Watson on Jeopardy: Round 3
“I FOR ONE WELCOME OUR NEW COMPUTER OVERLORDS”
added to his Final Jeopardy answer
competing against IBM’s Watson
Jeopardy, February 16 2011
If the goal was to show that a computer could understand human language and retrieve the correct answers to questions as well as humans can…mission accomplished.
Watson, the IBM computer system, easily beat Ken Jennings and Brad Rutter, two of Jeopardy’s greatest.
Yes, it was closer in this second game. It even looked like the outcome was in doubt a few times.
The final score on the second day?
Ken: $18,200, Brad: $21,600, Watson $41,413
The two game totals?
Brad: $21,600, Ken: $24,000, Watson: $77,147
We got some very interesting insight into how Watson thinks.
Amusingly, he did relatively poorly on a category called “ALSO ON YOUR COMPUTER KEYS”. Each answer was a key on a computer keyboard. Of course, Watson may not have much familiarity with computer keyboards…it’s not like he types. ;)
For example, there was this question (technically, an answer):
“In 1939’s Cartoon “The Pointer”, this guy got a new, more pear-shaped body & pupils were added to his eyes.”
You might not know…but knowing it was in the Magical Mouse-tery Tour category would give a good guess.
Watson got it right, and was quite sure he was right. His top three choices:
Family Guy 3%
Okay, so these could all be cartoons…but what made Hercules more than three times more likely than Family Guy? I assume “guy” and “cartoon” connected those. But I do think Watson must have been considering the category in making his selection.
What’s next for Watson?
IBM is going to work with Nuance (the company that makes DragonNaturally Speaking and Vocalizer, the voice of Amazon’s Kindle 3) , Columbia University School of Medicine, and the University of Maryland School of Medicine to prepare Watson to serve as a research assistant for doctors.
Doctors have an incredible amount information made available to them through medical journals. That information, though, is typically narrative. While there certainly may be graphs and tables, it’s the story that’s told that matters.
Watson can read all those journals very quickly…read and understand them.
Watson won’t decide on a patient’s medical care, but will retrieve information to help the doctor (and other medical staff) to make a decision.
I guess it’s good Watson won’t be directing the treatment (after all, he thought Toronto was a possible answer as a US city, although he really wasn’t sure). I just clicked on the link for the press kit on Watson, and got this:
The page you requested cannot be displayed”
For more information on Watson and Jeopardy, you might find this book interesting:
Speaking of technology, here’s one of the remarkable things. The second game was played yesterday…and it is discussed in the e-book I bought for my Kindle today. That doesn’t mean it was written since yesterday…it’s possible the author got to see it being played. But with traditional publishing, they would either have had to wait until after the game to publish the book, or published it and then some sort of supplement (people wouldn’t have wanted to buy the full book twice). With e-publishing, it becomes possible to publish a partial book ahead of time…and then update it for your purchasers.
Watson playing Jeopardy (and winning) is an important milestone. I always used to tell people that in the near future, everybody would know everything (or at least have access to everything). However, I would say, they wouldn’t be able to tell what was important…that’s why I, as an educator, would be safe in my job.
Now, however, I’m less sure about that, at least in many circumstances. Asking a person a question and getting an answer may become like watching a power-lifting competition. It’s cool to see people performing beyond the abilities of the average person…but deep down, you know it isn’t really practical…a machine can do it better. 😉
Just kidding…I don’t just answer questions, I provide insight. That could come in a sense, though. If ten thousand people had previously asked the same question, and the answer were recorded, a Watson-level computer could scan all of those answers in seconds, find the common answers, and even evaluate the questioner’s response to the answer.
That may be the key…not to have a computer come up with new ideas, but to be able to have them retrieve the ideas humans have already developed…and judge which ones are considered “best” by respondents.
Fascinating times ahead…
This post by Bufo Calvin originally appeared in the The Measured Circle blog.