predictions, FTW!

In grade school, we were all taught the scientific method, right?

The idea is that you observe something, and that makes you have some thought about how it works, and that thought gives you ideas about what other things might be true that you could observe, and then you go and look for those things, thus making more observations, and the cycle continues.

All too often, though, it becomes very hard to "have the thought about how it works" (come up with a compelling and parsimonious theory), and then the next stage of predicting observations that would be true, if your theory is correct, falls by the wayside.

In a really compelling paper, Peter Lipton explains why it's not okay to just by-pass this seemingly hard step (theorizing).

His basic argument is that, once you have all the facts, you can contort a theory in any way you want to get it to fit all the data. So the fact that the theory agrees with data is not necessarily impressive.

However, if you are predicting the results of experiments that haven't been done yet, you don't have that luxury, which forces the predictions to be justified by firm logic rather than "it fits the data" (because the data hasn't been collected yet!)

After months of hard work, I have finally succeeded in making some concrete predictions for doable experiments. I will discuss this a bit more when the paper (eventually) comes out, although those who were at my talk today in the Redwood Center already have some idea.