GCN LAB IMPRESSIONS

Watson's 'Jeopardy!' win impressive, but was it set up to win?

Watson, get in here, I need more from you!

So the big news this week in the land of computers is that IBM’s Watson beat two accomplished “Jeopardy!” champions at their own game. Even more impressive was the fact that this was a two-game competition played over three days and the computer won both games.

You would think that a techie like me would be overjoyed at seeing a glimpse into the rise what show loser Ken Jennings called “our new computer overlords.” But actually I was disappointed by the whole dog and pony show, which was set up in a blatantly unfair way to favor the computer.

Although Jennings did well (I think the other contestant, Brad Rutter, played a bad game), there was never any question who would win once I understood the rules – which both “Jeopardy!” and IBM were pretty vague about.


Related coverage:

IBM’s Watson vs. the human brain: Tale of the tape

IBM’s ‘Jeopardy!’ match more than game playing


Now, first off, I will say that this is a great moment for computers. It really showed what they can do in certain circumstances. But would we be similarly impressed if a desktop PC running Microsoft Excel could crunch 50 pages of numbers faster than a math whiz? Because it could.

The point of a game show, or any contest really, is that all contestants are on the same footing and the best person, or computer in this case, wins. But Watson had an unfair advantage: It was being fed the questions electronically.

I wanted to see Watson hear the questions using speech recognition and process them the way humans do. But Watson was instead fed the words that made up the question in ASCII text and then went about searching a database, albeit a good one, looking for patterns and coming up with the proper response. All very rudimentary work for a computer, actually, and not much different than what Google and Bing do everyday right now.

The fact that Watson had a buzzing device is irrelevant. It already knew how it would answer before the question was finished being read, and the humans were still gathering input. And considering that the questions on last night’s show were actually pretty easy for “Jeopardy!” and that Jennings and Rutter obviously knew most of the answers, what Watson really excelled at was buzzing in faster than the humans.

If you truly put Watson on the same footing as humans, with no special input allotted for it, I think Rutter and Jennings would have smoked it. Watson would still have a slightly better chance using those rules than, say, beating a human at a game of basketball, but the “Jeopardy!” champs would have eaten it alive.

I’m not really upset that the computer won. It’s kind of cool, actually. But I expected more from Watson. I thought perhaps we would see real human-like intelligence from a computer, which includes understanding context and natural language.

When you feed ASCII text to Watson, it can only look for patterns in the stream of data. When it receives the pattern “Tennessee Williams” in ASCII, it’s receiving just a string of ASCII bit patterns. Watson has never experienced a play, or even seen people acting. It doesn’t even know what a play is. It can only guess at the best answer based on the input characters without any meaning attached to them whatsoever.

Now if a computer instead was fed data through a microphone or a camera, it could theoretically learn from the patterns it experiences if programmed to do so. It might even start to understand context. What is sentient intelligence, really, other than recognizing patterns and context and then doing things that an organism “likes” while avoiding patterns that it “doesn’t like?”

Although I don’t know if what that system would develop would be real intelligence, that’s the only true path to making computers more like humans, and thus better able to understand us. Watson has none of that true intelligence. It’s just a dumb machine whose feelings won’t be hurt in the least if this column is broken down into ASCII code and fed into its database.

When Watson wins “Dancing With The Stars” or even “The Amazing Race,” I’ll be impressed. Winning “Jeopardy!” under the rules of that last game is simply doing the same brainless tasks computers do everyday anyway.

About the Author

John Breeden II is a freelance technology writer for GCN.

Reader Comments

Mon, Feb 28, 2011 Martial Artist Seattle

Thank you for answering a question that occurred to me when they first stated that Watson was fed the question as an electronic file. I think a simpler but viable alternative to your solution of making Watson depend on speech recognition would have been to do some preliminary testing of how long it took Watson to parse the Answer as opposed to how long it took the human opponents to do the same (while not waiting for Alex to finish reading the clue). Some reasonable statistical measure based on length of Alex's answer could then have been translated into a time delay during which the human competitors would be reading what they see, at the end of which time delay the file would have been input to Watson. That would have made for a fairer game, even if not perfectly fair.

Fri, Feb 25, 2011 Josie Not at IBM

All IBM did was design a computer that could win at Jeopardy. So what? This is the same thing they did with Deep Blue and chess. They created a computer with a specific purpose and then it did better than a human at that purpose. That's what computers do. But its still a dumb machine. IBM could create a computer that could win at The Price is Right too. It would be filled with all the price data for every product in the world. Then when you feed "Ford Explorer" into its database, it would spit out $18,983.12 and bid better and quicker than any human. And it would not be a big deal, just like Watson is no big deal either.

Thu, Feb 24, 2011 Andrew New Jersey

Watson is correct that "Toronto is a US city". Toronto, Ohio is about 50 miles West of Pittsburgh, PA. Watson hasn't learned that "Toronto, ON" is an especially big, important city. There are many US cities that share a name with Canadian ones. Personally, I hear "London" and think of the UK captiol, not London, ON. Social context is a very hard rule to apply.

Tue, Feb 22, 2011 CSK Denver, CO

Sorry if this has already been asked and I did not read all the comments - when did Watson get the question fed to it? Is it at the point where everybody sees the question, and Alex Trebek starts reading it? Or is it some time after, like when Alex Trebek finishes reading? It seems that it is unfair right there if the computer gets fed the question just when the question is shown to the other contestants, as it takes no time at all for it to "see" the question.

Mon, Feb 21, 2011 Psmith

If the question is; "Can a machine accomplish a task more quckly than a human being?" Well, that question was answered long ago. So, no news there. If the question is, "Can a computer 'out-think' Jeopardy contestants, then that is another question. I believe that the contest would have been more informative if Watson were constrained to OCR, and more importantly, if Watson were constrained to an average button-push time, i.e., a period established by prior averaging of the elapsed time taken by humans. Then we would be able to reach a conclusion regarding the efficacy of Watson's 'reasoning'.

Show All Comments

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above