AlphaZero : A computer program which is not quite a computer program

We recently hosted Harsh Patel, a cousin. He is also a research intern at ISRO. As we had dinner and decided to play a couple of rounds of chess, he sparked my curiosity with a curious tale of AI – @AlphaZero. An AI by Google DeepMind.
It’s a neural network based Chess algorithm which learnt chess by playing against itself in 3 hours!

I knew how computers played chess. They use a brute-force method, where they will calculate variations for each of the possible positions and then choose a move that gives them the best possible outcome. [A numerical value is assigned to each outcome, the computer usually chooses moves with the best probability to give it a positive outcome]. The focus is on making a choice that you KNOW will give you a better position at the end. And hence remembering it.

This is also how grandmasters prepare. Loving chess, I knew that this what is taking away the beauty of a once creative game, which is now evolving to be more of a memory exercise, where the player who can memorize the deep lines of a complicated variation usually wins.

That’s not how AlphaZero played. We watched analysis of a few of its’ games against Stockfish Chess Engine, an excellent conventional chess engine.
In it’s play, one could see the romanticism of the masters of 1700s. The wild sacrifices creating often novelties and positions that are unheard of, in modern chess – it seemed to understand the soul of chess.
This encounter left me stumped. AI isn’t just about efficiency; it’s about unlocking new ways of thinking, creating, and pushing boundaries. AI might not just play like a grandmaster, but compose like a legend.

Will it transform the way we do business in today’s world? From creating customer journeys to marketing stories, will it give rise to creativity at a scale not seen before?

Is that scary? Is that exciting?! Love to hear in the comments