The success of evolution by natural selection tells us that one can optimize one's response to an environment whose properties aren't understood, by simply trying everything and noticing what works, what prevails. Genetic programming is the computer-science version of natural selection. And A/B testing is a simple version of genetic programming.

http://en.wikipedia.org/wiki/Genetic_programming

> This article appears to push people to consider beyond that: new things.

Yes, but the issue isn't new things, it's which new things. For that, you need testing. If people were either rational or predictable, this argument would fall apart. But they aren't.