The latest man-versus-machine contest is happening right now, this one on the game show "Jeopardy!" For three consecutive evenings starting on the 14th, the computerized contestant "Watson" strut its cognitive stuff on the air. Watson will match wits with Ken Jennings -- who won 74 games in a row, the longest streak ever -- and Brad Rutter, who holds the record for "Jeopardy!" winnings, more than $3.2 million.
As many as 20 IBM scientists have worked on Watson over the last few years. Why would a company commit so much time and money to such a, well, trivial pursuit? Well, think about just what it is that "Jeopardy!" really tests in people. It's nothing less than the natural human ability to negotiate a vast amount of general knowledge -- to make connections and assess relationships between bits of information and discard useless facts and focus on what's most relevant -- and all very rapidly. In short, it tests people's intellectual suppleness to make educated guesses.
We all do this every day, of course, mostly outside our conscious awareness, but that doesn't make it any less remarkable. Indeed, playing "Jeopardy!" successfully requires much more cognitive flexibility than the ultimate egghead's game, chess. That's because chess is a rule-based game with finite possibilities. Human knowledge may be trivial but it's infinite.
If the IBM scientists are successful, they will have figured out one of the most robust cognitive strategies at work in the human mind, which scientists call "anchoring and adjusting." Our thoughts are dynamic, constantly in motion, and anchors keep us from ranging wildly and randomly.
To illustrate, imagine you are actually a contestant on "Jeopardy!" It's Final Jeopardy! and you're in a dead heat with the two other contestants. The category is Founding Fathers, which is good because you know a bit about 18th century history. Alex reveals the Final Jeopardy! answer: George Washington was elected president of the United States in this year.
Uh-oh. You don't know for sure. You're going to have to make an educated guess, but how do you come up with your best guess in just seconds? Most of us don't know precisely when the father of our country was elected. It's not the kind of information we need to keep accessible in your memory every day. When we don't have the precise answer we need, we use the next best thing: a factual anchor. That is, we probably have some other related facts stored in our heads. We probably know, for example, that the Declaration of Independence was signed in 1776.
Okay, 1776, so that's at least a starting point. But you know it's not a game winner, and the clock is ticking. So you start adjusting. Was it the next year, 1777? Five years later? Could it have been as much as 20 years later? What was going on in those years anyway? In your mind you come up with a range of plausible answers, moving away from the anchor in increments, and as time runs out you jot down the best of the possibilities.
This is basically what Watson, Jennings and Rutter are doing as we watch them. But it's an imperfect strategy. "Jeopardy!" contestants -- all taken together -- get about eight of 10 questions correct. That's impressive, but it also means they get almost 20 percent wrong. Psychologists have been running a variety of lab experiments to explore why this is so. In one, for example, they had volunteers answer general knowledge questions similar to this history question. They discovered, first, that people generate anchors in different ways. Usually, as with the 1776 example, they have a fact stored in their brain that they sense is vaguely connected to the question at hand. If they have nothing in their memory to anchor on, they simply guess.
What's interesting is that even if the anchor is an arbitrary guess, people still adjust around it. A person's anchor exerts a kind of cognitive drag on the mind as it tries to adjust, so that the farther one gets from the anchor, the less plausible it feels to keep adjusting outward. Scientists label this "satisficing" -- as in, the answer is satisfying enough to suffice. Scientists have been trying to figure out why some people are better at this than others. For instance, in one study they used a personality test to sort out the most reflective students from the least, figuring that the ones who tended to mull things over would do better at making educated guesses than those who answered more impulsively. That's exactly what they found. They also asked the same trivia questions of students who had spent the day boozing, and then compared them to sober students. Or they distracted some students and not others. They found, not surprisingly, that the sober students and the less distracted students did better at satisficing. In other words, making educated guesses takes a lot of mental effort, and many things can sabotage that effort.
So back to Final Jeopardy! You're a reflective type, or you wouldn't even be up there. You sure didn't need a psychologist to tell you not to have a martini before testing your wits on national television, and the only thing competing for your attention this moment is George Washington's election. You anchor, you adjust, and you take your best shot: What is 1788?
And we've got a new "Jeopardy!" champion!
This article is adapted from Wray Herbert's book "On Second Thought: Outsmarting Your Mind's Hard-Wired Habits," published recently by Crown.
The 404 758: Where it's Wilson vs. Watson on Jeopardy 404 (podcast)
'Jeopardy' invites supercomputer Watson to destroy all humans
IBM readies Watson for post-Jeopardy life - Feb. 14, 2011
Watson (artificial intelligence software) - Wikipedia, the free ...
On Jeopardy, Watson Rallies, Then Slips - NYTimes.com
IBM's Watson supercomputer destroys all humans in Jeopardy ...
ps but as that Baseball guy said...."it`s not over` till it`s over"
I'll be glad when it's back to plain old humans. I don't want to watch an IBM commercial for a full 1/2 hour.