Remember Deep Blue? The computer that beat Gary Kasparov? It seems IBM’s next target might be a Jeopardy-playing computer.
Whether this computer can ever beat Ken Jennings at Jeopardy is irrelevant. If it were commercialized, this thing could change everything overnight.The New York Times article goes into it. Here’s the thing. Being good at Jeopardy requires several skills, one of which is being able to retain and cross-reference information. Watson is amazing at that. Better than a human being, right now. Second is being able to understand questions. It might be better at understanding a tricky question than my two-year-old son, but not much. It’s better than any other computer I’ve seen.
When I played the demo hosted at the New York Times, I won, but it came down to the last question. Mostly it came down to the questions that included puns and, let’s face it, misuses and abuses of language.
But in the real world, we don’t ask questions like Alex Trabek does on Jeopardy. At least we don’t if we don’t want our colleagues to hit us with a broom. And in the real world, we don’t mind re-phrasing a question when we have to, if it gets us better answers.
The article in the Times cited a possible application. Feed Watson all available medical journals and textbooks. It could then dispense medical advice. But would a surgeon trust it when seconds count?
I think that’s the wrong question. In trial runs playing Jeopardy, Watson isn’t at its best when seconds count, which is why Ken Jennings will probably beat Watson every single time.
But imagine situations where there’s lots of available time. A patient is describing symptoms. Enter the symptoms into Watson. What does Watson think? But more importantly, why does Watson think that? Watson should spit out the opinion and the articles that led it to that conclusion. Let the doctor read the articles and come to a reasoned conclusion.
What about when seconds count? Run drills through Watson when seconds don’t count, so doctors can practice their imprecise science and get better. Don’t rely on the technology directly when seconds count–rely indirectly instead.
But doctors aren’t the only ones who can benefit from Watson. I once worked someplace that referenced every shred of data it had through a search engine called htdig. It was next to useless. It could give me a list of documents that contained words I was looking for, but had no way to rank them. It was marginally better than connecting to a file server and using FIND or FINDSTR or grep from a command line. Which was something that’s worked since at least 1990, possibly longer.
Today I work someplace that has a Google search appliance. It’s marginally better than htdig. But not much. When a complicated question comes across my desk, I still spend 8 hours digging through semi-relevant documents in search of an answer.
Watson provides a different approach. Ask Watson how far apart two computers have to be in order to avoid TEMPEST, by policy. Because of its ability to link related concepts, it would be able to spit out an answer, and an excerpt from each document that led it to believe that. A question that takes me hours to answer (unless I know it off the top of my head) takes minutes to answer instead.
Even when Watson is wrong, it’s still useful. It got that opinion from somewhere, right? Read those documents. It could be the problem is that the available documents contradict themselves. So Watson could expose holes in policy and/or technical documentation that nobody is aware of.
The problem with the Information Age is that humans now are burdened with information overload. There’s too much useless information out there. A technology like Watson offers the possibility of filtering through all the noise and showing us what’s relevant. And, used creatively, it could tell us what we know but forgot to write down anywhere.
At first the idea of a computer capable of making decisions and beating Ken Jennings at Jeopardy scared me. And it probably should. But that’s not what Watson is. It’s not good enough right now to do either of those things, and, frankly, I think morally we shouldn’t make a machine and put it in charge of making life-or-death decisions for us.
But it’s good enough to change the world right now. So I think it needs to be commercialized, however that looks. One of the problems is cost, since it requires $1 million worth of hardware to run on.
Offer it as a $10 million box for governments and huge companies to use to untangle their mess of documents. The U.S. government should be clamoring to feed all it knows about Pakistan, Afghanistan, al Qaeda and Osama bin Laden into it, then ask where Osama bin Laden is, if only to see what answer it gives. It may not be able to answer that question, but I’ll bet it could answer lots of other important ones.
Feed the entire contents of The New York Times into it and charge a subscription to ask it questions. I’m sure Google could find a way to commercialize it by feeding the contents of Google Books into it.
For that matter, IBM could feed the documentation for all of its products into a standalone instance of Watson, and call it a technical support site. In reality it would just be the world’s foremost expert on AIX, DB2, Tivoli, Lotus Domino, and whatever else IBM owns these days. Why would I ever spec a competing product when I could ask IBM any question and get really good answers in seconds?
I hope IBM realizes what it has here. I really hope IBM realizes what it has. But I fear it may not.