For years, the game of chess has been seen as a litmus test for how far AI can go against the human intellect. When IBM’s Deep Blue supercomputer beat reigning Chess world champion Garry Kasparov in 1997, it was deemed a pivotal moment. The Wall Street Journal called it “one giant leap backward for mankind.”
It hasn’t been a total route for humanity, however. Just a month ago, Norwegian chess grandmaster Magnus Carlsen beat ChatGPT in a chess game without losing a single piece. Interestingly, the AI bots are fighting, too. Earlier this month, ChatGPT (backed by OpenAI’s GPT-o3 reasoning model) beat Grok, an AI chatbot developed by Elon Musk-led xAI, in a chess tournament.
But how does an average chess-loving person fare against AI bots at chess? Well, it’s frustrating. One of the most recurring themes you will see on chess forums is that bots play “differently from humans.” Of course, when you’re playing against a chess computing bot like Stockfish 16 that can assess over ten million positions per second, not many players stand a chance.
On the other hand, some seasoned players say beating chess bots is easier because they follow a pattern, and that one must know how to survive the initial assault to beat them. But at the end of the day, AI bots don’t play by conventions. A researcher at Carnegie Mellon University has now come up with a less alien solution – an AI chess bot that plays like a human.
Say hello to Allie
The bot named Allie is the brainchild of Yiming Zhang, a PhD candidate at the Language Technologies Institute (LTI) in CMU’s School of Computer Science. Interestingly, Zhang found himself eager to play chess after watching Netflix’s popular series, “The Queen’s Gambit.” But soon after dipping into the world of online chess, he found himself frustrated by chess bots.
After playing against them, he realized that these bots play unnaturally. Moreover, the underlying tactics behind a chess engine often make it nearly impossible to beat them, thanks to their training, which involves winning at all costs by doing increasingly complex calculations.
That’s where Allie differs from your average chess-acing bot. It has been trained on 91 million transcripts of chess games played between humans. As a result, the way it contemplates moves, makes attacking advances, and defends positions feels like an average human player.
“Allie is trained on log sequences of real chess games to model the behaviors of human chess players across the skill spectrum, including non-move behaviors such as pondering times and resignations,” says the research paper. During evaluations, researchers found that Allie actually “ponders” at critical situations in the game.
Go ahead and test your mettle
The fact that Allie has been trained to think like a human doesn’t mean it’s a weak player. Far from it, actually. It can hold its fort against everyone from amateurs to grandmasters. “Against grandmaster-level (2500 Elo) opponents, Allie with adaptive search exhibits the strength of a fellow grandmaster, all while learning exclusively from humans.”
Since being deployed publicly, it has amassed more than 11,500 online chess games on the online platform Lichess, where you can also try your skills against it. So far, it has over 6,500 victories against human players, lost just over 4,000 games, and more than 500 battles have ended in a draw.
“For beginners, it’s not interesting or instructive to play against chess bots because the moves they make are often bizarre and incomprehensible to humans,” Zhang explains. Interestingly, Allie is completely free and open-source, which means other researchers can build atop it.
Do keep in mind it’s only accepting invites for Blitz games. Moreover, if you want to learn how the human-like AI chess bot makes its moves before going against Allie, you can watch it in action versus other human players at Lichess. And if you want to take a peek at the code, head over to the GitHub repository.
