Mutation Rate as Search Strategy

mental-model

Variation can be regulated as a search strategy: increase experimentation when the current configuration is failing, then suppress it once conditions reward what already works.

When survival becomes difficult, bacteria do something that sounds reckless: they increase their mutation rate. When the environment improves, they reduce it again. The apparent instability is conditional rather than constant.

E1

Failure changes the value of randomness

Mutation creates new configurations, most of which may be useless or damaging. But when the existing configuration is already failing, preserving it perfectly offers little protection. More variation increases the search space—and therefore the chance of producing a variant suited to the hostile conditions. Once the environment becomes favorable, that calculation reverses: further randomness is more likely to disrupt a working configuration than rescue a failing one. Mutation rate therefore acts like a dial between exploration and preservation. Mutation Filtering supplies the essential second half of the model: variation generates candidates, while selection determines which persist.

E1

Where it shows up

Bacteria under stress

Difficult conditions correspond with higher mutation rates; good conditions correspond with lower ones. The system searches more aggressively when its present answer is inadequate and protects that answer when it succeeds.

E1

Randomness is not adaptation

A higher mutation rate does not make any particular change intelligent or beneficial. It only produces more variation, including harmful variation. Without a process that preserves useful results and removes damaging ones, extra randomness is merely extra damage.

E1

Tie exploration to evidence of failure

Choose one recurring decision and define two modes in advance: when results fall below a concrete threshold, test a deliberately different option; when results recover, stop widening the search and repeat the configuration that worked. Change the experimentation rate, not everything at once.

Episodes that teach this