By Ari Jacobs
It was midway through Jeopardy this past Monday when Alex Trebek gave his contestants the following clue: “The first modern crossword is published and Oreo cookies are introduced.” The category was “Name the Decade,” and Ken Jennings, the multi-million dollar Jeopardy winner revisiting his glory days, incorrectly responded “the 1920s.” Another contestant picked up the question and, without hesitation, answered: “the 1920s.” A faint gasp burst from the frozen crowd. “No,” begged Trebek, “Ken said that!” The contestant did not respond — he couldn’t. Well, it can’t, because it’s not human. Its name is Watson, and it is a computer.
For the first time in history this week, Jeopardy hosted the ultimate man versus machine showdown. The humans, former Jeopardy champions Ken Jennings and Brad Rutter, found themselves squared off against Watson, a room-sized I.B.M. supercomputer. And woe it was for humankind as Watson upended his opponents. By the second day of competition, Watson had $35,734 compared to Jennings’s and Rutters’s meager $4,800 and $10,400, respectively.
So, now what? Has A.I. reached its practical potential? Are we on the verge of an I, Robot revolution? What exactly is Watson, anyways?
According to Robin Zebrowski, Assistant Professor of Cognitive Science here at Beloit College, Watson is a milestone for artificial intelligence. Watson’s sheer intellectual prowess will amaze any ordinary onlooker. Yet for cognitive scientists, Watson represents even more. Watson is the ultimate manifestation of A.I.’s top-down approach to building intelligence, the “old-fashioned” method as Zebrowski puts it. The old-fashioned method is very simple: shove as much information as you can into memory storage while maximizing computational speed. And with 15 terabytes of textual data, Zebrowski says that Watson is like “all of Wikipedia and all of the edits… something like 60 gigabytes of data compressed.”
Watson certainly seemed to answer Trebek’s translucent questions with an identifiably human intelligence. Would Alan Turing be impressed? Since the early 1900s, cognitive scientists have compared artificial and human intelligence using the eponymous Turing Test. Summarily stated, the Turing Test analyzes the intelligence of a computer by seeing if the computer can make a human believe it is itself human. Ignoring the physical limitations of Watson (which is simply a box), the input-output facet of Watson is eerily familiar. During Jeopardy, a camera display would show the top three choices Watson had picked for an answer. A “confidence level” would follow each choice, indicating the statistical probability that Watson would choose a specific answer. Is this how we define human intelligence? One could imagine that Watson deduced innumerable choices to those three, thereby copying what seems to be a humanistic, mathematical way of thinking.
But Watson’s incapacity to communicate with his contestants undermines its potential as a double of human intelligence. The reaction of the crowd towards Watson’s repetition of “the 1920s” clearly illustrates this limitation. Zebrowski notes that Watson lacks adaptability, a vital component of human intelligence. “Because Watson couldn’t hear his competitors, he couldn’t adapt.” Watson is a brain in a vacuum. It is a master of “natural language processing” as Zebrowski says, but it won’t shake your hand, and it certainly won’t have a dialogue with you about the beauty of Renaissance art. In terms of the Turing Test, Watson ultimately fails.
Now here’s an interesting thought experiment: replace Watson’s box with arms, legs and a head. Bestow Watson the body, voice and cadences of a human and repeat the A.I. experiment on Jeopardy. Would the crowd react differently then? Even without adaptability, it seems reasonable that dressing Watson with human characteristics would promote familiarity with the crowd. This begs the ethical question: how much would it take for Watson to become human? Zebrowski admits, “it’s my big worry with Watson… people might start acting as if it’s genuinely intelligent.” Anthropomorphizing Watson might “make people attribute a mind before there is a mind.” In that case, the boundary between artificial and human intelligence becomes blurred. That would make the situation much more reminiscent of an I, Robot movie plot.
Until that time comes, Watson stands as a powerful symbol of modern A.I. engineering. Will Watson be the first of many supercomputer or robot contestants on game shows such as Jeopardy? Zebrowski herself is elated by the progress that artificial intelligence has made. “I think people should realize that this is a really exciting thing that’s happening right now. After 15 years of stagnation, A.I. took a leap forward.”