As I mentioned last week, I recently sold the proposal for my next book. Here's today's announcement in Publishers Weekly. I'll be working with Amanda Cook at Houghton MIfflin, who edited The Numerati. And I'll be focusing on the IBM team that's building a computer (named Watson) to take on human champions in Jeopardy. This involves teaching a machine to understand incredibly complex and puzzling clues in English, go on a mad hunt through petabytes gigabytes of documents stored locally (The machine contestant, just like the humans, must carry its information in its own "head," and can't go online.), come up with a list of "candidate" answers, determine which one merits the most confidance, and then decide whether to buzz against the humans. This must all happen within about three seconds.
Here's an example of the type of Jeopardy clue Watson must decode: "This is the third time Sec. Chase has submitted it. But this time I'll accept it."
The answer, phrased as a question, is "What is his resignation?" Now, perhaps Watson figures out, working with statistics, that Secretary Chase is Lincoln's treasury secretary, Salmon Chase, who resigned from his cabinet. Perhaps it also sees that the verbs "submit" and "accept" are often matched with "resignation." With current search technology, it would be relatively easy to point people toward Web pages where the answer may lie. But to form the specific answer, and to bet on it, is another matter altogether.
I've been watching Watson in "sparring" sessions against humans, and I find it fascinating. The challenge involves natural language processing, question-answer technology, massive parallel computing, and a showdown against humans in Jeopardy. This fits nicely into the theme I've been wrestling with for a while: What do we need to know? As machines like Watson are increasingly at our service, perhaps on cloud services through handsets, what will we need to store in our own heads? Are humans who jam their heads with facts engaging in anachronistic behavior?
Here's the announcement:
Houghton
Gets Nerdy in Two-Book Deal
Amanda Cook at
Houghton Mifflin
Harcourt bought two books last week, both from James
Levine at Levine Greenberg. In the first
deal, Cook took North
American rights to Stephen
Baker's Final Jeopardy: The Man-Machine Battle
for the Future of Our Minds. Baker (The
Numerati)
examines IBM's quest to build the world's smartest
supercomputer—in-house
the effort is called “the Watson Project.” The author, per HMH, was
given “unique” access to the company's scientists and managers, and
the effort is planned to culminate early next year with the
supercomputer, aka
Watson, taking on whoever is the reigning Jeopardy champion at the time.
Cook also took North American rights to Sebastian Seung's Connectome:
A Neuroscientist's Search for the Self. Seung is a computational
neuroscientist at MIT and, as his publisher explained, he's currently
involved
in an effort to map the brain and all its connections. HMH hopes his
book will
“do for neuroscience what Brian Greene's The
Elegant Universe did for physics.”
Below: IBM's video on the project
|