Via twitter, @unlearningecon makes a good point:
Potential problem with agents fully exploring the opportunity set is 'localised' optimising which could bias it in general
If I open a new piece of state space (or close one off), then it's likely agents "near" that piece of state space are the ones that are explore it (or leave for nearby open areas). This makes the process of evolution "localized". In a biological system, species in different ecosystems evolve in ways largely independent of species in other ecosystems, which may eventually result in some kind of conflict or catastrophe. More importantly, evolution happens from the current set of species (exploring parts of state space near existing species), creating strong path dependence.
In economic systems, e.g. firms will explore state space near existing firms, which may or may not find the best solutions.
This issue with "localized" optimization turns into a good argument in favor of government intervention in markets to both alleviate issues with path dependence (e.g. by providing workers with help if their industry needs to be reduced in favor of some other more socially optimal industry) and providing new seeds to grow in less explored areas of state space (e.g. funding science and innovation). I missed the opportunity to make this point in my book, but it is a good one.
In future posts, I will discuss unlearningecon's other points.