We’ve all been reading about the AlphaZero, the incredible superior being created by Google’s DeepMind to rule the chess world.
Well, it turns out the Artificially Intelligent behemoth of Biblical-proportions is human after all – or at least in a figurative sense.
Dvoretsky's Endgame Manual 5th Edition
AlphaZero may have destroyed Stockfish and crushed every computer on the planet. But, as Grandmaster David Howell has revealed, it hasn’t swept every human aside every single time.
GM Howell, a guest at Chessable’s meet-up event last month, told in his column for UK newspaper The Sunday Times that he and three other plucky Brits held AlphaZero to an heroic draw.
Given everything we’ve seen and read about the neural network this seems incredible. So how did this happen?
On GM Howell’s team were the businessman Christopher Flowers, the president of the English Chess Federation’s Dominic Lawson and International Master and chess impresario Malcolm Pein.
Having enlisted help of two masters with
So the stage was set. The awesome-foursome entered to the sound of Florence and The Machine ready to take on the beast. It was humanity’s last stand – how would they tackle it?
Following GM Howell’s suggestion, they played 1. e4 and went for the notoriously drawish Berlin Defence…
And to their utter surprise, AlphaZero obliged.
“Mercilessly, we took every opportunity to hoover off the pieces,”
He even described their tactics as “the chess equivalent of a side close to a desired nil-nil draw in football taking the ball down to the corner flag and standing on it”.
The Alpha males
“This result revealed a shortcoming (of a purely sporting nature) in AlphaZero. It doesn’t know that it is infinitely stronger than any opponent it might come across.”
But was this a human mistake to make? Mr Lawson thinks not.
“An equivalent human would never play an opening that allowed a much weaker opponent a route to a sterile position. He (or she) would play a more double-edged opening, even taking a theoretically inferior line, if it promised complications that would confuse a lower-rated adversary.
“That’s not (yet) something the neural networks of AlphaZero have been built to
GM Howell said: “To our
Here’s the second – and far more interesting – game: