Thursday, January 05, 2006
The Failure of Research
There are is no Journal of Failures, an unfortunate state of affairs as what doesn't work can provide just as much insight as what does. Of course, I have a personal interest in salvaging knowledge from failure, as at this moment my simulated annealing process is failing to converge (example results, till I delete them). I was hoping to make ICML'06 but it is unlikely I'll have results by then. Well, that's the way it goes. Research is a high risk activity. If it wasn't, it wouldn't be research! A lot of the structure of our research institutions are designed to reduce the risk of failure. Even if your experiments don't work, you'll still get paid. You might not get a promotion, but that is preferrable to being destitute. This structure has been very successful. Thousands of people now spend their careers as researchers, compared to handfuls in previous centuries. The resultant benefits to society have been vast. All up it's a pretty good deal. However, decreasing risks necessarily involves also decreasing potential gains. I wonder if Universities lose out on the best and brightest due to this policy?