|
Posted: Feb 10, 2011 |
[ # 46 ]
|
|
Senior member
Total posts: 974
Joined: Oct 21, 2009
|
That’s what I’m thinking also Jan.
Incredible speed and incredible data correlation engine though.
But to determine sentence meaning on its own, I don’t think it has, I agree. Powerful, but narrow AI.
Perhaps a great first step though.
Well, I say incredible speed, but when I think that it is massively parallel with 3,000 cores, I question how well they have coded there algorithms But then again, this is more of a proof of concept I guess, they would probably optimize later. I know myself, I do everything in Perl first, then convert to low level languages. I use a lot of string manipulations, regexes, or whatever, just to get a concept working, then go low level.
|
|
|
|
|
Posted: Feb 10, 2011 |
[ # 47 ]
|
|
Senior member
Total posts: 697
Joined: Aug 5, 2010
|
When I hear this guy talking about ‘more is better’ with respect to algorithms, my guess is that they use lots of different methods to calculate a result. All these calculations are done in parallel, all accessing the DB. Maybe they even split up each algorithm over multiple data packets (they have enough cores and no need for writing). As a final step they will probably have some kind of weighting mechanism to pick 1 result (maybe it’s as simple as the one that’s most often returned by the various algorithms). That’s just a pure guess though.
|
|
|
|
|
Posted: Feb 10, 2011 |
[ # 48 ]
|
|
Senior member
Total posts: 974
Joined: Oct 21, 2009
|
Another thing is, if Watson wins, there are two factors to its advantage:
a) visual recognition - human has to convert the light hitting his eyes , into text, THEN process (you mentioned Watson is spoon fed the ASCII)
b) hitting the button - there is a time between when the human mind knows the answer to when then neurons send the signal from his brain to his hand muscles.
I know this is nothing to do with the CORE idea, but even milliseconds mean a lot in a game like that!
A friend here at work says unless it reads , does visual recognition AND has a robot arm to push the button, only then is the game completely fair
|
|
|
|
|
Posted: Feb 10, 2011 |
[ # 49 ]
|
|
Senior member
Total posts: 697
Joined: Aug 5, 2010
|
|
|
|
|
|
Posted: Feb 10, 2011 |
[ # 50 ]
|
|
Experienced member
Total posts: 61
Joined: Jan 2, 2011
|
If you listen closely, the guy talking about Watson does not speak very well and tends to agree with the interviewer in many instances. I think Watson is doing more than he says but he can’t express it very well.
|
|
|
|
|
Posted: Feb 10, 2011 |
[ # 51 ]
|
|
Senior member
Total posts: 974
Joined: Oct 21, 2009
|
Yes, I observed that as well. It seemed inconsistent. I’m going to hunt for more you tube footage
|
|
|
|
|
Posted: Feb 10, 2011 |
[ # 52 ]
|
|
Senior member
Total posts: 623
Joined: Aug 24, 2010
|
It sounded to me like the guy being interviewed differentiates “Watson the Jeopardy player” and “The NLP functions that produce Watson’s knowledge base” as different programs.
Out of curiosity, what do you consider unsupervised learning, Victor? Is learning supervised when there is teacher interaction, or does supervised mean there is active algorithm manipulation by a user happening?
The early (conscious) learning experiences of human children are largely supervised, and especially so in regard to language. I have a home video of my parents pointing out various objects in our living room, asking two-year-old me to name them and describe what they’re for. When I didn’t answer, they slowly said the answer and asked again a few minutes later. I could have been a simple Hal bot for all my language capabilities were worth. Yet clearly (I hope!) toddler-me became a system capable of intelligent discourse (with the help of a good algorithm, of course—the human brain is an amazing thing).
|
|
|
|
|
Posted: Feb 11, 2011 |
[ # 53 ]
|
|
Senior member
Total posts: 974
Joined: Oct 21, 2009
|
If there is any human intervention, or a human involved in help break down a sentence, that is supervised.
apologizes, I should have used the word “assisted” instead.
Your parents didn’t guide you, or actually help you break apart sentences and tell you how the parts tie together.
So if a human has to look at a sentence, break it apart, and specify the hierarchy of goals and/or algorithms, then that is for sure assisted learning.
supervised learning (as I said I should have used “assisted”) is ok, in fact, probably the only way it can learn.
agreed, he does seem to be indicated two separate pieces of code, one for KB updated and one for jeopardy-puzzle parsing.
|
|
|
|
|
Posted: Feb 11, 2011 |
[ # 54 ]
|
|
Senior member
Total posts: 623
Joined: Aug 24, 2010
|
Okay, that clarifies things.
|
|
|
|
|
Posted: Feb 12, 2011 |
[ # 55 ]
|
|
Thunder Walk
Senior member
Total posts: 399
Joined: Feb 7, 2009
|
Not sure if it’s already been mentioned in this (or another ) thread, but there was a PBS Nova program about WATSON that you can watch online, “Smartest Machine on Earth.” 53:07
[ul][li]http://video.pbs.org/video/1786674622/[/li][/ul]
|
|
|
|
|
Posted: Feb 12, 2011 |
[ # 56 ]
|
|
Senior member
Total posts: 623
Joined: Aug 24, 2010
|
Very cool, thanks for the link!
|
|
|
|
|
Posted: Feb 12, 2011 |
[ # 57 ]
|
|
Senior member
Total posts: 971
Joined: Aug 14, 2006
|
Thanks for the link David Thunder. The world is so advanced, but unfortunately our friends from the legal departments still believe they are helping the world by organizing rights per country: :-(
Image Attachments
|
|
|
|
|
Posted: Feb 12, 2011 |
[ # 58 ]
|
|
Senior member
Total posts: 494
Joined: Jan 27, 2011
|
Yep, I got that same screen as well :(
|
|
|
|
|
Posted: Feb 12, 2011 |
[ # 59 ]
|
|
Senior member
Total posts: 623
Joined: Aug 24, 2010
|
You said it, Erwin. I’m going to try later with my university VPN. That usually gets me past these blocks…
|
|
|
|
|
Posted: Feb 12, 2011 |
[ # 60 ]
|
|
Thunder Walk
Senior member
Total posts: 399
Joined: Feb 7, 2009
|
Is it possible to embed the link here? I can’‘t seem to get it to work.
<removed embed>
See more NOVA.</p>
|
|
|
|