Stephen Baker

The New Yorker's Adam Gopnik includes Final Jeopardy in his essay about what constitutes human intelligence. I got word last week that the magazine needed a copy of the book for fact-checking. So I was wondering what it would be. But when the mail came, my wife threw me a couple of bank statements to distract me and started reading the New Yorker herself. When she started reading aloud the sentences about my book, I hurried over. But she said, "Wait, I'm not finished yet...."

***

Other odds and ends...

Watching college basketball over the weekend, I couldn't help but notice a dominant new trend in basketball vocabulary. Tall players are now referred to, time and again, as "long." Arizona, we're told, struggled with Connecticut's "length." Now long has long had some relation to tall. We talk about people being "long and lanky." But for the most part, in the language I grew up speaking... Well, I'll illustrate it.

Long

Tall

Tall on the basketball court, long while sleeping

2 comments [view]

Freedom ... and a Watson Webcast

March 22, 2011

For the first time since July of 1985, I'm unemployed. It might sound better to call myself a freelancer, or an author. I'm both. But the fact is, I don't have a book contract. The last time I was this free was in the summer of '85, when I scoped out a possible job at an El Paso newspaper and promptly took off with a friend for a bicycling trip down part of the West Coast, from Portland to Napa Valley. We took a leisurely pace, about 50 to 60 miles a day, sometimes less, and we camped out in state parks.

Jedidiah Smith State Park, Cal., one of our stops as we rode south

That trip did wonders for me. Maybe it's just a coincidence, but within a month of the trip, I'd talked my way into a job in El Paso, and by the next spring I was dating my wife-to-be and negotiating my dream job for BusinessWeek in Mexico City. I think I was in the right frame of health and mind.

The question for me is what to do with this free time now. There's plenty of reason to start applying for work or drawing up book proposals immediately. The economy stinks, I have mortgage payments and college bills I didn't have 26 years ago, etc etc. Still, this is a chance I haven't had for ages. That's what I'm grappling with with these days. I'll welcome all ideas. I just read with interest, for example, a blog post about David Lynch's push on meditation. Hmm. I thought, I probably have time for that...

***

That said, in a couple of hours, 2 o'clock eastern, I'll be on an Information Week Webcast with IBM's David Ferrucci and three of his colleagues talking about Watson's post-Jeopardy career. If you have the time, log on.

Digital Music Insider takes a look at... Watson

March 17, 2011

Jack Isquith, the savviest music industry analyst around, got curious about the next generation of computing--and called me up for an interview on his Digital Music Insider blog. We ended up talking about "death panels," among many other topics.

And in what may be a sign of the book's convergence appeal, the Gamer Vortex site gave it a thoughtful review. Many thanks.

Watson and the Barometer

March 14, 2011

An op-ed I wrote that runs in today's Wall Street Journal.

In the weeks since IBM’s computer, Watson, thrashed two flesh-and-blood champions in the quiz show “Jeopardy!,” human intelligence has been punching back—at least on blogs and opinion pages. Watson doesn’t “know” anything, experts say. It doesn’t laugh at jokes, cannot carry on a conversation, has no sense of self, and commits bloopers no human would consider. ( Toronto , a U.S. city?) What’s more, it’s horribly inefficient, requiring a roomful of computers just to match what we carry between our ears. And it probably would not have won without its inhuman speed on the buzzer.

This is all enough to make you feel reinvigorated to be human. But focusing on Watson’s shortcomings misses the point. It risks distracting people from the transformation that Watson all but announced on its “Jeopardy!” debut: These question-answering machines will soon be working alongside us in offices and laboratories, and forcing us to make adjustments in what we learn and how we think. Watson is an early sighting of a highly disruptive force.

The key is to regard these computers not as human wannabes but rather as powerful tools, ones that can handle jobs currently held by people. The “intelligence” of the tools matters little. What counts is the information they deliver. In our history of making tools, we have long adjusted to the disruptions they cause. Imagine an Italian town in the 17th century. Perhaps there’s one man who has a special sense for the weather. Let’s call him Luigi. Using his magnificent brain, he picks up on signals—changes in the wind, certain odors, perhaps the flight paths of birds or noises coming from the barn. And he spreads word through the town that rain will be coming in two days, or that a cold front might freeze the crops. Luigi is a valuable member of society.

Along comes a traveling vendor who carries a new instrument invented in 1643 by Evangelista Torricelli. It’s a barometer, and it predicts the weather about as well as Luigi. It’s certainly not as smart ~~as him~~, if it can be called smart at all. It has no sense of self, is deaf to the animals in the barn, blind to the flight patterns of birds. Yet it comes up with valuable information.

In a world with barometers, Luigi and similar weather savants must find other work for their fabulous minds. Perhaps using the new tool, they can deepen their analysis of weather patterns, keep careful records and then draw conclusions about optimal farming techniques. They might become consultants. Maybe some of them drop out of the weather business altogether. The new tool creates both displacement and economic opportunity. It forces people to reconsider how they use their heads.

The same is true of Watson and the coming generation of question-answering machines. We can carry on interesting discussions about how “smart” they are or aren’t. But it’s academic. They make sense of complex questions in English and fetch answers, scoring each one for the machines’ level of confidence in it. When asked if Watson can “think,” David Ferrucci, IBM’s chief scientist on the “Jeopardy!” team, responds: “Can a submarine swim?”

As these computers make their way into law offices, pharmaceutical labs and hospitals, people who currently make a living by answering questions must adjust. They’ll have to add value in ways that machines cannot. This raises questions not just for individuals but for entire societies. How do we educate students for a labor market in which machines answer a growing percentage of the questions? How do we create curricula for uniquely human skills, such as generating original ideas, cracking jokes, carrying on meaningful dialogue? How can such lessons be scored and standardized?

These are the challenges before us. They’re similar, in a sense, to what we’ve been facing with globalization. Again we will find ourselves grappling with a new colleague and competitor. This time around, it’s a machine. We should scrutinize that tool, focusing on the questions it fails to answer. Its struggles represent a road map for our own cognitive migration. We must go where computers like Watson cannot.

Why Watson will be smarter post-Jeopardy

March 11, 2011

A lot of people assume that Watson, a machine engineered to play Jeopardy, might look foolish once it switches to a different job. But I'm thinking now that it might be smarter off the game-show circuit. The reason: On Jeopardy, Watson was on its own. When it didn't understand a clue, it was in trouble.

But in the real world, Watson-like machines will be able to ask questions--teaming up with humans, not confronting them. This is crucial, because the hardest thing for a machine is to capture the context of a question. An immense amount of Watson's Jeopardy processing went into what computer scientists call "disambiguation." Did the two words "George Washington" refer to the president, the bridge, the university, or perhaps the botanist George Washington Carver? Sometimes the grammar makes this clear. Sometimes not. Often Watson's confidence in a response was lower than it should have been because it was processing streams of contradictory evidence.

But in the work place, Watson will be able to work with with the human mind, which will help it greatly. In this collaboration, its specialty will be its range. Only a machine can it races through gigantic databases and millions of articles and Web pages. And it will count on humans, with our superior language and contextual expertise, to guide it.

I got a glimpse of this next-gen Watson yesterday, while taping an Information Week Webcast at IBM (to be shown Mar. 22). David Ferrucci, the chief scientist on the Jeopardy team, pulled out his laptop and showed me a prototype of a medical application for the computer. In just a couple of months, IBM researchers filled Watson with medical data and trained it as a diagnostician. (No doubt its education is at an early stage. But what interested me was the back-and-forth between the computer and the doctor. Starting out, the doctor enters a number of symptoms, and the computer comes up with a 78% confidence in one disease. (I couldn't write them down quickly enough to name them here.)

But this is where things diverge from Jeopardy. The computer asks for more information. When it learns, for example, that the patient lives in Connecticut, its results change. Lime Disease, which is all too common in Connecticut, jumps to the top position. The doctor can keep refining the information given to the computer, and get more targeted results. In a sense, this is what we do when we add words to our search queries on Google. But while a search engine points us toward relevant documents, Watson will "read" those documents on its own, come up with conclusions--and then point us back to its sources if we want to retrace its analysis.

In its next career, as I wrote recently in the Boston Globe, Watson role as a tool will be much clearer than it was on Jeopardy. But that tool will be turbocharged with human brainpower.

9 comments [view]

When Watson goes to work

March 4, 2011

I wrote this piece for Harvard Business Review blog. It talks about how people are focusing on side issues--such as whether Watson is intelligent, or whether its speed on the buzzer was an unfair advantage. But too often they're missing the big one: A machine can now answer the kinds of questions that people have long specialized in. What does that mean for us, our jobs, and our education systems?

***

Saguaro National Forest (west), March 3, 2011

2 comments [view]

Kirkus reviews Final Jeopardy

March 3, 2011

Just received an advance of this nice review of Final Jeopardy by Kirkus:

Review Date: March 15, 2011

Are you ready for machines to take over the world? How about just a game show to start with?

That’s just the scenario of BusinessWeek senior technology writer Baker’s (The Numerati, 2008) account of the difficult birth of Watson, the IBM computer that just won a championship round on Jeopardy. Cleverly, the author’s narrative works regardless of the outcome—for either way, the setup is the same: After the birth of Deep Blue, the supercomputer that beat grandmaster Garry Kasparov in a game of chess in 1997, IBM scientists set about building another machine. This one, like all machines, basically knows nothing—but, intriguingly, can approximate thought all the same. Imagine, as Baker describes it, how we might parse this clue: “This facial ware made Israel’s Moshe Dayan instantly recognizable worldwide.” You’d have to know something about who Dayan was and probably have been around in the day when the monocular Yul Brynner look-alike walked the earth, whereas Watson would merely go through millions of iterations of binary data by way of a process that, as Baker notes, is “scandalously wasteful of computing resources” to arrive at the correct answer: eyepatch. Scandalously wasteful, perhaps. But imagine a few generations down the line, when Watson will have spawned machines that, to name just one real-world application, can store the texts of every medical-journal article ever written—weighing the newer ones more favorably than those from, say, Victorian England—to aid diagnosticians in their work. But how to get the machine to be able to parse real-world data and skirt the shoals of puns, subtleties, metaphors and all the other tricks human language allows? There’s the rub, and Baker provides a fine, often entertaining account of the false steps that led Watson, ever the literalist, to read Malcolm X as “Malcolm Ten” and to confuse Charles Dickens’s Oliver Twist with the Pet Shop Boys.

Like Tracy Kidder’s Soul of a New Machine (1981), Baker’s book finds us at the dawn of a singularity. It’s an excellent case study, and does good double duty as a Philip K. Dick scenario, too.

***

I write this from Tucson, where I've come to tape the keynote for the DBTechCom virtual conference, Apr. 20-22. This is the first time I've been to this part of the world since I spent a couple weeks here as a 14-year-old. It's changed a bit. But I have a couple of snap shots from back then that look like this hipstomatic view from the hotel: parked cars, a building, mountains in the background.

Todd Alan Crain: Watson's Alex Trebek

March 1, 2011

One of pleasures of trundling up the Garden State Parkway, across the Tappan Zee Bridge and up to IBM Research for Watson's practice games was watching Todd Crain, the youthful stand-in host, do his take on Jeopardy's Alex Trebek. Crain, an actor and comedian who works on The Onion, understood the importance of this role. The games had to played according to strict Jeopardy rules and procedures, so that IBM could establish scientific data on Watson's performance. But that didn't keep Crain from cracking jokes, most of them at the poor machine's expense.

Crain (right) with Jeopardy's executive producer, Harry Friedman

In an interview with GameShowNewsnet, Crain reveals some of the cultural stress that came with running a game show inside of scientific research operation:

After one of our first matches where we were just playing with IBM contestants, someone from IBM came up to me afterwards and said, very seriously, "Do you have to clap so much?" I realized in that one sentence IBM is not used to working with (what's affectionately called, in this business) "talent." The subject of me being overly enthusiastic and playful with the contestants was a little off-putting to the IBM team. They were used to a research scientist hosting the games in front of other research scientists with research scientist contestants. That doesn't make for good TV, believe me. I made it clear that I wasn't an IBM researcher and that I was going to do my job to the best of my abilities, even if that was extremely different from what IBM had in mind. Eventually, we became VERY good friends because he saw the importance of the psychological effect I was having on our contestants and audience members.

He also discusses the dynamics of the test games, and how the humans teamed up against the machine:

With our games, something incredibly interesting happened. The game shifted from three humans competing against each other to two humans playing against a highly advanced Question-Answering System. The humans would, essentially, team up against Watson. I can't tell you how many times, during a match, I heard one human contestant say to another upon their discovery of a Daily Double, "You have to bet it all. If I don't win, I want you to." When was the last time you heard that happen on the broadcast version? The sense of "us against him" ran through every game we played.

0 comments [view]

How would Margaret Atwood fare against Watson?

February 24, 2011

I've been answering loads of questions about Watson, but here was a surprising one from Anna Sajecki, who blogged about the computer on Canada's PostCity.com. Who would win at a round of pure Canadian trivia, she asked, Watson or Canadian novelist and poet Margaret Atwood?

Had to give that some thought, but came up with this answer:

That's a tough one. I would suspect that in depth of Canadian knowledge, Margaret Atwood would clobber the computer (a dilettante if there ever was one). But I'm not so sure about her range. Is she up on the ski lingo at Whistler, seasonal festivals in Nunatsiavut, hurricanes that have battered the Maritimes? If so, she'd probably fare well against Watson — assuming she's fast on the buzzer.

0 comments [view]

Stanley Fish: Watson is simple computer program

February 22, 2011

Of all of the simplifications of Watson's question-answering methods, Stanley Fish's takes the cake. In his Opinionator column on the New York Times online, he compares Watson to the computer program that anticipates and finishes the words he is typing.

He writes: "It’s just a bigger and fancier version of my laptop’s totally annoying program. It decomposes the question put to it into discrete bits of data and then searches its vast data base for statistically frequent combinations of the bits it is working with."

I responded that Watson frequently has to carry out more sophisticated analysis:

Consider using that method for the following Jeopardy clue: "This is the northernmost of the four countries with which the United States has no diplomatic relations." Would Watson find the country in question by searching out "statistically frequent combinations" of "northernmost," "diplomatic," "United States," or perhaps lists of "four countries?" No, the program has to puzzle out the meaning of the sentence and embark on two hunts, one for the four countries in question, and a second for the one that lies farthest to the north. (North Korea)

Fish could have made his point about the superiority of the human mind without mischaracterizing the methods that go into Watson. It's worth noting that while preparing his article Fish engaged in human behavior that would be utterly foreign to the Jeopardy computer: He produced answers without researching them. Watson may on occasion be clueless, but it is never lazy.

0 comments [view]