Programmable alarm clocks
What are the stylistic differences between computers and top human chess players?
“Deep Blue was intelligent the way your programmable alarm clock is intelligent. Not that losing to a $10 million alarm clock made me feel any better.” Gary Kasparov
In my last blog I discussed how top human chess players were starting to play more like computers. Today, I’ll explore what it means to play more like a computer. Computers process chess games very differently from humans, so we would expect them to play in a different style, but what does that style look like?
I explore how computers play by comparing the recommendations of Stockfish (a top chess engine) with the actual moves from a dataset of 300,000 moves played across 3,600 games played by top Grandmasters in World Championship matches and qualifiers.
Two differences are clear from my analysis. First, computers are more likely to move their pawns than humans. While the top grandmasters moved their pawns on average 22 percent of the time, computers recommended pawn moves 25 percent of the time. This indicates that computers are more willing to make committal moves. Once you have moved a pawn forward you cannot return it.
The second difference is that computers are more likely to play on the wings of the board. The graph below compares the move locations of human moves and computer recommended moves. It shows the destination of the moved piece (for example if a player advanced a pawn from d2 to d4 it would count as d4).
Both computers and top players play most of their moves around the centre of the board. One of the key tenets of chess strategy is to occupy the centre of the board. Having control of the centre allows your pieces flexibility to attack and defend on either side of the board, whereas leaving your pieces on the edge of the board can strand them.
However, the computer recommended moves are more likely to deviate from this accepted wisdom. The right-hand side of the above chart shows the difference in frequencies between computer recommended moves and human moves. When playing as white, top humans only move pieces to h4 or a4 on the wings of the board 0.97 percent and 0.9 percent of the time respectively, computers recommend moving pieces there more frequently (1.22 percent and 1.16 percent of moves).
So, should top humans try to replicate this computer inspired style? Not necessarily. The stylistic differences of computers partly come from their in-built computational power. Computers can break from the conventional wisdom, as they can calculate the material benefits it will bring them 10 or 20 moves down the line. While top chess players can calculate impressively far, they always will be imperfect, especially when limited on time. Given these constraints, it makes sense to be cautious sometimes and to rely on principles where necessary. While humans can learn from computers, they will always need to approach the game with a little more humility.
interesting analysis,
also, in recent years, pushing the h-pawn, has became more and more fashionable, thanks to alpha-zero