[Sharlin]

Making occasional mistakes while applying formal knowledge seems to be qualitatively quite different from making suboptimal decisions because you’re improvising due to lack of such formal knowledge.

In some domains you don’t have a choice; there are too many moving parts, decision making is of necessity based on noisy and incomplete information, the domain is too messy for formal models to be useful. You just have to intuit and improvise. This is more often than not the case when human interactions are involved. Trying to impose engineering solutions to social problems usually ends in disaster.