Machine Intelligence and Mind Games

Former chess champ, Garry Kasparov, wrote this very interesting book titled Deep Thinking. Once a upon a time, chess was the holy grail for machine intelligence. Claude Shannon, founder of information theory, explained why. Chess is a sharply defined game, its objective is clear, and the last reason is best explained in Kasparov’s words:

“Since chess requires thinking, either a chess playing machine thinks or thinking doesn’t mean what we believe it to mean.”

There was also the hope that training a computer’s guns at chess may lead to other learnings, things far deeper than chess. Unfortunately though:

“Chess just wasn’t deep enough to force the chess-machine community to find a solution beyond speed… Patterns, knowledge, and other humanlike methods were discarded as the super-fast brute force machines took home all the trophies.”

Which is why Google’s AI, AlphaGo, that beat the human world champion at the Chinese game Go, a game “too big of a matrix to crack by brute force, too subtle to be decided by the tactical blunders that define human losses to computers at chess”, was “more interesting as an AI project”. Indeed, as I wrote earlier.

In an age where Google can find you images, we take machine intelligence for granted. But as Kasparov writes, that path was long, hard and at times amusing. Like the time they fed this machine hundreds of thousands of positions from (chess) grandmasters’ games to teach it. Here’s what happened when it played a real game:

“(It) launched an attack, and immediately sacrificed its queen! It lost in just a few moves, having given up the queen for next to nothing. Why did it do this? Well, when a grandmaster sacrifices his queen, it’s nearly always a brilliant and decisive blow. To the machine, educated on a diet of GM games, giving up its queen was clearly the key to success!”

Correlation ain’t causation.

The other area where the book is very interesting is Kasparov’s description of the mind games with the computers, Deep Thought and later Deep Blue:

“As a believer in chess as a form of psychological, not just intellectual, warfare, playing against something with no psyche was troubling.”

He switched from his usual style of play to create new scenarios, thereby depriving the computer of the option to dig into its database of older games. He tried to strangle it by creating situations that required analysis of moves deeper than what a computer’s memory could perform.

Ironically, the IBM team did similar things. If Kasparov played an anticipate move in certain scenarios, they programmed the computer to reply instantly. Why?

“This has a psychological impact, as the machine becomes unpredictable, which was our main goal.”

And Kasparov made an assumption: while a computer may make mistakes or miss things in the distant future of a game, it would never miss anything in the near future of a game. Therefore, he reasoned:

“If it is allowing you to play a winning tactic, it’s probably not winning at all.”

That sounds very reasonable, right? Except, that mode of thinking is why Kasparov resigned in a game that could have been drawn in a handful of moves. Why? Kasparov couldn’t imagine that a computer would have missed a way to draw just a few moves down the game!

In another game, Kasparov was convinced that IBM was cheating, that humans were intervening behind the scenes. He even said as much in an interview, saying the computer’s move felt like the “hand of God” goal of Maradona. Worse, his demand that the computer’s logs, a record of what it was “thinking” at each step, be made available to the judge, was turned down by IBM. So was this proof that IBM was cheating? Or, as Kasparov himself wonders, was the refusal just a part of IBM’s psychological war, a way to let Kasparov get all wound up with conspiracy theories and thereby not play his best?

“If they can get you asking the wrong questions, they don’t have to worry about the answers.”

I never knew the man v/s machine chess games had involved psychological warfare… from both sides!