Vehicle 6

Selection, the Impersonal Engineer

Chapter six shifts the question from how a mechanism behaves to how such a mechanism could be found in the first place. Braitenberg’s answer is selection: competence can accumulate without a planner who understands the result in advance.

9 min read

Vehicle demonstrationViable body

Selection is easier to see when the same landscape is held fixed and only the body keeps proving that it can cope with it.

DemonstrationSelection can be read off the trajectory: some bodies keep finding workable paths in the same world. This one survives because its coupling stays viable.

Once the vehicles become even slightly richer, the natural question is who built them. Braitenberg’s answer is that not every good design needs an engineer who understands it as a whole. Selection can do the searching.

A population only has to vary, confront a world, and keep what already works a little better. Over time, that is enough to produce mechanisms that look fitted, even when no agent foresaw the fit.

Key idea

Selection is an impersonal engineer because it can accumulate viable structure without representing the final design as a plan.

Search can be external to the vehicle

This chapter matters because it prevents us from overloading the vehicle itself with explanation. Some of the competence we observe may be the result of history rather than online reasoning.

That reframes intelligence as a layered story. The present mechanism may be simple. The hard part may have happened upstream in the process that selected it.

Figure 6aSelection pressure
generation 0mean 0.51

Selection does not need foresight. It only needs a target zone and a way to keep whatever already lands nearer to it.

Figure 6aA population drifts toward the target zone across generations even though no individual agent knows what the target means.

Good fits need not be understood

Selection is powerful precisely because it does not require full comprehension. It only requires differential retention. That is enough to produce designs whose apparent purpose exceeds the understanding of any single step that produced them.

Figure 6bOpen population
generation 0mean 0.38

Selection does not need foresight. It only needs a target zone and a way to keep whatever already lands nearer to it.

Figure 6bChange the target, mutation, and survival rate. The population still adapts because selection only needs a retention rule, not a designer’s blueprint.