Are Chess.com Adaptive Bots Truly Adaptive?
About three weeks ago, I published a piece entitled “Are Chess.com Bots’ Ratings Accurate?” where I discussed the strengths and playing styles of the standard bots. As promised, this is the second of the three-part series. Here, I will be looking at another category, the Adaptive Bots.
Adaptive Bots are deliberately designed not to play at a consistent strength. Instead, they make stronger moves when losing and weaker moves when winning. I was curious to know if this offers a better experience or the erratic play turns out to be so unnatural that it only frustrates players. To find the answers, I sampled three of the five bots, rated from 600 to 2000.
Jimmy (600)
Jimmy is the weakest adaptive bot. With my own experience, any engine at this rating will play thoughtlessly. Hence, I did not expect him to play any worse when he has the advantage, but he could possibly make smarter moves when losing. To get a feel for how his non-adaptive cousin fares, I also played Milica (550) to compare their games.
Assessment
Whenever I got the upper hand, Jimmy failed to show any noticeable increase in strength. This was a let-down. However, in other games, after force-feeding Jimmy with loads of material, he recognised he was winning. His adaptive instinct kicked in and he went out of his way to return material. In contrast, Milica did not play absurd moves such as Qxf7?? (Jimmy's third game).
In my opinion, Jimmy would make a good opponent for anyone who is extremely new to chess. The bot’s adaptability will allow players to quickly recover from blunders.
Tomas (1200)
As a mid-tier bot, Tomas “will give almost anyone a good game, and a tough positional battle”. While I could push him to the limit by playing strong games, I wouldn’t be able to objectively test his performance against weaker players. Hence, I invited one of my students, whom we will call Peter, to have a match with Tomas. Peter is rated about 900 on chess.com and loses to standard 1200 bots about 80% of the time.
Assessment
Tomas rose to the occasion when up against strong and aggressive opposition. He found ways to stay in the game for as long as he could. By my estimate, Tomas’ upper limit was about 1500, or about 300 points higher than advertised. On the other hand, his play did not weaken to the same extent when pitted against lower rated opponents. Against Peter, Tomas stubbornly stuck to his default strength. Overall, he is an adaptable bot but should be rated about 100 points higher.
Natasha (2000)
Standing tall at a rating of 2000, Natasha is the strongest of all adaptive bots. Understandably, her rating cannot be any higher since she “eases up sometimes”, especially in winning positions. To assess her strength and adaptability, I would experiment with the strategy of conceding an advantage early in the opening to have an easier game.
Assessment
When I played my strongest, I would quickly secure an advantage within the opening phase. Natasha responded by defending more tightly to create a challenge. Throughout the rest of the game, she ensured I had to work hard for the full point. In games where I purposely dug myself into inferior positions from the start, Natasha “eased up” a little too much and overcompensated for my lousy moves. She continued to make mistakes even after I played more accurately to regain the advantage. This irregularity makes it hard to pin down her rating. Hence, I can only give it a range – 1600 to 1900.
Conclusion
Fluctuating at about a 300-point range, the adaptive bots generally live up to their name. Of the three, Jimmy’s advertised rating is most accurate. Tomas is a tad underrated while Natasha’s quirk makes her an unrealistic 2000-rated player.
If you have been keeping tabs on the online chess community, you would have heard the buzz about Chess.com’s Celebrity Bots stomping around in the chess scene. In the final instalment of this three-part series, I will be taking a closer look at these supremo droids. Feel free to follow me if you’d like to be notified when the article, Chess Celebrities vs. Celebrity Bots, is published. In the meantime, do not hesitate to leave a comment and let me know which celebrity bots you’d like me to analyse.
