M2M Day 364: Playing my first game of algorithmic chess (i.e. Max Chess)

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.

Yesterday, after eight hours of processing, I was able to test the first version of my chess algorithm.

Sadly, it was only able to correctly evaluate chess positions as good or bad 70% of the time, which isn’t very good for my purposes.

Of course, this algorithm was created only based on 300 chess games, while I have access to over 200 million games. So, hopefully the algorithm will become more accurate as I feed it significantly more data.

Even if I can create an algorithm that is 99% accurate though, it’s still unclear how well this algorithm can be used to play chess at a high level.

In other words, even if I can identify and play “good” chess moves with 99% accuracy, is this enough to defeat the world’s best chess player?

In order to find out, I built a program today that let’s me test out my algorithm within actual gameplay. Here’s how it works:

1. Since I’m playing with the White pieces, I make the first move. To do so, I suggest a move to the program.
2. The program then runs this move through my algorithm and decides whether the move is good or bad.
3. If it’s good, the program automatically plays the move. If not, it asks me to suggest another move until I find a good move.
4. Then, my opponent responds with the Black pieces. I enter Black’s move into the program.
5. After Black’s move is recorded, I suggest my next move, which is again played if the move is evaluated to be good.
6. And so, until the game is over.

Here’s what the program looks like running in my Terminal:

In this way, I can test out different chess algorithms to see how they perform against chess computers of different strengths.

If I can find an algorithm that, when used in this way, allows me to defeat Magnus Age 26, within the Play Magnus app, then I can proceed to learn the algorithm, so I can execute it fully in my brain.

Right now, since the algorithm only has an accuracy of 70%, it’s still very bad at chess. I played a game, blindly using the algorithm, against Magnus Age 7 within the Play Magnus app, and I was very quickly defeated.

Still, it’s an exciting milestone: I finally have all the pieces I need to actually play algorithmic chess (i.e. Max Chess). Now, I just need to improve the algorithm itself…

Read the next post. Read the previous post.

Max Deutsch is an obsessive learner, product builder, and guinea pig for Month to Master.

If you want to follow along with Max’s year-long accelerated learning project, make sure to follow this Medium account.