Can a computer be smart or can it only be programmed to act that way? (Books - Final Jeopardy by Stephen Baker)

I rolled my eyes at my first glance at Final Jeopardy: Man vs. Maching and the Quest to know Everything sent to me by Houghton Mifflin Harcourt. It concerns itself with the now well-known stunt planned by IBM to program a computer named Watson to play Jeopardy against human contestants? What silly stuff and why would I want to read about that? However, being sick enough this past week to need to stay in bed for three days, I found myself searching for reading matter that didn't require deep study, as much of my reading for school and the lab do, and that didn't ring the knell of probing profundity ( I was staying away from finishing Thomas Mann's Magic Mountain, for example, which I must do some time soon). So I picked up Stephen Baker's Final Jeopardy and it was just the ticket.

The reason you might want to read about IBM's development of Watson to pull off a televised stunt akin to the chess match between Big Blue and Gary Kasparov in 1997, is that, aside from any backstage details you might like to know like why he was named Watson or who designed his 'face,' Baker is really telling a story about the nature and uses of intelligence. Since IBM was producing so-called artificial intelligence (AI) they had to know what it was they were faking. Baker, as IBM did, used the television contest as a scaffold. For IBM it narrowed the scope of their project and imposed a schedule. For Baker it lent his narrative thrust toward a conspicuous end point. It's almost a shame that we have to know the outcome, but Baker writes well enough to squeeze suspense out of the story by connecting us to the stakes experienced by scientist David Ferrucci, and his team of programmers, designers, former contestants, stand-in game show host and, of course, the inevitable public relations armies for both IBM and Jeopardy. However, what I enjoyed most about the book were Baker's lay-descriptions of the evolution of computing machines and how they differ from human brains: what kind of knowledge goes into them, what sorts of computations can be expected of them, what kinds of mistakes they make and the learning potential (if any) derived from making them. That learning, indeed, seems to separate the men from the boys (if you'll pardon the sexist expression) in what constitutes programming talent for this kind of project.

For certain types of questions, Ferrucci said, a search engine could come up with answers. These were simple sentences with concrete results, what he and his team called factoids. For example: "What is the tallest mountain in Africa?" A search engine would pick out the three key words from that sentence and in a fraction of a second suggest Kenya's 19,340-foot-high Kilimanjaro. This worked, Ferrucci said, for about 30 percent of Jeopardy questions. But performance at that low level would condemn Watson to defeat at the hands of human amateurs.

A Jeopardy machine would have to master far thornier questions. Just as important, it would have to judge its level of confidence in an answer. Google's algorithms delivered users to the statistically most likely outposts of the Web and left it to the readers to find the answers. "A search engine doesn't know that it understood the question and that the content is right," Ferrucci said. But a Jeopardy machine would have to find answers and then decide for itself it they were worth betting on. Without this judgment, the machine would never know when to buzz. It would require complex analysis to develop this "confidence."

Baker does a good job explaining some of the theory of how semantic knowledge is stored in and retrieved from neural networks at a basic and readable level, and - key to his story - is making clear the difference between designing a system that imitates such a process in silicon versus designing one that looks as though it is performing human cognitive problem, but is actually accomplishing that differently that human neurons do.

Equally as interesting was the discussion Baker's book provokes about what knowledge is for and highlights what I came to see as the ultimate poverty of Ferrucci's tightly-focused imagination .

"You can probably fit all the books that are on sale on about two terabytes that you can buy at OfficeMax for a couple of hundred dollars. You get every book. Every. Single. Book. Now what do you do? You can't read them all! What I want the computer to do," he went on, "is to read them for me and tell me what they're about, and answer my questions about them. I want this for all information. I want machines to read, understand, summarize, describe the themes, and do the analysis so that I can take advantage of all the knowledge that's out there. We human's need help. I know I do!"

You can't get all the information and the best computer can't either David, so calm down. Knowledge is not just possessing facts, nor is it analyzing them, or summarizing them. Get the Cliff Notes if you want a summary. In fact, merely determining which units are the facts of a narrative is itself an analysis. What are the facts of Oliver Twist, David? Is Fagin a fact? Is the theft of a pocket handkerchief a fact? Is 'some more?' How about the fact that the law is an ass? Does that mean that the law actually is an ass? Which law? Or that Beedle Bumford says it is? And if he says it is - who is Beedle Bumford and why does he think so? 'Facts' and the juicy stuff that can be derived from them are determined by an intersection with the point-of-view of the individual using them. To be fair, Ferrucci understands these limitations and his project embraces the challenge of beginning to reverse-engineer just such incongruities. Ferrucci offers some of his hopes later on in the book of the practical uses of a computer's possessing what he assumes is an overwhelming body of knowledge for the human brain. For example, making (or I would suggest checking) difficult medical diagnoses against the most up-to-date knowledge in the literature or sifting through legal precedent. However, machines will likely be better at generating lists for hypothesis development than they will be at making inferential leaps. The gains made in problem solving by sudden departures from the current knowledge tree or the standard method are legendary - that's the 'creative' part of creative problem solving and is the very stuff of the 'aha moment.'

Baker makes a similar argument around his chapter on a computer called Blue J, developed by Jennifer Chu-Carroll (among others). She was trying to determining the ideal reading curriculum of her computer, and Baker cites the difficulties of interpreting idiomatic language and dialect, metaphor, and paradox concluding, somewhat tongue-in-cheek (I hope) that

...most books had too many words - too much noise - for the job ahead.

But it's the noise I like, Stephen.

Other scientists, such as Joshua Tenebaum at MIT think that one day computers will generate concepts and make inductive leaps, but its hard to imagine as Baker writes of the scientists in this book attempting to program Watson to parse the words of a single Jeopardy question well enough to determine the proper category of knowledge to search (let alone to answer it). Baker makes the journey of Watson and his creators into entertaining reading. The most indelible take-home message from this book to my mind (there's that intersection with point-of-view I was mentioning) was how much more than computing speed is necessary before computers can comprehend human speech, make inferences, and...well, you know... take over the world.
Tenenbaum said it best.

"If you want to compare [AI] to the space program, we're at Galileo," Tenenbaum said. "We're not yet at Newton."

Baker's book is thoughtful, informative, swift-moving, and really amusing, without being cheap. I'm passing my copy (thank you, HMH) along to my non-scientist Uncle who was himself a contestant on Jeopardy a few decades ago. I think he should get a kick out of it.