Deep Thinking

This blog post is about Garry Kasparov's Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins. Yet, it's not a review, exactly. Nor is this a response. These are just my thoughts I had during the deeply enjoyable time I spent reading this excellent book.

I've got a confession to make. I really don't care much for playing chess. Mostly because I'm terrible at it. By the time I was old enough to recognize what made the game interesting, I was far behind and uninspired to catch up. Getting good at chess certainly requires thinking on one's feet and dynamic strategic changes. Those aspects interested me. But success in chess truly requires a lot more study and memorization as well. Knowing concretely that you've already established that a certain move could only lead to weaker positions in the future (i.e. memorizing) is a key part of serious play. For most serious chess players, the so called opening is a rather familiar walk through well traveled paths, eventually leading out into the vast less explored "middle game".

While I'm not a chess player, but my interest in artificial intelligence has, to a certain degree, demanded that I learn something about this prolific game. I'm interested in hearing about approaches to play. The best humans in the world today stand no chance against the best computer players. Yet, the advent of computer play has brought innovation in human play and especially human training. I find these ideas fascinating. I'm further impressed by the abstractions advanced players create to represent their knowledge in compact algorithmic formats. This doesn't even touch on the ways in which chess is interesting from a computational complexity point of view.

Having settled, I couldn't imagine better off to see inside mind Grand Master this book. Gary does not try to represent himself as a technologist, AI researcher, or software engineer, he has a surprisingly good understanding about a lot of the technology and innovation that was is the developing the machine that is able to outmaneuver him.

I have heard many people diminish the accomplishment of Deep Blue by pointing out that Gary wasn't just beaten by a machine. He was beat by a machine, the team of people that built it, and the company that funded the project. While this is true, Gary never makes this argument. For him, this is a unique adversary, but one he would go to battle as you get done with many before it, through preparation.

My favorite part of the text are the moments Gary discusses how he perceive the challenge in front of him, and the strategy he adopted for this series of games. He is very calculating at every step, and speaks critically about certain choices made during the games.

The famous mismatch between Gary and Deep Blue was not actually the first time they had competed. Yet this is one of the ways in which Gary prepared differently from how he would to face a human opponent. The engineer behind the Deep Blue were not just trying to make a good chest playing program. In some respects they were trying to build a "Gary defeating machine". For this reason much of his initial clay what's conservative in an attempt to get the machine to show its hand, and to avoid the possibility it's a deep search techniques had discovered some very long in Gary's typical openings. This is just one of the many exciting tidbits you shared about the preparation for this famous match.

I was very interested to learn that any high-profile match, Gary attempt to find and bring so-called "novelties" to the board. All major players have their games documented and available online. It's quite easy for chess player to study each other's former game, Recognize their strengths and weaknesses, observe their general strategy, and discover novel secrets that player may have. If a player determined that a certain board configuration looks bad for one player at a glance, that is actually a dominant position given deeper analysis, this can be described as a novelty. It's a sort of crap player might delay for their appointment to overlook. Any good trap, once you do, is likely to be recognized in future matches. This is a very curious concept that is totally absent (to my knowledge) from approaches to algorithmic playing games. The idea of finding a novelty and keeping it secret until needed, is the form planning that I don't believe any computer presently does.

The book is also an excellent treatise on the impact computers have had on this game. The introduction of computer play has been impetus for improvement in human. The computer turned out not just be an opposing opponent, but also a powerful training tool.

I deeply enjoyed this book and cannot recommend it enough for people interested in game, machine learning, artificial intelligence, or adversarial systems.