I'm sure that many of you watched two of Jeopardy's greatest living human players go up against Watson last night. So far, there's been a lot of differing opinions on it. Some are awed by the ability of a computer to go up against humans, in their own language, and hold its own. Others shrug and say "Get back to me when it can do something useful".
For those of you that aren't in awe, you probably should be. And those of you that are in awe, may be in awe for the wrong reasons.
Follow me after the jump, and I'll explain why.
To begin with, Watson's a pretty impressive piece of machinery. Or, more specifically, a cluster of them. It's 100 Power 750 servers, full loaded and decked out. That, plus the racks, the networking gear to tie it all together, the additional storage, etc. -- you're looking at $20 million to $24 million worth of hardware. If not more. It has some amazing computational power -- it's an absolutely massive database that it has to wrangle (and it's why humans skunk it on questions that take Alex Trebek a short time to read -- it isn't quite real-time yet).
But Watson's real promise isn't the fact that it can play Jeopardy. That's simply the flash to raise interest in the project, and it also gave the brilliant minds behind the project some focus. As anyone who has decided they want to learn to program or play an instrument knows, it's a lot easier to ,ale "I want to make a program that does this" or "I want to learn how to play Hotel California on the guitar" happen than "I want to learn to program" or "I want to play guitar".
Its real promise isn't even artificial intelligence. And if you're in awe of it because it seems like it is intelligent, or close to it, you're in awe of it for the wrong reason. Watson isn't intelligent, and isn't anything close to intelligent. Your dog is more intelligent than Watson is.
Watson is a machine designed to be good at one thing -- analyzing variables in context given a dataset of nearly infinite variables and contexts. And if that sounds like a downer, it isn't. That is something to be in awe of.
Right now, Watson gets the clue fed to it in a text file. You could make it listen and analyze sound, but you'd be massively increasing the cost, and as a proof of concept, it's worthless. We already know we can analyze sound and teach a computer what to do with that sound. That's why you have voice-activated music players in case, voice dialing on phones, speech-to-text software, etc. It's flashy, but useless for what Watson is showing off.
It also doesn't see anything, or more specifically, doesn't do any video/picture analysis. Again, this is something we can already do -- Google Goggles and other image/video analysis programs and algorithms. Could it? Sure, at a massive increase in cost, but we already know this can be done, so it didn't need to be added on to this proof of concept.
But Watson can analyze data (in this case, language), relate it to knowledge already within its database, make determinations about the context of the variable/word (run like a foot race, or run for office? Fried like an egg, or fried like a sunburn?), and not only come up with correct output, but assign a probability to it based on the knowledge it has.
Watson is a computer making an educated guess, not simply searching for a matching answer to a given problem like a calculator.
Now, how does that have real world applications?
In the short term, it's a new form of search and data organization. Doesn't seem like a big deal, but that's what Yahoo thought when a startup called Google had a new way of searching and organizing data.
In the long term? Far reaching applications. Take, for example, home care of dementia/Alzheimer's patients. The poor oldster has turned on the sink, wet their hands, put some soap on them...and has no idea what they were doing. And yes, I've seen that happen.
Watson v8.0 with contextual video and audio processing and an array of sensors in the home sees/hears/detects all of these clues, and maybe the ones before it like a toilet flush 35 seconds prior, and that the oldster entered the bathroom and headed for the toilet, and that the sink is in the bathroom, so water isn't being run to wash dishes or thaw food, and that it was cleaned recently, so that's not why the water is running, etc, etc...
...and tells the oldster "You just went to the bathroom, and you are washing your hands. Rinse your hands in the running water, and turn off the water by turning the handle."
Or it's a miniature version that can be trained to detect brainwave patterns in ALS patients (or even coma patients), and output speech. Or you can just speak into it, if you can do that. And the output is in whatever language the user wants it to be in, because it can not only analyze the brainwaves, but it can recognize slang, turns of phrase, and what have you.
That is what Watson is the beginning of. Right now, it's 100 high powered servers that cost more than most of us will ever see in a lifetime. Given Moore's Law and the exponential growth of computing power, it isn't too long before $25 million Watson is a $200 iThink.
This is a proof of concept of a large breakthrough in data analysis. It is at the pinnacle of achievement, and is an exquisitely crafted machine designed by some of the greatest minds of our time. I look at it in awe like I would a Michelangelo, or Cristo Redentor, or when I hear one of the great classical compositions.
It's not something to replace humans, but is a wonderful, beautiful creation of them to enhance the experience of humanity. Jeopardy is just the beginning.