Man vs Machine - Part II, The Mid-Level Computer Programs

Playing the Computer Programs ---

The new computer bots under 1000 rapid rating were detailed in Part I of this series, with every bot and Komodo level in the lower ranges being played at least once. I tried playing one or more bots a day, in the style of the 30 Day Challenge of RandomJeff. This produced a wealth of data and information on the new bots making an appearance since early November, 2019. In that time, 20 bots below 1000 were played as well as Komodo levels 1 through 6, for a total of 26 computer programs. A shortlist of all bots and K levels in the beginning range were included in Part I.

Completed Games with the Mid-Range Programs ---

This post focuses on the computer bots and Komodo levels in the moderate ratings range of 1000 to 1700. Many bots have their own playing styles, with some very aggressive, while others are programmed for gambits or particular openings. Some were balanced in style, and a few played at almost an expert level during the opening, but then weakened considerably by the end game. There are 20 bots and seven Komodo levels (7-13) in the mid-playing range, for a total of 28 programs. The following game is illustrative of a mid-range bot’s abilities.

Wendy-Bot. This bot's cartoon-like depiction is featured in the main visual picture of this post. I played Wendy in both rapid and blitz. The rapid game is provided below, and is typical of the mid-level bots. I held the initiative from the start. She began losing pawns on move 16, and by the end game, I held an overwhelming advantage. Wendy then sacked her last minor piece for a pawn. While a blunder, perhaps that is what the bio meant by “throwing pieces in the fire of an attack.” Mate was with a R and promoted Q.

Komodo. In addition to playing the bots, several Komodo levels were played in both rapid and blitz. Since this program has been so extensively used as a AI opponent, a short summary is appropriate as a baseline for all computer activity.

Komodo went on-line Feb. 1, 2018, and many K levels have played hundreds of thousands of games in bullet, blitz, rapid, and several variants. After a request was made by several members, an adjustment was made to Komodo in July, 2019, to provide more variety in its openings. This generated many off-beat openings as well as a tendency for the Queen to come out very early. A drop in ratings occurred in some moderate to low Komodo levels, maybe 100 points or so.  A forum post of mine describes the variations being given to Komodo.

Then in mid-October, many of the moderate levels began losing to much weaker players. A dramatic drop in ratings occurred over a few days, with some levels falling 200 to 400 points. The speculation among the blogs and forums (but unconfirmed by staff) is that programming was again adjusted, this time to reduce the computer levels from being too bunched up in their ratings. Previously, the first 9 levels covered 1600 + points of rapid ratings. The next 11 levels were in a tight range of only 400 points or less, from 1600 + to 2000, with Komodo Boss and Stockfish then being at CM levels. After Mid-October, there is a more even distribution of ratings, with levels 1 through 13 now in the 0 – 1600 + range, while levels 14-20 are typically in the 1650 to 2000 range.

In early Nov. 2019, Komodo levels 21-24 were introduced when the new bots went on-line, with initial ratings of CM to Master levels. Many of the Komodo levels went off line for several days while the bots were introduced, then came back as most of the bots were rotated out of the line-up for a while. That may have been meant for players to try out the new bots, but it took a week or so for Komodo levels 1-20 to stabilize and become fully available again.

The following game was with Komodo 11 in November, 2019, as the bots were going live. Note the dramatically different style of play this level has compared with pre-July and October time frames when it was over 1700 in ratings.  I doubt if I could have beaten this level back then. I would like to think it is from an improvement in my abilities, but more likely, it is from the programming changes in recent months.  

Game Statistics ---

In a span of 39 days beginning Nov. 1, 2019, I have played 46 separate computer programs across 63 rated games and one practice game. The following table has the break-down ...  

• 64 total games in 39 days, 60-3-1 (29-1-1 blitz; 30-2 rapid; 1-0 match)
  • 60 computer games, 4 human games ---
  • 46 separate computer programs – 34 bots; 12 Komodo levels (K 1-12); 
  • 31 blitz games – 23 bots, 8 Komodo levels (K 1-3; K 7-11)
  • 32 rapid games – 26 bots, 3 K levels (K10-12); 1 open challenge w a person; 1 SCL game; 1 practice game w HS chess club member.
  • 1 finish of a daily w a person

The Shortlist ---

The following computer programs range between 1000 and 1700 in playing strength, rapid rating. I have played most of them in both rapid and blitz games. Stats and references to the bio are taken from the Chess.com web-site. Commentary is from my games as well as other people who provided their thoughts to me. 

Komodo7 ---

• Previously played 16 rapid games with level 7 when it was rated much higher;
• Had a record against level 7 of 12-3-1; computer win rate in this match was 18.7%.
• When level 7 lost (12 games), its accuracy was between 22% and 86%, averaging 59.4%.
• When level 7 drew (1 game), accuracy was 47.3% (I let it stalemate by mistake)
• When level 7 won (3 games), its accuracy was between 50.7% and 85.5%, averaging 69.6%.
• Played a blitz game recently; strong K side attack against K7 led to lasting material and positional adv., then mate; it’s accuracy, 58.1%, best moves 45.3%, 4 mistakes, 2 blunders.
• 412,970 blitz games (as of 11-29-19); win rate, 59.1%; recent blitz rating, 1032.
• 181,373 rapid games (as of 11-29-19); 55.1% win rate; recent rapid rating, 1031.

 Elena-Bot ---

• Played blitz game; she played the first three moves of the Sicilian, then made a series of weak moves, then many missteps, letting me gain material nine times over the game.
• Her accuracy was 48.8%, best move 38.6%; 2 mistakes, 3 blunders
• Also played blitz; she was aggressive in the opening, but made several mistakes and blunders; her accuracy was 61.8%, best moves 41.9%; 2 mistakes, 2 blunders.
• 3,378 blitz games (as of 11-20-19); win rate 47.9%; recent blitz rating, 1065;
• 1,763 rapid games (11-20-19); win rate 48.4%; recent rapid rating, 1036.

 Laura-Bot ---

• The bio said to expect aggressive tactics. I did not find that to be the case, holding the initiative the entire game, then launching a K side attacked which mated her.
• Her accuracy was 30.5%, best move 33.3%; 3 mistakes, 2 blunders.
• 196 rapid games (11-19-19); win rate, 45.4%; recent rapid rating, 1040.
• In the blitz game, she played book for 6 moves, then blundered repeatedly; no aggressive tactics noted in this game, either.
• Her accuracy was 25.0%, best move 27.3%; 2 mistakes, 1 blunder (I felt blundered 6 times).
• 1,316 blitz games (11-23-19); win rate, 47.1%; recent blitz rating, 1007.

 Maria-Bot ---

• In rapid game, she made 6 weak moves, then 2 blunders, losing a minor and rook. End game was an easy win. Bio says she loves to sac and attack. In my game, she just lost pieces.
• Her accuracy was 67.35%, best move 42.9%; 1 mistake, 5 blunders.
• Also saw no attacking strength in the blitz game; accuracy was 60.9%, best move 46.4%; 5-6 weak moves, 1 mistake, 3 blunders.
• FonsecaSF played 50 + blitz games with Maria, feeling she makes direct attacks but defends poorly; has accuracy of between 40% - 79%; best moves, 49% - 60%.
• 905 Blitz games (as of 11-21-19); 44.6% win rate; recent blitz rating, 1177.
• 317 rapid games; 39.1% win rate; recent rapid rating, 1052.

 Junior-Bot ---

• Rapid game; Bio says he brings out the Q early, which he did; He held the initiative until Q’s were exchanged, then lost 2 pieces and several pawns, for a loss in the end game.
• His accuracy was 69.7%, best moves, 39.4%; 3 mistakes, 5 blunders.
• He brought out the Q in blitz game, too; blunder, with simple grab of the Q by my N.
• His accuracy in blitz was 82.2%, best moves 51.8%; 0 mistakes, 2 blunders (really, more).
• 3,075 blitz games (11-22-19); win rate, 43.6%; recent blitz rating, 1150.
• 1,273 rapid games (11-22-19); win rate, 48.0%; recent rapid rating, 1081.

 Wilson-Bot ---

• This bot played the Dutch in a rapid game; he then castled K side, right into my attack; then made several other mistakes which gave me the advantage throughout the game.
• His accuracy was 58.71%, best moves 44.3%; 2 mistakes, 5 blunders.
• 226 rapid games (as of 11-20-19); win rate, 46.0%; recent rapid rating, 1084.
• Also played a blitz game; He played the Bird’s as white, going down 20 points in material before being mated with a King in the open.
• His accuracy in the blitz was 55.8%, best moves 40.6%; 2 mistakes, 4 blunders.
• 564 blitz games (as of 11-21-19); 49.4% win rate; recent blitz rating, 1167.

 Komodo8 ---

• Previously played 10 games with level 8 when it was rated much higher; rapid rating was then between 1510 and 1631.
• Match play was 8-2; computer win rate was 20.0% (much lower than the below average).
• When level 8 lost (8 games), its accuracy was between 12.9% and 88.2%, averaging 62.7%.
• When level 8 won (2 games), accuracy was 74.9% and 92.5%, averaging 83.7%.
• Also played one blitz game; it quickly lost pieces and position, allowed an easy mate after promotion. Its accuracy was 68.3%, best move 45.9%; 1 mistake; 3 blunders.
• 163,518 rapid games (as of 11-29-19); 61.8% win rate; recent rapid rating, 1052.
• 337,307 blitz games; win rate, 65.8%; recent blitz rating, 1046.

 Jade-Bot ---

• Her bio says she has been taught aggressive openings by her coach.
• In both the rapid and blitz game I played, Jade was indeed was aggressive in the opening. But she so prone to mistakes and blunders after the opening that she was overwhelmed by positional play and simple tactics.
• In the blitz game, her accuracy was 54.1%, best move at 40.4%; 7 mistakes, 2 blunders. Win rate 58.7%; recent blitz rating, 1145.
• In rapid game, her accuracy rate was 30.5%, best move was 31.6%; 6 mistakes, 2 blunders.
• 1,345 rapid games (11-13-19); 50.6% win rate. Recent rapid rating of 1147.

 Vinh-BOT ---

• Bio says the bot likes the Kings Indian; he does, as he played an Indian as both white and black.
• Bot played a rapid game as white, with a King’s Indian Attack; book through move 6, but lost pawns on 13 and 20, then a piece on 21 and not stopping promotion on 28-31.
• His accuracy 67.7%; best move 35.3%; 2 mistakes, 1 blunder (felt like 4 mistakes, 4 blunders)
• Also played a blitz; his accuracy was 37.1%, best move, 35.7%; 4 mistakes, 5 blunders.
• 1,245 blitz games (11-22-19); win rate 45.8%; recent blitz rating, 1169.
• 354 rapid games (11-22-19); win rate, 45.1%; recent rapid rate, 1166.

 Komodo9 ---

• Previously played level 9 in rapid when it was much higher in ratings; won 5-2-1; 62.5% win rate for me (reverse of below stats favorable to Komodo).
• When I won, Komodo’s accuracy rate averaged 70.5%; when drawn, its accuracy was 41.8%; when I lost, its accuracy rate averaged 73.9%.
• More recently, played rapid game with level 9; accuracy was 45.8% when it lost.
• Then played blitz game; It played book in KID for 8 moves before varying, but hen blundered repeatedly; Its accuracy was 77.6%, best moves, 42%; 4 mistakes, 3 blunders.
• 212,819 blitz games (as of 11-30-19); win rate, 67/0%; recent blitz rating, 1211.
• 107,805 rapid games (11-30-19); win rate, 65.1%; recent rapid rating, 1156.

 David-Bot ---

• In rapid, the bot brought out the Q early, made 10 weak moves in 13, then lost its Q in a pin.
• His accuracy was 29.8%, best moves 34.0%; 1 mistake, 2 blunders.
• In blitz, he had an off-beat opening, made several weak moves; then went down a rook +.
• Accuracy in blitz was 78.7%, best move 47.5%; 1 mistake, 1 blunder (I felt 2 + blunders).
• 4,771 blitz games (as of 11-24-19); win rate, 51.4%; recent blitz rating, 1309.
• 1,790 rapid games; 53.7% win rate; recent rapid rating, 1220.

 Ali-Bot ---

• This bot outplayed me through most of a rapid game, going up 2 pawns. It played miserably at the very end, however, allowing me to equalize, promote a pawn, then mate it.
• The bio says the bot is developing a positional style. I agree, but the bot also can attack once it has a favorable position.
• His accuracy was high, at 92.4%, best move 56.6%; but 4 mistakes and 1 blunder lost the game.
• His positional play in a blitz game then produced a win, as he played book for 8 moves, then had a good game until I ran low on time and began making mistakes.
• His accuracy in the blitz was 82.1%, best move 44.4%; 4 mistakes, 1 blunder.
• 7,133 blitz games (11-24-19); 63.3% win rate; recent blitz rating, 1360.
• 2,463 rapid games; 56.8% win rate; recent rapid rating, 1233.

 Mateo-Bot ---

• Played rapid; He lost a piece in the opening, then 2 more pawns; He gradually exchanged pieces, as per the bio; promotion and mate in the end game was easy.
• I also played a blitz game; he made several blunders which lost material and position, then lost in the end game; accuracy 90.8%, best move 52.1%; 1 mistake, 5 blunders.
• FonsecaSF played this bot several times in blitz; 77.3% to 88.7% accuracy; 41.5% to 54.1% best moves; Weak, easy, but without blunders of lower bots; he does trade pieces.
• 2,022 blitz games (11-25-19); 54.9% win rate; recent blitz rating, 1278.
• 897 rapid games, 56.0% win ratio; recent rapid rating, 1282.

 Komodo10 ---

• Recently played a rapid game after the ratings drops in Komodo level. The computer played aggressively; after exchange of pieces, I had three extra pawns, with promotion for mate.
• Its accuracy was 81.1%, best moves 50.9%; 4 mistakes, 2 blunders.
• Also played blitz; The Q came out early, but was countered; w several weak moves, K10 lost 3 pawns, then had a weak end game; Accuracy 59.5%, best move 39.6%; 6 mistakes; 1 blunder.
• 215,348 blitz games (11-22-19); win rate, 73.4%; recent blitz rating, 1320.
• 104,638 rapid games; win rate, 70.3%; recent rapid rating, 1289.

 Pierre-Bot ---

• Rapid game; he played … f5, as the bio suggested; He played book through 7, then lost ground.
• His accuracy was 58.0%, best moves 35.2%; 2 mistakes, 1 blunder (felt like 3 + blunders).
• The bot also played f4 / f5 in 8 blitz games with FonsecaSF; accuracy ranged from 72.8% to 85.2%; best moves from 46.5% to 58.5%.
• I also played a blitz game; he again played 1. f4. He lost several pawns, but no obvious blunders; I was short on time and was unable to promote, drawing by repetition.
• His accuracy was 82.0%, best moves 52.4%; 6 mistakes, 3 blunders by him.
• 2,966 blitz games (as of 12-1-19); 57.1% win rate; recent blitz rating, 1386.
• 1264 rapid games; win rate 52.5%; recent rapid rating, 1350.

Antonio-Bot ---

• Rapid game; the bot played KID as white for 6 moves before varying off book. It then steadily lost material, until I had overwhelming power to mate with 3 pieces swarming around the K.
• His accuracy, 24.9%, best moves 26.8%; 2 mistakes, 1 blunder (really, many more).
• Played blitz, too; The bot lost ground over a long game; I held the advantage throughout. His accuracy was 86.0%, best moves 55.3%; 1 mistake, 5 blunders.
• 940 blitz games (as of 11-27-19); 62.6% win rate; recent blitz rating, 1471.
• 2,696 rapid games; 55.6% win rate; recent rapid rating, 1390.

 Wendy-Bot ---

• Rapid game; After losing material, she sacked a N for a pawn, going down further, which is perhaps what the bio meant by throwing her pieces into the fire of an attack.
• Her accuracy was 68.4%, best move 41.2%; 2 mistakes, 1 blunder.
• Also played blitz; she played poorly, w a back-rank mate on 31 after never castling. Her accuracy was only 12.1%, best moves 23.3%; 3 mistakes, 3 blunders.
• FonsecaSF feels this is the first club level bot he played.
• 4,158 blitz games (as of 12-2-19); 57.2% win rate; recent blitz rating, 1460.
• 2,185 rapid games; 52.4% win rate; recent rapid rating, 1392.

 Komodo11 ---

• Recent rapid game; It played book for 9 moves in Sicilian, Dragon, Soltis line, then varied off book, losing pieces. I was 11 points ahead in end game, then promoted a pawn for mate.
• Its accuracy was 52%, best move 36%; 0 mistakes, 3 blunders.
• Also played blitz game; I forked two minor pieces, then took several pawns, with easy end game; It’s accuracy was 57.9%, best move 38.6%; 4 mistakes, 2 blunders.
• 121,717 blitz games (as of 11-24-19); 75.4% win rate; recent blitz rating, 1366.
• 56,873 rapid games (11-24-19); 72.0% win rate; recent rapid rating, 1446.

 Pablo-Bot ---

• Bio says he likes gambits; Did not see any in rapid game; He lost in the end game; His accuracy was 63/8%, best moves 33.8%; 3 mistakes, 3 blunders.
• 3,209 blitz games (as of 12-4-19): win rate, 66.3%; recent blitz rating, 1650.
• 1,357 rapid games; win rate, 58.8%; recent rapid rating, 1483.

 Komodo12 ---

• Rapid game; K12 is much better than K11, and possibly better than current rating levels. Its accuracy was 91.2%, best move 51.4%; 3 mistakes, 0 blunders.
• 91,084 blitz games (as of 12-4-19); 78.5% win rate; recent blitz rating, 1552.
• 41,704 rapid games; 73.1% win rate; recent rapid rating, 1489.

 Isabel-Bot ---

• 5,460 blitz games (as of 12-7-19); win rate, 67.9%; recent blitz rating, 1593.
• 2,341 rapid games; win rate, 68.5%; recent rapid rating, 1511.

 Joel-Bot ---

• Bio says he likes the King’s Indian Attack, and likes blitz more than studying.
• 4,827 blitz games (as of 12-7-19); win rate, 63.2%; recent blitz rating, 1613.
• 1,652 rapid games; win rate, 61.5%; recent rapid rating, 1520.

 Isla-Bot ---

• Bio says she loves to attack in the opening, but will defend if she loses initiative.
• 8,594 blitz games (as of 12-6-19); win rate, 76.7%; recent blitz rating, 1753.
• 4,136 rapid games; win rate, 70.9%; recent rapid rating, 1540.

 Nora-Bot ---

• Bio says she has been studying games of Magnus Carlsen and has learned lots of his tactics, but not his patience. Bio also warns to watch out for her queen.
• 3,868 blitz games (as of 12-8-19); win rate, 70.4%; recent blitz rating, 1796.
• 1,428 rapid games; win rate, 70.2%; recent rapid rating, 1543.

 Jonas-Bot ---

• Bio says he has crazy, aggressive moves.
• 4,507 blitz games (as of 12-8-19); 68.4% win rate; recent blitz rating, 1755.
• 2,311 rapid games; 67.6% win rate; recent rapid rating, 1676.

 Komodo13 ---

• 65,752 blitz games (as of early Nov. 2019); 81.7% win rate; recent blitz rating, 1775.
• 25,517 rapid games; 74.0% win rate; recent rapid rating, 1681.

 Arthur-Bot ---

• Bio says he is a balanced and careful player.
• 4,577 blitz games (as of Nov. 7, 2019); win rate, 73.4%; recent blitz rating, 1744.
• 1,828 rapid games; win rate, 67.3%; recent rapid rating, 1663.

 Lorenzo-Bot ---

• Bio says he loves a good attack and has won many games in the King’s Indian.
• 1,729 blitz games (as of 12-8-19); win rate, 76.5%; recent blitz rating, 1852.
• 416 rapid games; win rate, 72.5%; recent rapid rating, 1678.

More to Come ---

The next part of this series on Man vs Machine will contain a write-up on the better computer bots and Komodo levels, detailing their playing abilities. This series will thus have at least some basic info on all 60 computer bots, with the ones I played containing more details. I will round out the series with a summary of my thoughts on the existing computer levels and the new computer bots, done in a plus / minus or pro and con style. 

Stay tuned for Part III of Man vs Machine!