A/B testing is the primary method by which technologies can automatically optimize themselves. In online advertising, for example, a company can run several ads at once and choose from them the single ad producing the most revenue. (I have designed several systems for optimizing this process statistically).
That process is just like natural selection (aka evolution): introduce several random versions and select for amplification those which work best. In other words, temporarily increase the entropy of the state-space, then decrease it to come closer to an optimum. In both cases, increased performance comes at the cost of decreased entropy and limited diversity.
This approach to optimization is so general that brains must use it not only in the outside world, but inside themselves in a continuous fashion. Regardless of specific mechanism, any continuous representational system must constantly balance "smoothing" (entropy-increasing) functions like mutation and blurring with "sharpening" functions like selection and amplification, resulting in an optimally smooth and simple model.