IBM’s Watson supercomputer has pulled off quite a feat, beating Jeopardy’s top two contestants ever. But it can’t answer every question (or form the right question for every answer, as in the case of Jeopardy). Try these, Watson:
- Which doctrines in Scripture are “weightier matters” and which are like tithing on spices?
- What is the main purpose of man’s existence?
- What is the purpose of the ending of 2001, A Space Odyssey?
- Which is a better series, The Chronicles of Narnia or His Dark Materials?
- What is your favorite song?
- How do you feel?
(Of course, Ken Jennings and Brad Rutter might have trouble with these, too. I’ve never asked them.)
Most of these questions are questions of evaluation, and evaluation requires some standard or criterion against which to judge. Watson could access sales data and determine which song is the most popular in the history of recorded music sales. It could perhaps access an article or ten pontificating on what Stanley Kubrick and Arthur C. Clarke had in mind with the Star Child. But it wouldn’t really be answering the questions like we would expect a human to do so.
As usual, Stanley Fish comes through with a good quote on Watson:
as the philosopher Hubert Dreyfus explained almost 40 years ago, a “computer is not in a situation” (“What Computers Can’t Do”); it has no holistic sense of context and no ability to to survey possibilities from a contextual perspective; it doesn’t begin with what Wittgenstein terms a “form of life,” but must build up a form of life, a world, from the only thing it has and is, “bits of context-free, completely determinate data.”
I don’t know if all of this is true; Watson has no doubt carried in some contextual assumptions from its creators. But it’s getting at a truth, namely that Watson is missing some pole stars. It has only a spotty and incomplete worldview. That’s bound to happen if your worldview is entirely inductive, built up from a collection of billions of facts. A person must (and will) have a set of organizing principles which undergird his facts.
All of that is true, but I think what believers in computer intelligence would tell you is that it is possible to give computers principles so they could at the very least simulate a worldview.
The more radical folks of the Singularity would probably go further and point to the fact that A.I. is increasing exponentially every year and computers could very easily surpass humans in this century and redefine humanity as we know it.
I still think computers are merely programmed, so even if they surpass us in intelligence (however you define that) and “take over the world” as a result of the knowledge and principles we give them, they will be historically dependent on humans. John Searle also had a good take on Watson, in a WSJ column I believe.