If you do experiments and you're always right, then you aren't getting enough information out of those experiments. You want your experiment to be like the flip of a coin: You have no idea if it is going to come up heads or tails. You want to not know what the results are going to be.
-- Peter Norvig, in an interview about being wrong. When I saw this, I thought it sounded a lot like entropy pruning in decision trees, where you don't even bother asking questions that won't make you update your probability estimates significantly. Then I remembered that Norvig was the co-author of the AI textbook that I had learned about decision trees from. Interesting interview.
Wow, I'm glad this kind of analysis is showing up in mainstream publications.
Norvig is describing an important insight from information theory: the amount of information you get from learning something is equal to the log of the inverse of the probability you assigned to it (log 1/p). (This value is called the "surprisal" or "self-information".)
So, always getting results you expect (i.e. put a high p on), means you're getting little information out of the experiments, and you should be doing ones where you expect the result to be less ...
This is our monthly thread for collecting these little gems and pearls of wisdom, rationality-related quotes you've seen recently, or had stored in your quotesfile for ages, and which might be handy to link to in one of our discussions.