By Steven Kurutz
Last year, writer Stephen Baker approached Houghton Mifflin Harcourt with an idea for a book about an IBM computer designed to compete against humans on Jeopardy! “I said, ‘There’s going to be a big match in January 2011 and we should really hurry and get a book out by September,” Baker recalled. “They said, ‘Forget it.’”
The publishing house said the story would be old news by the time Baker’s book was published. Instead, HMH suggested Baker report and write everything but the final chapter before the match taped so the book could be in stores the day after it aired on television. The result, “Final Jeopardy: Man vs. Machine and the Quest to Know Everything,” will go on sale in hardcover tomorrow. E-book readers, meanwhile, have been able to buy the book already; the final chapter will be beamed to them tomorrow.
Speakeasy talked with Baker about the publishing experiment.
The Wall Street Journal: How did the unorthodox production schedule influence your reporting and writing?
I had the first 10 months of the year to report 11/12ths of the book. We had to have magazine-type deadlines -– the first three chapters by now, the next three chapters by then. You’re working on rewriting chapters one through three, while reporting chapter seven and writing chapter six. We get all the stuff done by November 15th and edited by December 1st. The match was taped on January 14th and the following weekend I wrote the final chapter.
If the last chapter is going out to e-book readers electronically, couldn’t you fiddle up until the final minutes?
It’s still tied to the manufacturing and distributing process of paper because Barnes & Noble said, okay, we’ll go with this partial e-book for the Nook on the condition that you have the paper book in the stores the next day.
You published an earlier book in a more traditional fashion. Did you enjoy working this way?
I don’t expect to do a slow book again. I think this is going to be the way books are done -– in a much more timely schedule. Book authors are going to have to be faster and tied to the news in the way that magazine writers traditionally have been.
Isn’t one of the strengths of books the contemplation time they afford to writers?
Ideally, yes, you get a woman like the one who wrote “The Immortal Life of Henrietta Lacks.” She spent eight years on that book. It’s a richly reported thing and it benefits from all the gestation. But for every book like that there are many others that have long gestation periods that don’t sell very well, that can’t get space on the shelves, that don’t get media attention and can’t really finance themselves. A certain number of lucky authors or ones with a brand can say to Random House, ‘I’m going to take four years and will come back with something you’re glad I spent four years on.’ But that’s a tiny minority.
Did it matter to you if the computer, Watson, won or lost the Jeopardy match?
It doesn’t matter in terms of the march of this technology. For me, personally, I wanted Watson to win –- both because I knew the team and had worked with the team closely and consider Watson to be a human endeavor. But also I think my book will do better if it’s about a machine that triumphs. People might say, gee, machines are really going to change our world. What do they do and how did they build this thing?
Has there been a downside to the process?
This galley is much more typo-ridden than the galley for my previous book, “The Numerati.” That book went through one or two more rounds of copyediting before the galley. This galley is down and dirty.
Readers, what do you think of how Watson has performed so far? Leave your thoughts in the comments.
For more on this story go to:
What a shame to spoil the best quiz program on TV by putting in a computer. I have always thought this game was the best because of the challenges and the human beings playing.
Gives meaning to the saying, “The computer beat me in chess, but it was no match for me in kick-boxing.”
@Mike Chiarito - Exactly. The people here who talk about the buzzer or databases don’t have a clue how difficult it is just to have a computer understand the question with its puns and double meaning of some words, much less find an answer from natural language knowledge base. This is just amazing.
Anybody who thinks it’s easy I challenge you to write a program that would look at google links that you get after you type in an actual question (not words) similar to the one on Jeopardy (you can even phrase it as a question) and find the one link that actually contains the answer.
@Jeff Knudson “The only reason Watson has been winning handily is because of his extremely quickly buzzing-in capability”
If you’d actually watch it when they explained buzzing, you’d know that there is a mechanical finger that is actually pressing the buzzer which takes time. A special effort was made to eliminate any advantage with electronic buzzing. Also, if you watched today, you’d notice that there were questions where Watson had an answer but wasn’t quick enough.
@Anonymous “Computers have long surpassed humans at rote tasks. Jeopardy trivia is largely a rote task”
There is nothing “rote” in figuring out what the question is about. It’s not just about a quick search or quick calculations. Parsing a natural language sentence, figuring out what is actually asked based on context, processing the text in its knowledge base to come up with the answer, this is where the difficulty is. Jeopardy questions have puns, the words in questions and in articles have double meaning.
There are zillion things that we know that seem obvious to us which are very difficult for computers. If you watched Nova, you’d see how during practice rounds a year or so ago (don’t remember), one of the questions was “The first lady that…..”. The computer came up with the answer “Richard Nixon” because the concept of male/female which is obvious to us isn’t so to the computer. There are zillion of such cases (”He shot his wife by the river” - did he kill his wife or did he simply take her picture? We look at context, and from the context the meaning becomes obvious to us. It’s not so to the computer. There was a lot of machine learning involved.
As to neaural networks - I don’t really know if they were used, but the machine learning was certainly used during preparation. They talked about it in detail on Nova on Friday. BTW, the researcher who wrote the algorithms for the determining the bets during final jeopardy is well known in the area of neural network research. This is his article on how the bets are determined:
http://ibmresearchnews.blogspot.com/2011/02/watsons-wagering-strategies.html
You must know or studied programming, to fully apprecite the work IBM’s programmers have created. Truly an amazing technological breakthrough, this is the future.
Anyone who has ever developed or maintained software for a living has got to be absolutely amazed at the accomplishments of Watson’s programmers. I enjoyed all three days immensely. One of the biggest winners will likely be Homeland Security where there are way too much data that could ever be analyzed so far.