Wednesday, February 16, 2011

What do political science and Watson have in common?

Will Wilkinson again does a bang-up job blogging for The Economist (though I find it disturbing to see his American English edited into British). In this post he pulls research from the blog of Jason Sorens, a polysci professor at University of Buffalo. Sorens did research on whether America fits the "exceptional" mold people as people on the right and left portray it. I find this interesting since President Obama irked the ire of the right by having a Niebuhrian viewpoint on American exceptionalism. I highly recommend reading Wilkinson's summary.

I like Sorens' work because it puts conventional wisdom on its head: Americans don't love government any less than Europeans, our government is only smaller because of our historical religious diversity. In terms of income disparity in America actually looks good when you control for the fact that we've enslaved, massacred, and quarantined parts of our population. I haven't looked at his equations or formal articles, but I'm only jealous it's a polysci person running those regressions instead of an economist. He thought outside the box and came up with results we wouldn't well predict from independent variables we wouldn't have thought to look at.

But it dawned on me last night while watching the IBM love-fest that is Watson on Jeopardy! that a supercomputer like Watson, being filled with all the data on the planet, could be used to do this same kind of research. I imagine it in two ways:
1. He runs trillions of random regressions and reports those that show up the most significantly. Of those trillions of regressions, someone would have to figure out which ones make sense or are actually related and aren't just noise.
This is, in a sense, like the controversial poetry written by computers. Essentially, if you have infinite monkeys on typewriters eventually one of them will write Shakespeare.

2. As he learns the nuances of the human language, thought patterns, and relationships Watson could eventually sort through his trillions of regressions and report the ones that make the most sense. Studies that economists and doctors and others would come up with if they had enough time to think about it. That would be huge. You could have Watson-generated and Watson-refereed journal articles. And eventually there would be no need for actual researchers, right? We just let the machines do it.

It's a scary, Wall-E world sort of thought.

No comments: