By Stephen Baker
Picture the smart, unassuming person at a meeting, who says, “Well, I’m no expert, but once I saw this case where….” That person is doing something that until recently was uniquely human: Soft-pedaling an idea.
Humans beings can soft-pedal because we know what we know (or at least think we do). We also know what we don’t know. And then there’s this entire domain of knowledge in which we know a thing or two. That gray area in the middle is important, because that’s where we can dabble. We can come up with insights and discuss them with people who are better informed. This process widens a discussion beyond the cloistered world of experts. It can lead to insights, the generation of hypotheses, and innovation.
One of the very special aspects of Watson, IBM’s Jeopardy computer, is that it “knows” what it doesn’t know–or, more precisely, simulates this knowledge through statistical analysis. Looking beyond Jeopardy, to Watson’s career in business, this gauge of its confidence is one of its most valuable features.
Say Watson is working in a hospital emergency room. A person comes in with a combination of symptoms that no one has seen before. Someone lists them for Watson. (The machine doesn’t have voice-recognition now, but that could be engineered in a matter of days.) Watson scours its base of documents and research papers, finds various combinations of these symptoms, and lists possible diseases or disorders that the person might have. Each one is accompanied by a confidence gauge. It turns out in this case Watson has only 14% confidence in, say, lupus.
That 14% amounts to a big shrug of Watson’s electronic shoulders. It does not know and is admitting as much. And maybe the doctors have done tests and know that it’s not lupus. But maybe below Lupus, with only an 8% confidence rank, is some other disease that they hadn’t considered. It may be wrong. It may be idiotic. But it may also lead to a thought, a connection. After all, that 8% came from some combination of the symptoms that Watson found in its research. In effect, Watson–even in its ignorance–has come up with a list of hypotheses (along with pointers to its sources). If even one of these hypotheses nudges doctors toward a correct diagnosis, the machine has provided a service–even without “knowing” the answer.
When Watson plays Jeopardy on Feb. 14, 15, and 16, many will focus on the machine’s speed and precision on the clues it gets right. But its hidden strength is its knowledge of its own knowledge–including its own limitations. Like us, it’s able to hazard a guess and, equally important, label it as such. That’s one area where Watson stands out.
Stephen Baker is author of Final Jeopardy–Man vs Machine and the Quest to Know Everything.