The 10 Most Famous Chess AIs
Chess has long been considered a benchmark for artificial intelligence. The game’s combination of simplicity of rules but complexity of strategies has made it a great testing ground for AI.
Over the years, many different chess AIs emerged, each pushing machine chess skill into new territory and unlocking new insights.
Let’s explore the 10 most famous chess engines:
1. Deep Blue (1997)
Deep Blue, developed by IBM, made history in 1997 when it defeated world champion Garry Kasparov in a six-game match, becoming the first AI to beat a reigning world champion under tournament conditions.
A few years earlier, Kasparov said “No computer will ever beat me”. But during the match, Kasparov was surprised by the humanlike intuition of some of Deep Blue’s moves, famously declaring that the machine, at its best, “played like a god”.
Deep Blue was using highly customised hardware to achieve sufficient search depth, often calculating 30–35 moves into the future before making its play.
For me personally, reading about Deep Blue’s match with Kasparov was my first true exposure to machine intelligence, undoubtedly planting a seed that would grow into what I do today.
Deep Blue with its special hardware, although impressive, wasn’t something regular chess players could make use of.
Let’s look at the next candidate, representing the PC analysis revolution:
2. Fritz (1995-present)
Fritz has been a staple in the chess software market for decades. Developed by Frans Morsch and Mathias Feist, Fritz has consistently ranked among the top chess engines. Its user-friendly interface and strong play have made it popular among both amateur and professional players for analysis and training.
As one of the first strong chess engines to run on a regular PC, Fritz has been an important partner for many chess players in not least opening preparation.
Today, that role has largely been taken over by the next candidate:
3. Stockfish (2008-present)
Stockfish is an open-source chess engine that has dominated computer chess rankings for years. Known for its exceptional strength and constant improvements through community contributions, Stockfish has become the go-to engine for many players and analysts. Its open nature has also made it a valuable tool for chess research and development.
Traditionally being an engine relying on extremely deep and quick search, with a handcrafted evaluation function, Stockfish in 2020 made the biggest skill jump in many years by incorporating a fast, light-weight variant of neural networks, forgoing handcrafting and emulating the evaluation function of neural network engines like Alpha Zero – the next AI on our list.
4. AlphaZero (2017)
Developed by DeepMind, AlphaZero revolutionized the chess AI landscape at its arrival. Unlike its predecessors, AlphaZero learned chess from scratch, playing against itself and developing its own strategies. In just 24 hours of self-play, it was able to defeat Stockfish in a 100-game match, showcasing the power of machine learning in chess.
AlphaZero’s release had huge impact on both AI and human chess. In a sense, it was a victory for humanlike intelligence against the machine, as it showed that extremely deep search can be overcome with extremely deep intuition.
The intuitiveness of Alpha Zero’s strategies also made them transferable to humans in a way that previous computer tactics weren’t. AlphaZero gave human masters higher confidence in executing daring positional gambits such as exchanging material for activity or weakening the kingside with a h-pawn launch, by proving that these strategies were viable even against superhuman calculators like Stockfish.
AlphaZero was a closed-source project by DeepMind with very scarce information released.
Fortunately for the chess AI space, that was soon remedied:
5. Leela Chess Zero (2018-present)
Started as an attempt to replicate AlphaZero, Leela Chess Zero (or Lc0) is an open-source neural network chess engine, the torchbearer of its category. It uses machine learning techniques similar to AlphaZero but has relied on distributed computing power from volunteers for its training.
Together with Stockfish, Lc0 has been at the forefront of chess AI skill and innovation in the years since its inception, and is still a contender for the strongest chess engine in the world.
6. Komodo (2013-present)
Komodo, created by Don Dailey, Mark Lefler, and GM Larry Kaufman, has won numerous computer chess championships. Komodo’s ability to adjust its playing style based on the position has made it a favorite among chess professionals for analysis.
7. Houdini (2010-2019)
Developed by Robert Houdart, Houdini was once considered the world’s strongest chess engine. It dominated computer chess rankings for several years and was known for its aggressive and dynamic play style. Although development has ceased, Houdini remains a respected name in the chess AI world.
8. Rybka (2005-2011)
Rybka, created by International Master Vasik Rajlich, was a groundbreaking chess engine that dominated computer chess rankings in the late 2000s. However, its legacy is controversial due to allegations of plagiarism from other open-source engines. Despite the controversy, Rybka’s impact on computer chess is undeniable.
9. Fat Fritz (2020-present)
Fat Fritz is a neural network chess engine based on Leela Chess Zero but with its own unique training. Developed by Albert Silver and the Chessbase team, Fat Fritz combines the strengths of traditional engines with the learning capabilities of neural networks.
10. Maia (2021)
Maia is a research project attempting to train Leela Chess Zero-like neural networks on human games, to replicate the human playing style at different levels. The Maia project showed that such neural networks make more humanlike errors than the crippled engines normally used for lower-rated bots.
Use Cases of Chess AIs
These Chess AIs weren’t just built for beating humans (possibly with the exception of Deep Blue). They’re made for humans to learn from as well.
The traditional viewpoint has been that chess engines are great for identifying tactical blunders, and working out concrete opening variations, especially in positions that aren’t immediately intuitive to humans. Only recently, the usefulness of chess AIs has expanded to offering deep positional insights and themes.
Chess AIs have also been used as sparring partners ever since their inception. In the 80s, physical chess computers boomed, marketed as ideal practice partners when you couldn’t find a human opponent to play with.
The Future of Chess AIs
The evolution of chess AIs from Deep Blue to Alpha Zero follows the rapid advancements in artificial intelligence. While early AIs focused on brute force calculation, modern engines incorporate pattern-recognizing machine learning, and even attempt to mimic humans.
As AI technology continues to advance, we can expect chess engines to become even more sophisticated.
The next frontier in chess AIs might be to provide humanlike explanations and insights into why some moves are better than others. This is a notoriously difficult thing to do, but immensely valuable if done properly.
Who will be the first to make the next breakthrough?