First Test Of Computer Jeopardy Player Goes Well; Watson Beats Mere Humans
from the show-off dept
Last month, we wrote that the IBM computing project Watson was ready to take on real Jeopardy contestants in February of this year (just a few years after the project first came to life). While the big test isn't until Valentine's Day, apparently they had a dry run, and things are looking pretty good for Watson, who beat Jeopardy champions Ken Jennings and Brad Rutter. It wasn't a complete domination. Apparently Watson just narrowly edged out Jennings, though Rutter couldn't keep up. Of course, between now and the real test, Watson can be tweaked. Jennings' and Rutters' brains are pretty much set, and I'd imagine that their ability to cram more useless trivia in their brains is outmatched by Watson.Thank you for reading this Techdirt post. With so many things competing for everyone’s attention these days, we really appreciate you giving us your time. We work hard every day to put quality content out there for our community.
Techdirt is one of the few remaining truly independent media outlets. We do not have a giant corporation behind us, and we rely heavily on our community to support us, in an age when advertisers are increasingly uninterested in sponsoring small, independent sites — especially a site like ours that is unwilling to pull punches in its reporting and analysis.
While other websites have resorted to paywalls, registration requirements, and increasingly annoying/intrusive advertising, we have always kept Techdirt open and available to anyone. But in order to continue doing so, we need your support. We offer a variety of ways for our readers to support us, from direct donations to special subscriptions and cool merchandise — and every little bit helps. Thank you.
–The Techdirt Team
Reader Comments
Subscribe: RSS
View by: Time | Thread
Still, steps in the right direction. I wonder how long it will be before fully fledged AI computers will be interacting with us on a regular basis!
[ link to this | view in thread ]
From My Introductory AI Lecture this morning
"We will construct a program/robot that will perform task X (e.g. play Chess). The resulting program/robot will thus have artificial intelligence - since humans use intelligence to perform task X."
After project
Scenario 1
"We failed"
Scenario 2
"We succeeded - however on closer examination it turns out that the method our machine uses is really some form of brute force algorithm that can't really be called intelligent. It seems that you don't really need intelligence to perform task X after all."
[ link to this | view in thread ]
[ link to this | view in thread ]
You seem excited by this
[ link to this | view in thread ]
Re: You seem excited by this
[ link to this | view in thread ]
Re:
[ link to this | view in thread ]
[ link to this | view in thread ]
Re:
[ link to this | view in thread ]
Re: Re:
[ link to this | view in thread ]
Much more complex...
For example, a question might read: 1812
The AI would have to remember that the category was "wars in history" and then apply that to its database search.
Of course that one was simple, there are many more complex ones that I cant think of.
I do remember one category that was, in other words, "Poets with the same letters for their first and last name"
Now, unless you were expecting this category, how on earth could you program something to realize what that means.
[ link to this | view in thread ]
Re: Much more complex...
[ link to this | view in thread ]
What I'm wondering: Are they trying to anticipate common categories and programming this type of information into the computer or simply relying on some kind of intuition algorithm (?).
[ link to this | view in thread ]
Re: Re:
Computers aren't good at figuring out what information can be ignored and what words are being used in ways that aren't in the dictionary.
[ link to this | view in thread ]
Re:
There are lots of articles out there explaining why Watson was a very real AI challenge - and no they are not all just articles by IBM. You should definitely look more into it because it's actually pretty fascinating and it truly is a major breakthrough, if not in genuine "artificial intelligence" then at least in the natural language processing (which is one of the core building blocks of artificial intelligence)
[ link to this | view in thread ]
Re: Re:
[ link to this | view in thread ]
Re:
[ link to this | view in thread ]
Re: Re:
[ link to this | view in thread ]
[ link to this | view in thread ]
Re: Re:
Were the people who created the program independent of those who chose the questions? Were those who chose the questions given any guidelines as to what kinds of questions should be asked, what kinds of questions the computer has answers to, and how the questions should be worded? This test run was done at IBM's headquarters and so it was done in a controlled environment. Practically any beginner programmer can create an excellent Jeopardy contender in a controlled environment if the set of questions and answers were predetermined. We must know how prepared the computer was to answer the set of questions that were being asked ahead of time in opposed to how prepared it was to answer any random set of questions.
[ link to this | view in thread ]
Re: Re: Re:
otherwise you can come out sounding petulant and foolish.
[ link to this | view in thread ]
Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Re:
That's not what happened. And yes, the people creating the program ARE independent. Maybe you didn't notice but this is an IBM computer competing at Jeopardy - IBM does not own Jeopardy and Jeopardy has no interest in helping them win. Besides, both sides have made it clear that the challenge is not altered in any way from a typical Jeopardy match - so unless you are accusing everyone involved of lying to our faces, I have no idea what you are talking about.
This is a genuine challenge, but you for some reason insist on believing it's a massive cheat. What has given you this idea?
[ link to this | view in thread ]
Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Re: Re:
Clearly.
[ link to this | view in thread ]
Re: Re: Re: Re:
My point is that if the computer has been asked these questions before and the answers that the computers algorithms will provide are known in advanced (and the computer has been tweaked to answer these questions correctly), and the people contending against the computer have never been asked these questions, then it's not as surprising that the computer won. Sure, that still maybe a slight oversimplification, but the point is that the computer will naturally be more familiar with and tuned to answer the specific types of questions that its programmers will tend to ask along with being able to understand the specific wording that the programmers will tend to use (and respond with the expected wording).
[ link to this | view in thread ]
Re:
If you doubt please try to make a search engine that understands the meaning of words and can keep track of the many many double meaning things have.
[ link to this | view in thread ]
Re: Re: Re: Re:
Jeopardy was known to fix their games for a long time, this is history not story, so I would be careful not believing they could just lie to our faces they would if they could get away with it LoL
Other then that I doubt IBM would want that, because that system would be eventually be open to scrutiny and the truth would come out and destroy their good reputation that helps them sell billions in hardware. That is why I don't believe they are cheating here, they don't do that kind of thing because it would cost them a lot of money.
[ link to this | view in thread ]
Re: Re: Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Re: Re: Re:
And the whole concept of a "type" or "nature" of question is exactly the point here. Humans can abstract and recognize the "type" of a bunch of complex and varying things like trivia questions. To a computer, each one is a completely unique string of characters that can only be parsed with strict logical conditions - same goes for every single one of the facts it has in its database of knowledge.
This is obvious even at basic levels of computing: spellcheckers work well, grammar checkers do not. That's because spelling is a (for the most part) strictly defined set of rules - the computer simply checks each word against its dictionary, seeking an exact match. Grammar, on the other hand, is highly complex with varying degrees of strictness and countless structural options: there are "rules" but they are not nearly as simple or clear, always being full of exceptions and intricate conditions. Now, even though most people cannot explain all the rules of grammar, we instinctively know when a lot of things are right or wrong just based on how it "sounds" to us - and we can't even explain what we mean by that, or how that process works in our head. Computers have no such skill.
[ link to this | view in thread ]
Re: Re: Re: Re: Re:
[ link to this | view in thread ]
Re: Re: You seem excited by this
[ link to this | view in thread ]
Re: Re: Re: Re: Re: Re:
Many game shows have been known to do that, not just Jeopardy, though that wasn't my point here. My point here, from the article
"was a test run this morning at IBM Research's headquarters in preparation for a televised weekend challenge"
From my understanding, this wasn't done by Jeopardy, it was done by IBM. To what extent were the selected questions created by people not involved in the development of the computer hardware, software, and database info and to what extent did those people creating the questions receive guidance as to how to word the questions, how accepted answers should be worded, and what kinds of questions to ask and which ones to steer away from.
[ link to this | view in thread ]
Re: Re: Re: Re: Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Re: Re: Re: Re: Re: Re:
[ link to this | view in thread ]
Re: Re:
Another point worth making is that Watson is completely offline. While they feed it presumably more books than any human could read in their lifetime, it cannot just rely on the question having already been answered in an easy digest form on Google.
[ link to this | view in thread ]
Re: Re: Re: Re: Re: Re:
Two terms worth throwing in here for anyone wanting to learn more about the sort of processes I assume Watson uses: fuzzy logic and neural network.
[ link to this | view in thread ]
Re: Re: Re: Re: Re: Re: Re:
Why would IBM cheat on the not actual Jeopardy test? It's obvious why they wouldn't cheat: they can test it against the actual competitors and tweak the systems in preparation for the real test. Are you seriously suggesting that they're not confident of their chances in the real test and are using this trial run to make it look like they did well?
[ link to this | view in thread ]
Re:
I'm guessing that if they needed to then they could a combination. I think it is more likely that such a step would be redundant because such a thing should be easily and more effectively learned by the machine anyway, e.g. by inputting previous actual questions. If every question that had the word 'dig' in the title turns out to be about archeology then the computer will learn that.
[ link to this | view in thread ]
Re: Re: Re: Re: Re: Re: Re: Re:
What do you base that on? If their goal is to improve the system then they may well ask it questions that it is less likely to get right, a completely opposite bias to the one you assume.
[ link to this | view in thread ]
Re: Re: Re: Re: Re: Re: Re:
[ link to this | view in thread ]
Re:
[ link to this | view in thread ]
Re: Re:
I think you could make a pretty good argument that this one does. Which is what's so awesome about it.
[ link to this | view in thread ]