Neural Nets: A New Type of Chess Engine
Computers have revolutionized the game of chess. There are now chess engines capable of beating any player in the world again and again.
Since their inception, chess engines have worked in similar ways. They have become more advanced, but followed the same basic method: meeting the enormous challenge of chess probability with brute-force calculation.
In recent years, chess engine developers have asked a new question: What if a computer could think like a chess player?
What is a Chess Engine?
Also known as a chess calculator or chess computer, a chess engine is software designed to calculate chess moves. These powerful programs calculate massive numbers of variations to see which ones are likely to provide a material or positional advantage.
Chess computers have been around since the 1950s. Famously, in 1997, a chess engine called Deep Blue defeated the world champion Garry Kasparov in a widely publicised match. Today, chess engines can perform consistently better than the best human players in the world.
Chess engines are now central to the world of chess. Globally, people play against the computer, use analysis boards to evaluate positions, and carry out various forms of computerized chess training.
Related: Should You Play Chess Against the Computer?
What is a Neural Network?
Neural networks are a subset of artificial intelligence that attempt to mimic the human brain. Using a type of machine learning called deep learning, these networks can process massive data sets and producing outputs, which can be accessed through a user interface. They excel at pattern recognition, complex problem-solving, and making predictions.
How do they mimic the human brain? By using connected nodes, these networks mimic the billions of neurons that allow us to think. The system takes in information, processes it in different ways, and creates an output. In many ways, this reflects a simplified version of how we experience, understand, and interact with the world.
Traditional Chess Engines vs Neural Net Engines
Today, chess engines exploit this technology to create some of the most powerful non-human chess players. Let’s look at some key differences in how neural net engines differ from traditional chess calculators.
Traditional Engines
Traditional chess engines work by calculating incredible numbers of possible variations in a position. Through pure computational power, they can determine which move is most likely to win the game.
While you or I might consult an opening book to see how to start, a chess engine will calculate variations far into the game before deciding on its first move. They cannot solve entire games because of the sheer number of possible positions. Instead, they evaluate which series of moves is best based on deep calculations and an understanding of things like material, pawn structure, piece activity, and king safety. Nevertheless, the key idea behind recent chess engines that do not use neural networks is to evaluate hundreds of millions of positions and find the best one.
Neural Nets
Neural net chess engines have a more creative approach. Their design is to recognize patterns and understand the game through their own learning methods. Primarily, they are trained using self-play in which they essentially play out huge numbers of chess games and learn their own positional principles, probabilistic calculation, and tactics.
Neural nets do not aim to calculate every possible variation. Instead, they develop a deep understanding of the game that allows them to come up with advanced strategies. They do calculate (they’re very good at it), but they only explore promising variations based on what they have learned about the game.
In this way, neural nets reflect a more human approach to chess. Instead of purely mathematical calculations, these new engines develop their own unique way of playing.
Related: Chess Strategy Vs Chess Tactics: What's The Difference
Are Neural Nets Better?
Today, the strongest chess engine in the world is Stockfish. It is an open-source engine that primarily utilizes massive computing power to perform the most complex calculations. However, it now uses neural networks for evaluation in its latest versions. The neural nets support the calculations by helping Stockfish decide which of millions of possible variations will have the best chance of success. This has become a popular hybrid approach employed by the world's most powerful chess engines, such as Fritz and Komodo.
Original neural network engines like AlphaZero, which initially were stronger than Stockfish, are now not as powerful as some of the hybrid engines. It seems that the future of chess engines will combine artificial intelligence innovation with raw calculating power.
Chess Engine Competitions
In several events, chess engines compete to determine which is the best player. In recent years, engines that use neural networks have often claimed victory.
World Computer Chess Championship
The World Computer Chess Championship (WCCC) is an event that happens every several years in which chess engines compete against each other. It has been around since the 1970s and demonstrates a fascinating history, showing how chess engines have developed alongside computing, the Internet, and Artificial Intelligence.
Recent winners include Leela Chess Zero. This engine, built following the example of AlphaZero, uses a neural network learning model. Other recent winners use integrated neural network technology in their engines.
Top Chess Engine Championship
Another competition that pits chess engines against each other is the Top Chess Engine Championship (TCEC). In this more recently created competition, Stockfish competes and has taken the bulk of victories in the last decade. The most common runner-up and occasional winner? Leela Chess Zero.
Overall, these tournaments reflect the growing strength of neural network chess engines.
Neural Nets and the Future of Chess
Neural network technology is ushering in a new era of chess engines that may change the game for everyone. Stockfish, the most powerful engine in the world, now uses a hybrid approach in which neural nets support its evaluative functions.
Traditionally, computers are very good at tasks that humans are limited in, like mathematical calculation and storing vast quantities of information. It is through these skills that chess computers first became able to beat humans. Now, computers are surpassing humans in chess through a process that reflects study, creativity, and innovation.
Many of the moves made by purely calculating chess engines seem not to make sense. They have calculated so far into the game that we cannot identify the benefits of an individual move. The play of neural net chess engines can look slightly different. While still involving advanced calculations, these programs play in a way that reflects the intuitive genius of great chess players.
What Does This Mean for Chess Players?
For some, the rise of chess computers may make improving at the game feel pointless. For most of us, chess computers provide new ways to interact with the game.
Alongside the improving technology, human players have gotten better as well. The question of how far advanced neural net engines can become remains open. Playing against these new opponents, learning from them, and seeing how they change the world of chess are all incredible opportunities.