Watson

Ryan Lantzy

Ryan Lantzy

Junior
Jan 10, 2011
266
0
16
Allegheny Mountains
Has anyone been watching Jeopardy! this week? If you haven't seen or heard about it, IBM has a super computer playing as a contestant. This week features Ken Jennings and Brad Rutter, two of the most successful Jeopardy contestants to ever play the game. Their calling it the Jeopardy! IBM Challenge, and it IBMs entry is a massively parallel supercomputer designed and built by IBM named Watson (after IBMs founder).

After the first round, Watson was tied with Brad with $5000, and Ken trailed with $400.

As I sit here watching them complete the second round Watson stands at $36681 Brad $5400 and Ken $2400.

Watson has been programmed with facts and algorithms to parse natural language questions and find the best answer. He has no connection to the internet but has a butt load of storage 16 TB of RAM and over 2800 POWER7 processor cores.

There is one more day of the challenge tomorrow, if you are at all interested in computers or seeing Star Trek type technology come to life... this is your chance.

Truly amazing times we live in.

Just to give you some idea of the feat that has been reached here, researchers have recently determined that all the processors in the ENTIRE WORLD contain the computational power of about one human brain. So Watson is really the underdog in this race.
 
Last edited:
Re: Watson

I'd be interested to see what would happen though if Watson had to listen to Alex's voice instead of being fed the questions electronically in text format. Seems to me that Ken and Brad probably would be able to answer most of the questions themselves, but the computer just rings in faster than them.

It would also be interesting to know where they accumulated most of it's database content..... since it is strong in pop culture etc perhaps Wikipedia was a starting point for them.
 
Re: Watson

Jeff Babcock said:
I'd be interested to see what would happen though if Watson had to listen to Alex's voice instead of being fed the questions electronically in text format. Seems to me that Ken and Brad probably would be able to answer most of the questions themselves, but the computer just rings in faster than them.
Click to expand...
Agreed, Ken and Brad seemed to have no chance unless Watson wasn't sure enough to try to answer. Watson seemed to also get credit for mispronunciations that wouldn't be accepted for human contestants. I found the amounts Watson chose to wager on Daily Doubles and Final Jeopardy and how it picked questions (or more accurately, answers) rather interesting.

Overall I was impressed, but cautiously so without knowing the details of how everything worked and what was known in advance.
 
Re: Watson

I think I saw one article on this... Seems a logical progression of applying computers to deal with the human interface.

We haven't heard much about big Blue and the chess championship for a while, but that like this is brute force computer power, mixed with smart programming.

It should remind us of how remarkable the human mind is, if they are just now getting into the same ball park competition after such a massive computer effort.

JR

Note: recalling trivia is only a small (but important) part of our meat computer's work load.
 
Re: Watson

Chess is relatively easy to map into algorithms, text queries less so, particularly when the response must be phrased as a question for the provided answer.

And you're saying recalling trivia is trivial? :lol:
 
Re: Watson

Indeed chess moves are easier to define, but except for the end game, a brute force analysis of all the possible end point combinations is not that useful. Not only do you have the universe of moves by white, but the same universe of responses from black, so a multiple factor geometric progression of possible moves to fully explore.

I suspect programming a computer to play chess at master level was very instructive into how human players actually think and weight moves.

Trivia, or information recall, seems the simpler task, with the wild card of having to understand contextual speech, but if you think about that a Jeopardy game, does use a finite subset of all conversational speech, so is probably easier than trying to manage a customer service line cybernetically.

JR
 
Re: Watson

Begins, begun, and watching this thread to see if anyone has figured out that they are , wait whats that noise ......Arrgghh!!!

Oh my name, err coincidence maybe, or.


Travis Watson

Just trying to remember all the things I'll never know
 
Re: Watson

John Roberts said:
Indeed chess moves are easier to define, but except for the end game, a brute force analysis of all the possible end point combinations is not that useful. Not only do you have the universe of moves by white, but the same universe of responses from black, so a multiple factor geometric progression of possible moves to fully explore.

I suspect programming a computer to play chess at master level was very instructive into how human players actually think and weight moves.
Click to expand...

I've only done some simple game programming in games similar to chess, but typically these types of problems are handled using recurrsion and doing exactly what you suggest is fairly tedious... exploring all the possible combinations. Depending on the the size of the problem, optimizations can be made in the depth of the exploration, 4 moves ahead, 5 moves ahead, etc. At the end of each branch in the tree of possibilities, the result is scored. A simple example would be, who has more peices on the board at the end of the move. The computer can then sum up the scores of each move down the branch and pick the route with the highest end score.

Weighting of the branches can be done as well. Maybe a branch has a one leaf with a really high score, but al the leaves near by are terribly low. The probability of hitting that leaf might be low, so the branch is "pruned." Other methods of pruning branches before they even are evaluated are useful and this is where, as you suggest, some knoweldge about "how human players actually think" might come in handy.

While the formal parts of the game of Jeopardy use a subset of speech, answer and question, this may or may not reduce it's complexity.
 
Re: Watson

I believe you must factor in the response time, human vs. machine. The simple fact that the machine can "buzz in" in a matter of milliseconds lets it get in ahead of the humans more often than not. So it becomes not a matter of knowledge and recall but one of "mechanical advantage". How fast and consistent would you be if audio signal gating had to be done manually?????
 
Re: Watson

It requires great complexity in parsing natural language albeit a limited subset of that natural language. A different problem than chess computers. Probably much more useful commercially than playing games.

I bought one of those microcomputer based chess games back in the '70s and it would allow you to set different degree's of difficulty. Not too surprising, the higher degrees of difficulty took several hours for the slow circa '70s micro to crunch though enough data to be competitive with even unrated players. Playing serious chess with a computer back then was like playing by mail, while you wait overnight for the next move.

Brute force is most useful in the end game, but I suspect most players and chess programs just memorize the shorter list of standard classic openings for the early game moves. Tree trimming, is practiced to save computer time. With modern platforms that is less needed, while the future moves in chess games, just like warfare, ultimately depend on the enemy's response.

Only when the competition strays from a classic opening, does the computer need to worry, or take advantage of a potential mistake. Note: I don't think computers worry, but they could be programmed to act like they do.

JR
 
Re: Watson

Jeff Babcock said:
I'd be interested to see what would happen though if Watson had to listen to Alex's voice instead of being fed the questions electronically in text format. Seems to me that Ken and Brad probably would be able to answer most of the questions themselves, but the computer just rings in faster than them.
Click to expand...

I didn't realize that there wasn't speech recognition involved. The whole thing was being compared to a human like interface. Well the front end of the interface is being able to understand the request for information as a human would. Complete with all the distractions and anomolies of using ears as an input device. So, yeah, that really seems like an unfair advantage.
 
Re: Watson

Lee Douglas said:
I didn't realize that there wasn't speech recognition involved. The whole thing was being compared to a human like interface. Well the front end of the interface is being able to understand the request for information as a human would. Complete with all the distractions and anomolies of using ears as an input device. So, yeah, that really seems like an unfair advantage.
Click to expand...

It's actually a disadvantage. Human contestants can read *and* hear the question. Watson can only "read" the question. They can also garner info from Alex's inflection on "clue" words and how he reads the question. Additionally, Watson can't hear when another contestant says a wrong answer and use that info to hone his choice. It actually cost him a question in the first round of the first game they played. For one question in a category about decades, Ken answered "what are the 20's?" which was wrong and Watson buzzed in next with "what were the 1920's?" The voice to text part is pretty trivial now-a-days, something nearly any phone can do. It just ads a layer of complexity that would be totally unnecessary to prove what IBM's researchers sought out to do.

Watson is playing the game blind and deaf, and basically all he can read is braille and he still kicked their asses. This is a feat like no other recorded in history, I promise you that.
 
Top Bottom