but decided to try something different; rather than play tactically against the computer, I try quiet, positional play) Nc6 (Demonstrating that Na5 is a bad idea) 11.d3 a6 12.Na4 (To free the c-pawn) Ba7 13.O-O (I considered b3 with the idea of Nb2) Kf7 14.h3 (Luft and taking away g4) Rf8 15.b3 (Anticipating an isolated pawn on e5 after Black plays d5) Kg8 16.Be3 Bd4 17.c3 (Houdini doesn't like this move; the point is that Nb2 defends the d3 square. There is no threat against f3 or f2, but now I have to watch for cheapos on h3) Bxe3 18.fxe3 b5 19.Nb2 d5 20.exd5 Nxd5 21.Qc1 (I still can't see any tactical shots against my King but I'm wondering what the silicon beast has up its sleeve) Qd6 (Now I think that White is better) 22.Nd1! (I didn't calculate anything in depth here; I couldn't see any obvious tactical shots. No matter what I need this Knight on the Kingside. This defends e3 and c3 and prepares Nf2 holding h3) Rab8?! (I don't understand this move) 23.c4 (This seems to make the most sense; now Black must exchange a critical pawn and give me a solid pawn majority in the center) bxc4 24.bxc4 Nf6 25.Nf2 (Now I'm stuck; I played this quickly; perhaps it is not best) Qc5 26.Rd1 (Preparing d4) Qa5 27.Rd2 (I know Houdini is seeing ahead 10-12 moves, so this part of the game is pretty scary. Right now I'm just trying to maintain good piece coordination) Bf5 28.d4 e4 29.Nh2 (Intending to recycle the Knight to the f1 square where it defends e3----suddenly it hit me! I can hit e4 four times!) Kh8 30.Nf1 (I'm winning, and, with a big lead on the clock, I just have to watch out for the shots; this obeys the edict of Nimzowitch to overprotect the base of the pawn chain) Qb4 31.Nd1 Qd6 32.c5 Qe7 33.Rb2 (Rooks belong on open files) Rxb2 34.Qxb2 (Each exchange brings White closer to victory) Nd5? (Neither player factored in just how dangerous the passed pawn is) 35.Bxa6 (I can't understand what I am missing. Houdini is in mild time trouble---18 minutes until the time control; I have over an hour on the clock. Clearly Houdini's evaluation algorithm has gone haywire. Giving me an unstoppable passed pawn is a bad idea.) Rb8 36.Qe2 (Just connecting my pieces and anticipating Nb4/Nd3) Qg5 (With the obvious hit on h3) 37.Kh2 h5? (At long last the win should be a matter of technique) 38.Bc4 Nce7 39.Bxd5? (Missing the simple a4/a5/a6---Passed pawns must be pushed!) Nxd5 40.Qe1 (Just getting ready to push the a-pawn. I have good piece coordination and two extra pawns Re8 41.a4 Bc8 42.Nc3 Nxc3 43.Qxc3 (I was threatening to walk off with the e-pawn) h4 44.Qe1 (When I push the a-pawn I can get Black to give up pressure on the h3 square. I can combine threats to Queen the pawn and occupy the e5 square with my Knight) Ba6 45.Nd2 Bd3 46.a5 Re6 47.g4 (Necessity is the mother of invention; I need to defend the g2 square) g6 48.Ra2 Kg7 49.Rb2 Re7 50.Ra2? (I do not have a clue how to win this) Re6 51.Ra3 Ra6 52.Ra1 Qd8 53.Kg2 c6 54.Nf1 Ra7 55.Nh2 (Draw offer declined) g5 56.Nf1 Bxf1 57.Kxf1 Rf7ch 58.Kg2 Rf3 59.a6 Rg3ch 60.Kh2?? (Kh1 =) Qc7 0-1

I HATE chess computers!

Yes. I have several pterodactyls. And I've trained them. To SMD.

Houdini 3 picks Bf7ch about 1/3 of the time; it leads immediately to a draw by repetition i.e. 9.Bf7ch Kf8 10.Qd3 (f3? Nxe4 with a strong attack) Be2 11.Qh3 Bg4 12.Qd3 Be2 =)

Left free to pick its own openings Houdini will be tough to beat, but require it to follow human theory, it can be beaten. ECO volume C 2006 Edition gives the Berliner Gambit as "with compensation". Houdini and I agree that it is +-.

Here I require Houdini to play this inferior opening and I scored two pretty wins---one that Houdini played the most tenacious defence and one where Dr. Hans Berliner, World Correspondence Chess Champion recommended an even more inferior  continutation that  loses four separate ways. I have investigated two of them achieving winning positions by move 15.

1.e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. Ng5 d5 5. exd5 b5 6. Bf1 Nd4 7. c3 Nxd5 8.
Ne4 Qh4 9. Ng3 Bg4 10. f3 e4 11. cxd4 Bd6 12. Qe2 O-O 13. fxg4 Bxg3+ 14. Kd1 c6? (Houdini tries to save material instead of attack) 15. Nc3 Rfe8? Houdini should not allow me to simplify) 16. Nxd5 cxd5 17. b3 Bf2 18. Bb2 b4 19. Rc1 e3? (Bad idea to allow me to shut down the tactics) 20. d3 Rac8 21. g3 (At this point I realized that I am winning the d5 pawn)  Qf6 22. Bg2 Qd6 23. Rc2 h6 24. Kc1 Red8 25. Bf3 Rxc2+ 26. Kxc2 Qf6 27. Kb1 g6 28. Rc1 a6 29. Rc5 Qe6 30. Bc1 Kg7 31. Kc2 Bg1 32. Kd1 Rd7 33. Qg2 Bf2 34. Bb2 a5 35. Ke2 Re7 36. h3 Qd6 37. Bxd5 Bxg3 38. Bc4 Kh7 39. Rxa5 Qc7 40. Re5 * 1-0

Dr. Hans Berliner, World Correspondence Chess Champion according to Houdini and me, recommended a continuation that loses four separate and distinct ways!

1.e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. Ng5 d5 5. exd5 Nd4 6. c3 b5 7. Bf1 Nxd5 8.
Ne4 Qh4 9. Ng3 Bg4 10. f3 e4 11. cxd4 Bd6 12. Qe2 Be6? (Now White wins with 13.Nc3!, the move chosen here; also winning is 13.Qf2, 13.Qxb5ch and 13.fxe4) 13. Nc3 Nxc3 14. dxc3 Bxg3+ 15. hxg3 Qxh1 16. Qxb5+ Bd7 17. Qe5+ Kf8 18. fxe4 Rc8 19. Kf2 Kg8 20. Bf4 Bg4 21. Re1 Qh5? (A weak move against a human; without Queens there are no  complications) 22. Qxh5 Bxh5 23. Ba6 Rb8 24. b3 Rb6 25. Bc4 Bg4 26. Bxc7 Rc6 27. Bf4 h6 28. Bd2 Rf6+ 29. Ke3 Rg6 30. Rc1 Be6 31. d5 Bc8 32. Be1 Rg5 33. Be2 Kh7 34. Kd4 Re8 35. c4 Rge5 36. Bd3 f5 37. c5 a6 38. c6 fxe4 39. Be2 Rf5 40. b4 Rd8 41. Rc5 Be6 42. Bc4 Bc8 43. a4 Rh5 44. b5 axb5 45. axb5 Rh1 46. Ba5 Rf8 47.b6 Ra1 48. b7 * 1-0

There are special championships where engines play each other. There you can see Houdini getting beaten from time to time

Hi WGM Pogonina! If Houdini can be beaten in an opening it chooses for itself I predict that it will take out to move 50 for an evaluation of +- to  occur and another 50 moves to checkmate the  computer.

You're refuting your own point.  If you're forcing it to "follow human theory" (your words above) instead of selecting its own lines, you're not really beating Houdini, are you?

As I said in my book Magic the way to beat supercomputers is to think of the programming as a sieve. When the  computers were first built, "holes" in the programming were so large that a boulder could flow through the hole. As a big hole was plugged ten smaller holes appear. As you plug each of the smaller holes, 10 smaller holes appear and so on.

If you allow computers to pick their own openings they will pick inferior strategic openings. Then they will slowly lose ground. A tiny drop of water will flow through several of the smaller holes, they will combine  to make a larger drop, the drop becomes a trickle, the trickle, a stream, a stream, a river and a river a raging torrent.

That "raging torrent" is the loss of a pawn or the exchange for no compensation. Instead of being dead lost by move 15, it might mean it takes out to move 50 to win a pawn, but once you allow computers to  pick their own openings they will routinely make subtle positional errors that GM's can exploit. As a Class player, I am merely punishing faulty tactics.

In this game I require Houdini to play book in the Two Knights' Defense:

1. e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. Ng5 Bc5 5. Bxf7+ Ke7 6. Bb3 Qe8 7. d3 d6 8.Be3 Bg4 9. Nf3 Qh5 10. Nbd2 Bxe3 11. fxe3 Bd7 12. Qe2 Rhb8 13. O-O-O b5 14. h3a5 15. g4 Qe8 16. g5 Nh5 17. Qg2 a4 18. Bd5 a3 19. Rhf1 Kd8 20. Nh4 Nb4 21. Bf7 Qe7 22. bxa3 Rxa3 23. Nb1 Rxa2 24. Bxa2 Nxa2+ 25. Kd2 g6 26. Nc3 Nxc3 27. Kxc3 b4+ 28. Kd2 Ng7 29. Rb1 Ne6 30. Nf3 Nc5 31. Kc1 Be6 32. Nd2 Ba2 33. Ra1 b3 34.Kd1 bxc2+ 35. Kxc2 Be6 36. Qf3 Nd7 37. h4 d5 38. exd5 Qc5+ 39. Kd1 Rb2 40. Rf2 Bxd5 41. e4 Be6 42. Ke1 Qc3 43. d4 exd4 44. Qxc3 dxc3 45. Nb1 c2 46. Na3 Ne5 47. Rd2+ Kc8 48. Rxc2 Rb3 49. Nb1 Re3+ 50. Kf2 Ng4+ 51. Kg2 Rxe4 52. Ra7 Kb8 53. Rcxc7 Rb4 54. Nd2 h5 55. Kf1 Ne3+ * =

I recall going over a game you presented as you defeating Houdini and your moves matched Houdini's top choice for your side all the way until there was no hope for Houdini's side and after there was no hope for Houdini's side then suddendly your moves occasionally matched it's second, third, or forth choice (but most often it's top chose still). It appeared that you were using Houdini to assist you in your defeating of Houdini. When you are waiting for Houdini to make it's move are you not watching the engine analysis on screen and then using that to aid you in your move selection?

I never use Houdini's choices for guidance, only confirmation. If Houdini chooses a better move than I would make, I resign the game and start over. Most of my wins are simply memorized move sequences (not unlike what other players do when they prepare for tournaments) until I get to something new where I am on my own. Typically I am just memorizing a long  list of choices, but Houdini often screws me up by choosing something different.

A good illustration of the kind of way Houdini would screw me up is in the Fried Liver: 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 d5 5.exd5 Nxd5 6.Nxf7 Kxf7 7.Qf3ch Ke6 8.Nc3 Ncb4 9.Bb3 c6 10.a3 Qf6?! Houdini would avoid the obvious Nba6, give back the piece and enter a drawish ending. According to my "rules" I couldn't force the  computer to play Nac6 so I had to abandon the Fried Liver analysis until the computer got over its infatuation with 10...Qf6! 

When I beat Houdini on the Black side of the Wilkes-Barre, Houdini would simply play the same move sequence time after time, so I would extend the analysis a few moves each game. The final game  represents the culmination of about 4-5 previous "tries".

If I had to quantify my results over the past three weeks I am about +4 = 15 -60 in the Two Knights' Defense and far worse in openings other than the Two Knights' Defense. All of these games were played at 120/40  60/20 30. However, the vast majority of games I resign are evaluated as equal; to save time I don't try to hold equal positions.

The best game I ever played without "starting over" was in the Wilkes-Barre where I went from an advantage of +/= to +- over the span of 20 moves without any computer assistance. What you will see there is that my choices are radically different from what the computer would play. In fact the "winning" move c3 was roundly criticized by Houdini!

Thanks for the question!

They are tests of generally accepted variations presented in ECO, BCO 2, MCO 13, 14, 15, and NCO. If you don't consider these games real, then any theme tournament you would regard as not being "real" games. These games are, essentially, a theme showcasing the 4.Ng5 variation of  the Two Knights' Defense. But from now I'll let Houdini choose openings at random.

I lost four games in a row when I allowed Houdini to pick its openings but just missed a win in the fifth game; one careless move allowed Houdini to draw.