Naked Science Forum

Non Life Sciences => Geek Speak => Topic started by: CliffordK on 09/01/2011 02:31:56

Title: Will Watson Beat Jeopardy, Feb 14, 16?
Post by: CliffordK on 09/01/2011 02:31:56

At risk of being ostracised for starting a post with a few links... 
I heard on NPR today about a new "Challenge".

http://www.dailyfinance.com/story/media/ibm-watson-jeopardy-jennings-rutter-computer-robot-trivia/19760176/

http://www.ibmwatson.com

While on the surface, it may not sound like much.  I think the computer system is understanding natural language, responding with natural language...  (I believe in Real Time).

But, unlike a "Google Search", it has to come up with a 1 or 2 word answer to any question it is presented with.

Beyond a simple game, you can imagine the implications for future search engines...
Google / Ask Jeeves, etc. 

I frequently space out on a simple word...  so then I'll throw clues at a search engine...  hoping to find what I want.  The difference between a "natural search" and Jeopardy is that often one's clues are even imprecise. 

It turns out that Google does amazingly well at some things.....  if it could only generate 2 word answers. 

For example, I can ask "What was the name of the WWI invasion in New Zealand", and the first word that shows up on my screen is Gallipoli.  Whether or not I am in the right country, that is the answer I was looking for. 

Unfortunately, one is often wading through pages worth of gibberish too...
A Google search for "How does Voyager Get Power" will eventually lead to a Wikipedia which will give a few clues... (RTG) and a few more links and one might arrive at ²³⁸Pu, but it requires several steps by the person.

Perhaps a real-world application would be when one is trying to hunt for a sensor that is causing your car to act up.  Giving a year/model of the car, but also seeking info for a broader application. 

Or trying to determine what is causing your computer to go haywire. 

One of the recent applications on Google is how it is now returning search results before one even finishes the question...  which would be useful for the new Watson (except that on Jeopardy, often the last word is vital for finding an answer).

What do you think?
Will the computer beat the humans?

Title: Will Watson Beat Jeopardy, Feb 14, 16?
Post by: Geezer on 09/01/2011 05:14:22

Not a snowball's chance in hell. (Why do I seem to have an obsession with snow???)

The only way to do it is to build a better brain. That means building a synaptic processing model with even greater capacity than a human brain, and trying to do that, even with a large array of digital computers, won't even come close.

Human brains are massively parallel. Digital computers are pathetically serial.

There was a case recently where a woman got a Jeopardy question with only one letter! Good luck Watson.

Title: Will Watson Beat Jeopardy, Feb 14, 16?
Post by: CliffordK on 09/01/2011 05:53:19

Quote from: Geezer on 09/01/2011 05:14:22

Not a snowball's chance in hell. (Why do I seem to have an obsession with snow???)

Did you have a similar prediction on computers reaching the Chess Grand Master level?

(which, of course, would be neatly in the realm of computers, try a lot of options, and build a sophisticated algorithm to predict the best choice).

Title: Will Watson Beat Jeopardy, Feb 14, 16?
Post by: Geezer on 09/01/2011 06:34:53

Quote from: CliffordK on 09/01/2011 05:53:19

Did you have a similar prediction on computers reaching the Chess Grand Master level?

Actually, no. It always seemed to me that chess can be reduced to a serialized algorithmic process that digital computers (given reasonable resources) will be better at than humans. The human brain excels at making connections between apparently disparate data in a way that no computer has emulated.

Title: Will Watson Beat Jeopardy, Feb 14, 16?
Post by: Geezer on 14/01/2011 00:44:01

Uh oh! Perhaps I spoke too soon  [;D]

http://news.yahoo.com/s/ap/us_man_vs_machine

Title: Will Watson Beat Jeopardy, Feb 14, 16?
Post by: CliffordK on 18/02/2011 07:27:36

Well....
Apparently Watson made a few blunders, but in the end won the competition.

http://www.cnn.com/2011/TECH/innovation/02/17/watson.silverman/index.html
http://blogs.forbes.com/carolinehoward/2011/02/17/watson-jeopardy-ibm-game-show-reality-tv-smart-emotional-iq/
http://www.huffingtonpost.com/2011/02/15/watson-final-jeopardy_n_823795.html

Title: Will Watson Beat Jeopardy, Feb 14, 16?
Post by: Geezer on 18/02/2011 07:47:23

I was expecting that  [;D]

Actually, I think I got confused between "Jeopardy" and "Wheel of Fortune". I wonder if they'll put him on that show too?

Title: Will Watson Beat Jeopardy, Feb 14, 16?
Post by: CliffordK on 18/02/2011 10:30:49

I doubt it.

Assuming one didn't care about the language & wheel spinning interface, an electronic Wheel of fortune player would be too easy, and might be ideally suited due to the fill-in-the-blank method.  All one would need was a good dictionary and phrase book.

Title: Will Watson Beat Jeopardy, Feb 14, 16?
Post by: Geezer on 18/02/2011 18:06:03

I was thinking of the case recently where a woman got the answer with only one letter. Not a bad trick!