Tuesday, September 28, 2010

What's black and white and set in front of a cheesy background? This Blog!

In my readings for my thesis I came across a joke which is doubly funny when applied to the philosophy of artificial intelligence. I plan on explaining why this is the case afterwards, so be prepared to not find it funny in about five minutes. So hear it goes:

A man is about to get married and decides to go in and buy himself a tuxedo. Not having a whole lot of money, he goes to the cheapest tailor he can find. So after paying for everything he puts on his terribly fitting suit. He tries to complain to the tailor but the tailor just tells him the suit is fine, it's the mans terrible posture that's ruining things. The tailor instructs the man to bend over a little and stop bending his knee so much when he walks and to hold his arm a little bit more like so, and sends the man on his way. Walking down the street like this, a passerby says to a friend "Isn't that the most unfortunate case of physical deformity you've ever seen?" The friend replies "Yeah, but the suit fits nice".

Now this is funny. For reasons I hope are clear because I'd hate to ruin the simple level at which this joke works (though I'm still willing to sacrifice the deeper level at the drop of a hat). The deeper funniness has to do with a few different concepts of Artificial Intelligence: substrate dependence being the one I'm going to go into (out of want to not overwhelm anyone).

So basically, to get why an AI nerd, such as myself would find this funny I need to tell you about a longstanding debate between the two sects of AI theory.

In the red corner is Good Ol' Fashion AI or Strong AI. These guys are the ones who think that we should make human machines who fulfill their function in the same way humans do. In the blue corner you have New Fangled AI or weak AI. These guys think that AI should be focused on creating any functioning intelligent systems, regardless of whether or not they resemble, at the command level, organic intelligent systems, so long as the outcome is positive.

For example, in the article where I read the joke the first time, the author(s) were discussing a super computer that played chess. Organic systems (people) use a cognitive ability called chunking to pick up patterns in chess arrangements. The super computer did not use this technique. Instead it prioritized "smart" moves and looked 14 to 15 moves ahead and used a complex algorithm to choose the best path. Doing this got that super computer to an expert chess grandmaster level who trailed by two hundred or so points behind the number one ranked chess player in the world. This is a weak AI system because it doesn't seek to perform the task as a person would. The strong AI system did not have the computing abilities and tendency to simplify that the weak AI system did and achieved a score of about 700 (half that of the weak AI system) though it performed complex pattern recognition.

Now for the joke (yeah, there was a joke before, remember?). So, in light of the above paragraph we can see the joke in metaphor. For a proponent of weak AI, you have a lousy tailor who represents strong AI proponents who are forcing their penniless grooms of AI systems into poorly constructed constraints that do little to fit how the AI system best operates. And though you can compliment how well this system can act within a set of weird senseless rules, you're still looking at a perversion of what could very easily be a system that functions better in a different format. When the chess computers tried on the wares of strong AI, what they got was a sub-optimal system that 'fit the suit' just fine, but got its proverbial ass handed to it by the weak AI system. And laughter was had by all.

No comments:

Post a Comment