M2M Day 338: Should I try to learn chess like a computer?
This post is part of Month to Master, a 12-month accelerated learning project. For October, my goal is to defeat world champion Magnus Carlsen at a game of chess.
In the past few days, I’ve played many games of chess on Chess.com and spent a lot of time researching the game, approaches to learning, etc. The hope is that I can find some new insight that will enable me to greatly accelerate my learning speed.
So far, I’ve yet to find this insight.
Chess is a particularly hard game “to fake” because, in almost all cases, the better player is simply the one who has more information.
In fact, in a famous study by Adriaan de Groot, it was shown that expert chess players and weaker players look forward or compute lines approximately the same number of moves ahead, and that these players evaluate different positions, to a similar depth of moves, in roughly similar speeds.
In other words, via this finding, de Groot suggests that an expert’s advantage does not come from her ability to perform brute force calculations, but instead, from her body of chess knowledge. (While it has since been shown that some of de Groot’s claims aren’t as strong as originally thought, this general conclusion has held up).
In this way, chess expertise is mostly a function of the expert’s ability to identify, often at a glance, a huge corpus of chess positions and recall or derive the best move in each of these positions.
Thus, if I choose to train in traditional way, I would essentially need to find some magical way to learn and internalize as many chess positions as Magnus has in his over 20 years of playing chess. And this is why this month’s challenge seems a bit far-fetched.
But, what if I didn’t rely on a large knowledge base? What if I instead tried to create a set of heuristics that I could use to evaluate theoretically any chess position?
After all, this is how computers play chess (via positional computation), and they are much better than humans.
Could I invent a system that let’s me compute like a computer, but that can work with the processing speeds of my human brain?
There’s a 0% chance that I’m the first person to consider this kind of approach, so maybe not. But, there’s a lot of data out there (i.e. I have downloaded records of every competitive chess match Magnus has ever played), so perhaps something can be worked out.
Clearly, I can’t play by the normal chess rules if I want any shot of competing at the level of Magnus. (By “normal chess rules” I mean the normal way people learn chess. If I could just bend the actual rules of the game, i.e. cheating, then this challenge would definitely be easier…).
I’m skeptical that some magical analytical chess method exists, but it’s worth thinking about it for a few days, and seeing if I can make any progress.
Hopefully, soon, I’ll be able to formulate an actual training approach for this month. Right now, I’m still floating around in the discovery phase.
As soon as I have an interesting training idea, I’ll be sure to share it.
Read the next post. Read the previous post.