click for larger view

Komodo-10-64bit:

+0.30/40 22.Nac4 Kg8 23.Qa3 f6 24.Nf3 Nd3 25.Rfc1 Nxc1 26.Rxc1 Nb4 27.Ne1 g6 28.Nxc2 Nxc2 29.Rxc2 Kg7 30.g3 Rhd8 31.h3 Rd7 32.Ra2 Ra8 33.Nb2 Qc1+ 34.Kh2 Rb7 35.Nd3 Qxa3 36.Rxa3 Rb5 37.Nc5 Kf7 38.Kg2 Ke7 39.Ra4 Rb4 40.Ra1 Kd6 41.Kf3 Kd5 42.Kf4 h6 43.Ra2 Rbb8 44.h4 Rb4 45.g4 g5+ 46.hxg5 fxg5+ 47.Kg3 Rbb8

Thanks for the link, seen it before, still as unimpressed now as I was then. I see they made it GOTD. What's next? 'Sportsman of the Year.'

Here:

Substitute 'song. for 'game.'

Even the most primitive computer possesses knowledge. Actually anything connected with some kind of explicit or implicit database does.

Just as easy is the answer whether a computer is conscious - a clear "no".

With intelligence, there are debates. I'm on the "yes" side, because most of the "no" arguments IMO mix it up with consciousness. Intelligence is the ability to perform some tasks for which we humans require thinking. It is not necessarily being aware of that ability (after all, already the word "aware" suggests consciousness). In this sense, computers are intelligent, though their intelligence is specific, not general (IBM's Watson may be the first step towards general intelligence).

The pyramid is this:

Knowledge < specific intelligence < general intelligence < consciousness.

Each "bigger" part cannot exist if the smaller part is absent. Chess computers are on the specific intelligence level. Modern computer technology generally somewhere in between specific and general intelligence.

Anyway, humans have access to a really pretty good assortment of tools to operate.

Maybe you should keep your shoulds to yourself.

Weird expectations for a public forum.

<And to compare consciousness to flying to prove a point is quite an obtuse idea. Perhaps you should choose another>

Komodo also gained 2 rating points for its effort...

In competitions like TCEC's it seems like the machines use book for the first half dozen or so moves. Does the computer pick which book opening is played? If not, how is it decided?

Given the advance in machine learning, I'd guess that a high powered computer with an adequate database could narrow down the entire chess opening spectrum and distill the best array of lines that could gain an advantage for white and conversely always or just about equalize for black. This computer program would not be interested in anything else but finding those best "sure fire" openings. With criteria, say, position after move 16. or so.

Has anyone heard of any development efforts for such a program?

In my opinion, it is almost impossible, because the number of possible openings is almost unlimited. For machine learning, we need to know the best moves so that machine can learn them, which is almost impossible also.

I ran Komodo 12.3 MMax and Komodo 12.3 MCTS concurrently last night, each with Hash Table = 512, Threads = 2. As has happened before, Komodo 12.3 MCTS just stopped when it reached d=34 as a result of a wrong setting of the MCTS Hash parameter. Komodo MCTS uses both a Hash Table (to keep track of move transpositions) and an MCTS Hash Table to keep the MCTS search tree in memory. And, when the MCTS Hash Table becomes full, Komodo MCTS just stops. I thought I had this fixed earlier but somehow I must have changed it without realizing it (it's called old age). So I'll rerun the analysis of the original position tonight.

In the meantime, since speed is not really the object of the comparison, I had run 2 analyses with Komodo 12.3 MMax, one last night and one two days before from the original position:

click for larger view

The first analysis, as I previously mentioned, came up with an evaluation of [+0.96] from d=11 through d=54 for its "top" 3 moves, 1.Kd3, 1.Ba8, and 1.Bd5; an evaluation of [+0.92] from d=55 through d=80 for the same "top" 3 moves, and an evaluation of [+0.71] from d=81 through d=83 when I stopped the analysis.

The second analysis, which I ran last night, game similar results; an evaluation of [+0.91] from d=12 through d=47 for the same 3 moves, an evaluation of [+0.93] from d=48 through d=54 for the same 3 moves, an evaluation of [+0.90] from d=55 through d=74 for the same 3 moves, and an evaluation of [+0.67] for the same 3 moves from d=76 through d=79 when I stopped the analysis. Not much difference between the two analyses but some. An example of the "Komodo Randomizer" due to the non-determinism of multi-core engines, even though Threads was only set = 2.

For the first analysis with Komodo 12.3 MCTS last night the top 3 moves were 1.Kd3, 1.Ba8, and 1.Kc3 and the evaluations varied from [+0.82] to [+0.98] from d=14 through d=34 when the MCTS Hash Table became full and Komodo 12.3 MCTS stopped. There was often a 1 – 3 centipawn variation in the evaluation of the 3 moves to their ranking would often change. But remember that Komodo 12.3 MCTS "estimates" the evaluation in centipawns according to the win (or scoring) probability for each move so the difference could just be due to rounding errors. Another example of the "Komodo Randomizer" which this time affects the move rankings. But it's not significant at any rate.

For the second analysis with Komodo 12.3 MCTS this morning the same top 3 moves were indicated, and the evaluations varied from [+0.88] to [+0.91] from d=15 through d=34 when the MCTS Hash Table became full (I had modified the wrong parameter) and Komodo 12.3 MCTS again stopped. There was a 1 – 6 centipawn variation, usually only 1 – 2 centipawns, in the evaluations.

So it doesn't seem that using MCTS, at least as far as Komodo 12.3 MCTS is concerned, will cause the evaluations to be [0.00] or near to that no matter the search depth. And that, after all, was your question. I'm checking now to see whether my modification to the MCTS Hash Table size will allow it to exceed d=34 just in case it changes the conclusion.

A common enhancement to all 3 versions is the ability to play Sudden Death/Armageddon games (about time!). I find the following quote from the Komodo 14 page (https://komodochess.com/Komodo14.htm) very interesting:

"There is one new feature added since Komodo 13.3, "Armageddon" mode, which tells Komodo that White (or Black) must win, draws are scored as losses for that color. When draws are counted as losses for White, knowing this improves Komodo's performance as White by about 30 elo according to our tests. In Armageddon mode, the specified color (usually White) not only avoids clear draws, but adjusts Contempt by an amount we determined was most helpful for Armageddon play. Testing on one thread in Armageddon mode suggests that on a modern laptop, if White gets one minute total plus 1% increment, and Black gets half a minute plus 1% increment plus draw odds, then chances are fairly close to balanced."

I was surprised to see that they found out that, if White knows that draws are counted as losses, then Komodo's performance improves by about 30 Elo rating points. Also, if using only one thread (core) if White is given 1 minute + 0.6 (+1%) sec increment and Black gets 30 secs + (presumably) a 0.3 (+1%) sec increment plus draw odds, then each side's chances are fairly close to balanced. This 2:1 (2.00) White Time Control Ratio (WTCR) is much greater than the usual 5:4 (1.25) WCTR, the 10:7 (1.43) WCTR used in the last two Norway tournaments for their Sudden Death/Armageddon game tiebreaks, or even the theoretical 9:5.5 as suggested in https://www.researchgate.net/public.... Another indication that if a fair set of time controls is desired (i.e. a 50/50 chance of either player winning the game when Black has draw odds), the current WCTRs used in today's Sudden Death/Armageddon games must be increased.

Chess.com in their announcement of the release of Komodo Dragon (https://www.chess.com/news/view/kom... ) claims that Komodo Dragon beat Stockfish 12 in two testing matches. They also estimate that Komodo Dragon is about 200 Elo rating points higher than Komodo 14 when using 1 core and 170 Elo rating points higher when using 4 cores. When using 4 cores that would make it rated at 3589 compared with Stockfish 12 at 3516 in the latest (11-07-20) CCRL engine vs. engine tournament at Rapid (40/15) time controls but only 3647 compared with Stockfish 12's 3644 when using only 1 core at Blitz (40/02) time controls. It will be interesting to see how Komodo Dragon performs in the CCRL and CEGT engine vs. engine tournaments and (hopefully) TCEC season 20. May the best NNUE engine win.

1. c4 c5 2. Nc3 b6 3. Nf3 Bb7 4. d4 cxd4 5. Nxd4 Nf6 6. f3 a6 7. e4 d6 8. Be3 e6 9. g4 Nc6 10. h4 Nd7 11. Qd2 Be7 12. O-O-O h5 13. g5 O-O 14. Bh3 Nxd4 15. Bxd4 Qc7 16. g6 f5 17. Kb1 Qxc4 18. Rc1 Qb4 19. Qd1 Bf6 20. Bxf6 Nxf6 21. a3 Qa5 22. Qxd6 Kh8

click for larger view

23. Qc7 Rab8 24. Rhe1 Qc5 25. Qg3 Qd4 26. exf5 Qd3+ 27. Ka1 exf5 28. Qg5 Rbe8 29. Bxf5 Qxf3 30. Be6 Qg2 31. Qxg2 Bxg2 32. Bf7

click for larger view

32...Rb8 33. Ne2 Be4 34. Nf4 Bf3 35. Ne6 Rfc8 36. Rxc8+ Rxc8 37. Ng5 Bc6 38. Re7 Bd5 39. b4 Rb8 40. Rc7 Bg2 41. Kb2 Nd5 42. Rd7 Nf6 43. Rd6 Ng4 44. a4 Nf6 45. Kc3 Bf1 46. Kd4 a5 47. b5 Bg2 48. Ke5 Ng4+ 49. Ke6 Ra8

click for larger view

50. Ke7 Ra7+ 51. Kf8 Nh6 52. Rd8 Ng4 53. Be6 Ne5 54. Nf7+ Nxf7 55. gxf7 Kh7 56. Ke8 Rxf7 57. Kxf7 Be4 58. Rg8 g6 59. Kf6 Bd3 60. Ra8 Bf5 61. Ra7+ Kh8 62. Bxf5 gxf5 63. Kg6 Kg8 64. Ra8# 1-0