Tuesday, February 15, 2011

IBM

Jeopardy!

It's the game show where an answer is given: "The Father of Our Country; he didn't really chop down a cherry tree" and a short question is to be given by the contestant: "Who is George Washington?" So this involves using language, understanding and knowledge to get the solution.

IBM took the challenge of making a computer system that could compete, like a human player, in this game. It's called WATSON. It sits like a contestant and gets its input as a text file and processes it itself - no outside internet access or human intervention. And the results were amazing.

This brings an interesting question. What does it mean to understand something? Right now what WATSON is doing is looking all the word in the answer given up in its database and assigning a probabilty to the correct question. Outside the system it looks like the system understands it. But imagine WATSON played Jeopardy! in Chinese. And imagine instead of a computer running the program it was a non-chinese speaking person running the program by hand. It takes in Chinese characters as input, processes it by hand according to the program, and outputs Chinese. Would you say that the human understood Chinese?

22 comments:

  1. nice post, WATSON really is an important step to human-level AI. It is still very narrow AI, but it's ways of "understanding" language are almost revolutionary.

    ReplyDelete
  2. From watching Jeopardy last night, Watson still has a way to go...

    ReplyDelete
  3. wow thats an amazing piece of technology!

    ReplyDelete
  4. I've read about that thing. It's fucking amazing.

    ReplyDelete
  5. It's a cool demonstration of computing and programming power....but as you say it's just a series of instructions with no real understanding

    ReplyDelete
  6. Hm interesting. Philosophical question, for which I really don't have an answer.
    I would say no, although I don't really have a reason...

    ReplyDelete
  7. Heard about this, hopefully it doesn't bite us in the ass.

    ReplyDelete
  8. lol iv seen it on the show, its sick

    ReplyDelete
  9. Good insight, it seems we're at the point where AI can beat humans at specific intelligence tasks where its more useful to have a wide range of data rather than the ability to analyse. I think we're way off computers having human-like intelligence though.

    ReplyDelete
  10. I don't really think that watson understands, but it looks like we are on our way to one day creating something that does!

    ReplyDelete
  11. Yeah I was watching a special about this on PBS. I think it was NOVA. Pretty funny episode.

    ReplyDelete
  12. GIGO = garbage in, garbage out.

    ReplyDelete
  13. Fascinating stuff. In the end it's not understanding, but assigning answers based on probability.

    ReplyDelete
  14. With how intelligent programs are getting I must admit I'm not too surprised.

    ReplyDelete
  15. Clever bit of tech, but needs abit more work I think.

    ReplyDelete
  16. I have got myself an old IBM pc, must boot it up once again

    ReplyDelete
  17. interesting point. my CPU is old but is proccesing the information well

    ReplyDelete
  18. "assigning answers based on probability" is what i did through high school and college...

    ReplyDelete