So, have you heard about Watson? You probably have, but if not, here's the quick rundown: IBM designed a computer to compete on Jeopardy, against two of the highest-profile Jeopardy champions: Ken Jennings, the holder of the longest winning streak, and Brad Rutter, the highest-scoring Jeopardy player ever. Watson won handily.

This whole situation had the wonderful ability of striking a chord with just about anybody using the internet - the nerds picked it up because of the AI ramifications, and the masses picked it up because it's Jeopardy. Everybody has an opinion on the meaning of the contest, and its results, and just about everybody falls into one of two camps - they either downplay the significance of Watson, or blow the significance out of proportion.

Here's what you really need to know about the whole situation:

The most common comment I see online is some variant of "so what? A computer winning a trivia contest isn't anything special." Well, yes and no.

The fact that Watson won Jeopardy is not impressive at all. After all, Watson has a massive "memory" bank and lightning-fast reflexes - probably the two most important things for a Jeopardy player. Watson can recall information in an instant, can never forget, and is immune to the variety of pitfalls that humans face: pressure, pride, frustration, etc. Any physical, mental, or emotional characteristic of humans that could possibly degrade performance has been eradicated in Watson. In that way, it is superior, and its victory really should come as no surprise.

The fact that Watson won Jeopardy is not impressive... but the fact that it could even PLAY Jeopardy in the first place is VERY IMPRESSIVE. See how I bolded it? THAT'S how impressive it is.

Watson was not just "speed Googling" it's own database of information. That is beyond an oversimplification - it's a blatant misrepresentation. Consider the following category and question (technically an answer, in Jeopardy's reverse-question format):

Category: Superstitions
Answer: You can get four of these lucky charms from one animal.

All you humans out there probably identified the correct answer as "(lucky) rabbit's foot". Of course, it's possible that you went with "horseshoe" as well, but that'd be wrong, as we'll discuss in a moment.

If Watson was merely "speed Googling", here's what it would have to work from:

Search results for the answer phrase.

The first result is about a pickup line. The second and third are about 4-leafed clovers. None of the results on the first page suggest that "rabbit's foot" is actually the right answer, and none of these would've produced that correct answer.

What Watson needed to do (and successfully did) was analyze the question, pick out key words, run multiple algorithms in parallel on these various words, analyze all the results, and determine the confidence interval for each one. And ONLY then, if the confidence interval was high enough, would it buzz in.

The way a human would approach this problem is very similar - you would think about all of the different words, and determine the key phrases ("superstitions", "lucky charms", "four", "animal" being the main ones). You would then think through EVERYTHING you know, trying to find anything that has a clear link between them all.

One piece of knowledge that you almost inevitably have (that Watson would need to "discover" with an algorithm) is that most animals have four feet. And that fact gives you an extra clue that Watson doesn't innately get - that some kind of animal's feet may be part of the answer. If you repeat the above search, with the word "feet" added to the query, you get the following:

Modified search query

Now, the top two results are for horseshoes and rabbits' feet, which is much better, but still not enough. In order to determine the correct answer from these, you (or Watson) would need to recognize that you need something that comes FROM the animal. Horseshoes are put onto horses - they don't naturally occur on them. This is another piece of information you likely take for granted, but that Watson would need to determine.

As you can see, it's more than just running a Google search, and returning the top result. The fact that we now have a program that can search through TONS of information, draw connections between them, and determine the answer with pretty good accuracy, is something to be amazed by and proud of.

However, you should go so far as to believe that Watson can actually think. That's absurd. John Searle puts it pretty well, but all Watson can do is analyze data according to predefined rules. Watson (unless it was to be reprogrammed otherwise) is incapable of giving a wrong answer just for fun. It has no capacity to decide to do that. You could, by feeding it carefully selected prompts, control its every action.

Watson would be unable to throw the competition. If the rules of the game changed even slightly, Watson would be toast - it is completely unable to adapt to changes (again, unless it was reprogrammed). As stated above, Watson can't get flustered, flattered, happy, sad, angry, or complacent. And in this way, it is a long, long cry from true artificial intelligence. It will not become Skynet. It won't refuse to open the pod bay doors for you.

The fact that Watson exists is very, very impressive. There are plans to use this technology in medical fields, allowing for a computer system that can access vast amounts of historical patient data, analyze trends, and aid in diagnoses; this will be especially useful when it comes to patients that have more common symptoms, but less common diseases.

It's important to maintain perspective in things like this. Watson is indeed a triumph, and the fact that we have this kind of technology is something to be proud of. But in the end, humans designed and built Watson. And as impressive as Watson is, we humans are still that much more impressive.

Read the rest >>