Lao Tsu said that if you give a hungry man a fish, you feed him for a day, but if you teach him how to fish, you feed him for a lifetime.
Sure chess engines can show you the best move of a position and help you win if you knew that move (you eat for that day) but chess engines cannot teach you why that move is good and how you can make good moves like that for the rest of the game. That falls in the realm of a good chess coach that will hammer in the foundation, good practices, relevant exercises for training etc.
AI Chess Coach not here yet
No AI can teach chess yet. What we need is a good Chess LLM that can mimic the function of an expert chess coach. ChatGPT, Perplexity, Claude etc are general AI chat bots that returns back intelligent answers based on a given prompt by the human. We have specialised LLMs like Suno for creating music, Mid Journey for creating realistic as well as artistic graphics and pictures. We even have LLMs that can generate realistic videos.
There is Chess LLM and they are not here yet.
The above 2 minute paper about Google Deep Mind is impressive and certainly is the starting steps to seeing an actual AI Chess Coach. At the moment it is more a strong analysis maching like a chess engine but the way it categorise positions line "Long Term Sacrifice", "Colour Weaknesses", and hundreds of other categories makes it more than an engine meaning it mimics understanding why a position is good or the opposite.
Again to emphasis this is still not an Chess LLM that can be used for chess coaching.
The main reason Chess LLM are not here yet is the cost and huge computation power to be invested. So far no company has decided to do this yet. The 2 minute paper is the closest. It is more an analyrical engine than a teacher.
Knowledgebase vs Training Data
All these AI system rely heavily on training data. This is different from a knowledgebase. For example a chess knowledgebase will be:
- Opening books with pre-calculated moves and evaluations
- Endgame tablebases with perfect play for positions with few pieces
- Strategic rules and heuristics coded by chess experts
- Databases of historical games
This knowledge is often hard-coded or stored in specific formats optimized for quick access during gameplay or analysis.
Training data for a chess LLM, on the other hand, would be much broader and less structured. It would include:
- Natural language text about chess (books, articles, commentary)
- Game records in various formats (PGN, algebraic notation, descriptive notation)
- Conversations about chess (forums, interviews, coaching sessions)
- A wide variety of chess-related content not necessarily in a chess-specific format
Copyright
Of course one would ask : What about copyright? You can't just take a book like "How to reasses your chess" by Jeremy Silman and feed it to the AI to learn.
That would be what a machine learning programmer would do. Take all these literature and feed it to the AI for learning. That's the key point : "It's for learning".
And it's not going to regurgitate facts and direct quotes from all the data that AI has taken in.
The copyright problem has been addressed in other AI systems like Suno. Suno was fed tons of actual popular songs or artists so that it can produce music. Of course, many musicians are up in arms over this as some of the songs very closely resemble their style and voice. Same goes for Mid Journey in image generation. Lots of real photos were fed in as training data and some photographers even tried suing when they saw AI photos produced that were similiar with their original pictures. Most of the legal cases were dismissed and these AI companies are still thriving today.
What other training data can an AI chess coach use?
AI chess coach can cover many other areas like psychology, physical preparedness and technical data during an actual chess game.
Psychology: AI Chess coach constantly monitors the human player's thoughts like asking the player how he or she felt during, before and after the game recording and tagging those thoughts to all the individual games it recorded.
Data during game: The time taken for each move made, the evaluation for each position, etc.
Physical data: How much sleep a player had daily, heart rates during the day via health smart bands/watches.
Other data like food and diet etc.
These are things a human chess coach might do but definitely not this level of detail.
Creepy?
The big advantage
Realistically, let's take a typical talented junior player aspiring to become FM, IM or GM. He or she would be training 4 to 8 hours daily. They would have one or more coach which they would spend 1 to 4 hours a week to get advice and directions and things to focus on their daily traning. If it was an AI coach, it would not be limited to that 1 to 4 hours weekly but as long as you want. And to top it off the daily training can be monitored and dynamically changed real time to suit that student.
So in the end we would have an AI that knows chess positions as well aspects of the human student it is training at an almost unimaginable level.
So will we ever see the emergence of this futuristic AI chess coach? I think so. It's just a matter of time.
