Man vs Machine - Part III, The Better Bots and Computer Programs
Lc0 design, from Chess.Com web-site.

Man vs Machine - Part III, The Better Bots and Computer Programs

Avatar of pawn314159
| 7

Great Play from the Better Computer Programs ---

Part I of this series noted the initial days of the new computer bots, and focused on the programs at beginner levels of play. The Part II post detailed the mid-level computer programs on this web-site. This third post in the Man vs Machine series provides information on the playing strengths of the computer bots and Komodo levels > 1700 in ratings, as well as Komodo Boss and Stockfish. If I have actually played the particular program, info detailing the playing strength of the computer bot will be provided. Otherwise, basic info will be given on what is known about each bot.

On an interesting note, two new bots made their appearance in Dec. 2019, a month after the release of the initial 60 bots. These two are named and patterned after real people. Hikaru-Bot is based on the play of Hikaru Nakamura, a five time US champion. Danny-Bot takes after Danny Rensch, an IM and noted chess commentator. During initial release of the bots, staff indicated there would be a total of 60 bots "for now", but more would come later. If this any indication of what is planned for further release, it is a very exciting development!

Then on Dec. 15, 2019, Lc0 showed up in the live play area. This is astonishing news, as neural programming has not been commonly available to many non-professional players. With its appearance at Chess.com, any member can play whenever it is on-line. 

There are 22 bots and 7 Komodo levels (14-20) which are usually between 1700 and 2000 in rapid ratings. 4 new K levels (Komodo 20-24) started in the CM range in Nov. 2019, but are now slightly under that, between 1900-2000. Stockfish and Komodo Boss, are definitely in the CM to Master level of play. Lc0 had an initial rating of 2000, but has dropped to 1700 and below, possibly due to an early learning phase. With its self-taught neural abilities, it is also possible that this program will progress up the ratings as it gains playing experience. This makes a total of 36 computer programs in the better bot category.

Across all levels of play from beginner to Master levels, the live play area of Chess.com web-site current;y contains 62 bots, 24 Komodo levels, Stockfish, KomodoBoss, and Lc0. There are also 10 practice computer levels, which are unrated and un-timed, but are great for training purposes as they contain take-back and hint functions. All this adds up to 99 computer programs and levels available at Chess.com. This is a extraordinarily impressive array of chess-playing computers, likely the most complete grouping of bots commercially available anywhere in the world. So take your pick and play!

Completed Games with Higher-Rated Programs ---

In playing the best of the bots, I braced myself for withering attacks and humiliating defeats. While I did lose to some of the higher-levels, I was surprised I could at least play a good game. So long as I took my time, played positionally, and watched for tactics, I was competitive through much of the battle. The following games are illustrative of the better programs.

Lc0. This particular version of neural programming may be in its initial learning stages. In the below rapid game, the computer perfectly played the Sicilian, Najdorf variation, English Attack pattern for the first 12 moves. It then varied on 13, and lost its Queen a move later. I was shocked that a program designed for an intuitive grasp of the game would make such a freshman blunder. It then continued to lose pieces, even with post-game analysis believing Lc0 was making the best moves!  When faced with overwhelming odds, Komodo will deliberately shed pieces as a last-ditch effort at stalemate,, and Lc0 is no different in this regard. In several months from now, the program may have learned so much that it will become unbeatable. For the time being, however, the version of Lc0 on the Chess.com web-site is in the moderate range of play. 

Stockfish. This is likely the best of the programs available at the Chess.com web-site, and definitely plays at CM to Master-levels. In a rapid game with it, Stockfish had a nearly perfect accuracy rate of 99.1% and best moves of 75.4%. I pushed myself to play as well as I could, and the results were astonishing. Yes, I lost, but the program was not able to unleash tactical brilliancies. Instead, I was even on material through move 46, only gradually losing ground to an advancing pawn on the edge of the board which was difficult to stop. Then, Stockfish finally converted its positional advantage into the gain of a single pawn in the center. The computer is so strong, that's all it needed for a win, and I resigned on move 61. I was happy I did so well, having 97% accuracy and 63.9% best moves. Not too shabby against the best of the best!

Game Statistics ---

I have played several of the better bots recently, extending RandomJeff's challenge concept into a 60 Day Challenge. 78 games and 56 separate bots, programs, and computer levels have been played over the last two months. The statistics are as follows ---

  • 78 games in 62 days
  • 71-6-1 (29-1-1 blitz; 35-4 rated rapid; 6-1 unrated rapid; 1-0 daily)
  • 73 computer games, 5 human games
  • 56 separate computer programs and levels – 37 bots; 12 Komodo levels (K 1-12); 1 Stockfish game; 1 Lc0 game; 5 levels, practice computer (1-5)
  • 31 blitz games – 23 bots, 8 Komodo levels (K 1-3; K 7-11)
  • 46 rapid games – 31 bots (1 bot 3 times), 4 games w 3 K levels (K10, 11; K12 two times); 1 open challenge w a person; 2 SCL games; 1 practice game w a HS chess club; 1 unrated game w Stockfish; 1 rated game w Lc0; 5 games w on-line practice computer (level 1-5).
  • 1 finish of a daily w a person

The Shortlist ---

Basic data and information on the better-rated programs follows.  If I have played the program, stats on accuracy, best move, mistakes, and blunders are provided. Otherwise, info on win rates and recent ratings levels are included.

Komodo14 ---

  • 50,574 blitz games; 84.0% win rate; recent blitz rating, 1879.
  • 19,896 rapid games; 79.1% win rate; recent rapid rating, 1707.

Julia-Bot ---

  • Bio says she likes the Dutch.
  • 3,553 blitz games (as of 12-8-19); win rate, 71.8%; recent blitz rating, 1819.
  • 2,193 rapid games; win rate, 65.5%; recent rapid rating, 1721.

Wayne-Bot ---

  • Bio says he doesn’t mind playing a long endgame.
  • 1,716 blitz games (as of 12-18-19); win rate, 78.0%; recent blitz rating, 1890.
  • 622 rapid games; win rate, 70.2%; recent rapid rating, 1736.

Olga-Bot ---

  • Bio says her openings are unusual, but she knows her tactics and won't lose without a fight.
  • 3,402 blitz games (as of 12-19-19); win rate, 72.9%; recent blitz rating, 2014.
  • 1,097 rapid games; 69.0% win rate; recent rapid rating, 1746.

Xavier-Bot ---

  • Bio says he tries to sacrifice and checkmate as fast as he can.
  • 1,974 blitz games (as of 12-18-19); win rate, 79.4%; recent blitz rating, 1934.
  • 798 rapid games, 72.8% win rate; recent rapid rating, 1761.

Miguel-Bot ---

  • Bio says he studies tactics, endgames and game collections.
  • 1,266 blitz games (as of 12-19-19); win rate, 82.8%; recent blitz rating, 2028.
  • 788 rapid games; 76.1% win rate; recent rapid rating, 1775.

Komodo15 ---

  • 37,470 blitz games (as of 11-27-19); 86.2% win rate; recent blitz rating, 1941.
  • 13,475 rapid games; 82.6% win rate; recent rapid rating, 1793.

Fatima-Bot ---

  • Bio says she is Calm, patient, and determined to queen in the endgame.
  • 2,834 blitz games (as of 12-19-19); 85.1% win rate; recent blitz rating, 2053.
  • 1,410 rapid games; win rate, 76.2%; recent rapid rating, 1839.

Charles-Bot ---

  • Bio says he doesn't have time to study, but he hasn't forgotten how to play tactics.
  • Lasso_of_Chess played Charles-Bot Nov. 1, 2019; says this is a high-rated he sometimes offers a draw when you are losing. 
  • 2,389 blitz games (as of 12-19-19); 86.4% win rate; recent blitz rating, 1973.
  • 1,366 rapid games; 79.5% win rate; recent rapid rating, 1841.

Oscar-Bot ---

  • Bio says he defends carefully and rarely loses.
  • 404 blitz games (as of 12-19-19); 93.5% win rate; recent blitz rating, 2138.
  • 364 rapid games; 75.8% win rate; recent rapid rating, 1869.

Li-Bot ---

  • Bio says she loves to win short, attacking games.
  • 1,442 blitz games (as of 12-19-19); 87.9% win rate; recent blitz rating, 2051.
  • 1,053 rapid games; 75.3% win rate; recent rapid rating, 1892.

Manuel-Bot ---

  • Bio says he enjoys studying and reviewing the classic games and openings of the past.
  • 903 blitz games (as of 12-18-19); win rate, 82.0%; recent blitz rating, 2192.
  • 484 rapid games; 76.6% win rate; recent rapid rating, 1919.

Komodo16 ---

  • 30,127 blitz games (as of 12-19-19); 88.6% win rate; recent blitz rating, 2026.
  • 11,945 rapid games; 84.3% win rate; recent rapid rating, 1929.

Komodo17 ---

  • 23,199 blitz games (as of 12-19-19); 90.9% win rate; recent blitz rating, 2121.
  • 9,650 rapid games; 87.3% win rate; recent rapid rating, 1901.

Francis-Bot ---

  • Bio says he is a vicious attacker and will checkmate you the first chance he gets.
  • 1,140 blitz games (as of 12-18-19); win rate, 93.8%; recent blitz rating, 2136.
  • 396 rapid games; win rate, 79.0% win rate; recent rapid rating, 1922.

Komodo18 ---

  • 19,920 blitz games (as of 12-19-19); 92.0% win rate; recent blitz rating, 2190.
  • 9,588 rapid games; 90.0% win rate; recent rapid rating, 1939.

Komodo19 ---

  • 059 blitz games (as of 12-20-19); 92.9% win rate; recent blitz rating, 2169.
  • 10,244 rapid games; 89.9% win rate; recent rapid rating, 1954.

Sofia-Bot ---

  • Bio says she defends like a pro.
  • 725 blitz games (as of 12-18-19); win rate, 91.1%; recent blitz rating, 2191.
  • 615 rapid games; win rate, 79.6%; recent rapid rating, 1955.

Komodo20 ---

  • 48,102 blitz games (as of 12-20-19); 97.3% win rate; recent blitz rating, 2197.
  • 13,922 rapid games; win rate, 94.4%; recent blitz rating, 1964.

Hikaru-Bot ---

  • Based on Hikaru Nakamura, a five time US champion and GM who is highly rated in blitz and classical forms of chess.
  • 206 blitz games (as of 12-13-19); 96.1% win rate; recent blitz rating, 2266.
  • 79 rapid games; 91.1% win rate; recent rapid rating, 1967.

Luke-Bot ---

  • Bio says to expect a slow game which is squeezed by Luke.
  • 342 blitz games (as of 12-14-19); win rate, 96.1%; recent blitz rating, 2134.
  • 275 rapid games; win rate, 88.7%; recent rapid rating, 1974.

Komodo21 ---

  • 608 blitz games (as of 12-20-19); win rate, 95.5%; recent blitz rating, 2171.
  • 271 rapid games; win rate, 91.5%; recent rapid rating, 1975.

Arjun-Bot ---

  • Bio says his first moves will take you out of the comfort zone, followed by accurate play.
  • 1,623 blitz games (as of 12-18-19); win rate, 95.0%; recent blitz rating, 2132.
  • 775 rapid games; win rate, 95.1%; recent rapid rating, 1978.

Sakura-Bot ---

  • Bio says she plays chess like science, and in endgames, will pick apart opponents.
  • 728 blitz games (as of 12-18-19); 93.6% win rate; recent blitz rating, 2138.
  • 366 rapid games; 84.4% win rate; recent rapid rating, 1985.

Komodo22 ---

  • 604 blitz games (as of 12-20-19); 93.3% win rate; recent blitz rating, 2182.
  • 311 rapid games; 91.3% win rate; recent rapid rating, 1986.

Komodo23 ---

  • 799 blitz games (as of 12-20-19); win rate, 94.1%; recent blitz rating, 2227.
  • 432 games; win rate, 90.2%; recent rapid rating, 1985.

Alexander-Bot ---

  • Bio says he is all strengths and no weaknesses.
  • 2,485 blitz games (as of 12-18-19); win rate, 97.3%; recent blitz rating, 2142.
  • 980 rapid games; win rate, 93.8%; recent rapid rating, 1986.

Ahmed-Bot ---

  • Bio says he loves to play the Dutch and attack with as many pieces as he can.
  • 3,900 blitz games (as of 12-18-19); win rate, 97.0%; recent blitz rating, 2181.
  • 1,284 rapid games; win rate, 88.1%; recent rapid rating, 1988.

Noam-Bot ---

  • Bio says he plays very solidly, so be prepared for a long fight.
  • 702 blitz games (as of 11-18-19); 91.5% win rate; recent blitz rating, 2199.
  • 444 rapid games; 80.8% win rate; recent rapid rating, 1991.

Wei-Bot ---

  • Bio says she is looking for a knock out from move one. B
  • 1382 games (as of 12-18-19); win rate, 94.7%; recent blitz rating, 2205.
  • 378 rapid games; win rate, 85.4%; recent rapid rating, 1992.

Danny-Bot ---

  • Based on Danny Rensch, International Master and chess commentator.
  • 255 blitz games (as of 12-13-19); 90.5% win rate; recent blitz rating, 2240.
  • 62 rapid games; 85.4% win rate; recent rapid rating, 1993.

Komodo24 ---

  • 1705 blitz games (as of 12-10-19); 97.0% win rate; recent blitz rating, 2181.
  • 328 rapid games; 89.0% win rate; recent rapid rating, 1907.

Lc0 ---

  • Neural programming, designed for self-teaching and progressive learning abilities.
  • First games were in December, 2019; was assigned an initial rating of 2000, but has fallen in both blitz and rapid ratings.
  • In rapid game, Lc0 played Sicilian, Najdorf for 12 moves before blundering away its Q, then playing miserably. Its accuracy was 43.5%, best moves, 25%; 2 mistakes, 1 blunder (and really, several more, as it deliberately lost pieces to attempt a stalemate).
  • 923 blitz games (as of 12-22-19); win rate, 86.9%; recent blitz rating, 1835.
  • 264 rapid games; win rate, 65.5%; recent rapid rating, as low as 1593.

KomodoBoss ---

  • 95,538 blitz games (as of 12-18-19); 97.4% win rate; recent blitz rating, 2413.
  • 19,578 rapid games; 95.8% win rate; recent rapid rating, 2191.

Stockfish ---

  • Probably the strongest AI program at the Chess.com web-site; It also plays slower and more realistically than lower-level programs.
  • Rapid game; Accuracy was nearly perfect, at 99.1%, best moves 75.4%; 0 mistakes, 0 blunders. It simply out played me, patiently waiting 46 moves before going up a single pawn.
  • 152,828 blitz games (as of 12-25-19); win rate, 97.2%; recent blitz rating, 2425.
  • 45,526 rapid games; win rate, 96.8%; recent rapid rating, 2211.

More to Come ---

These posts have turned into a series on computer play. The next and last installment, Part IV, will  summarize my thoughts on the pros and cons of live play with computers.

So, say tuned for Part IV of Man vs Machine!