Efficiency is quite nice when you have no competitors, but when you do it comes with the nasty downside of predictability. If you always make the most efficient choice given your options, then your opponents will be able to predict your choices and wreck you easily. So, agents typically choose to sacrifice some efficiency in return for some much needed unpredictability. Examples:
- In predator-prey interactions, escaping prey typically add random noise to their movements, zigzagging, looping, and bouncing away. Although this is less efficient, it prevents predators from learning a systematic rule to predict exactly how prey will escape. (see here)
- In primate courtship, individuals may have evolved to add random variability to their thoughts and behaviors so as to avoid having their minds read by fellow apes using theory of mind. (see pdf here)
- In game theory, nearly every solution to interesting multi-person games has a mixed rather than pure solution. For example, in the iterated prisoner's dilemma the algorithm "generous tit for tat" cooperates sometimes even after an opponent's defection, but only randomly. Thus it is impossible to exploit, but avoids the infinite cycles of defection you get from pure tit for tat. (see here)
(I'm hoping to compile a list of all the canonical trade offs, which should be fun for the whole family. Above photo of peregrine falcon chasing a cormorant comes courtesy of flick user Nick Chill.)
