Monday, May 16, 2011

Why I'm not worried about Watson . . . yet

When IBM's Watson beat the human champions at Jeopardy!, what made made its victory a little unexpected was that instead of us playing a game within the world created by the computer, the computer competed in a game that we ordinarily play against each other, in our world. In a sense, it beat us at our own game.
But was that really so frightening? Unlike us, Watson did not know it was playing a game, in the sense that we view games as differing from “real life.” No one is yet claiming that Watson is sentient: it remains a logic engine with access to a vast database of factual information, and instructions about how to respond to real-world ambiguities; its answers depend on probabilities stemming from how well or poorly it understands the questions asked.
The real challenge for Watson was not so much in knowing the answer as it was in understanding the question: once its programmers had taught it how to make sense of natural language, and the computer could translate that input into the sort of logical query required to analyze the vast amounts of data available to it, it became a matter of processing power to make it practical to compete in real time against human opponents. And its power was such that within the limited universe defined by the rules of Jeopardy! it became very hard for a human being to beat it.
We’ve yet to create, or encounter, self-conscious artificial intelligence of the sort Hollywood likes to portray. Watson plays Jeopardy! well because the game exists within a carefully limited set of rules that constrain what is possible, not because it finds the challenge consciously stimulating, or fun.
We play our games for different reasons than computers. Yes, we like rules: Basketball courts are ninety feet long, with ten-foot goals, and five players on each side who score by putting the ball in the hoop; there are no forward passes in rugby; chessboards have sixty-four squares; two cards are dealt face-down in Texas Hold ’em. For human beings, the challenge is playing within the rules. That’s not a challenge for Watson, which ultimately depends on rules. Within them, it uses raw computing power to explore all possible solutions before settling on the most probable or efficient: its real challenge would be playing where there are no rules, or making them up as you go. Humans do that all the time.

1 comment:

Raoul Rubin said...

Reminds me of the "Chinese Room" thought experiment. It is a rather involved setup, but it addresses whether a computer can understand its own algorithmic processes.