In different contexts the function that assigns values to types of behavior has been known under different names: In genetics it represents fitness, utility in economics, error functions in engineering, performance indices in motor learning, payoff in game theory, and power in politics. Some of these functions might not be rigorously defined but as long each agent follows a local decision rule that will improve the generalized fitness value no explicit definitions are necessary. Of course these local decision rules might that evolved through past evolution might not guarantee survival in a changed environment especially if those changes happen rapidly compared to evolutionary time scales.
Note that the words "maximize/minimize" have not been used specifically since such a strategy often is quite risky on longer time scales. For instance in natural evolution "super predators" such as saber-toothed tigers emerged at different times and independently on different continents. But apparently they all went extinct within a very short (on evolutionary scales) time. It seems to be a general rule that the more chaotically an environment changes the more successful are sub-optimal but less brittle evolutionary strategies.
The general rule for improved performance of a complex adaptive system seems to be that it is better to increase the probability of a desired behavior by assigning it a higher value for the fitness function than trying to explicitly specify any procedure that the system should follow. In Tom Ray's Tierra model computer programs are produced without specific design rules and with significant levels of errors but under the constraint of a fitness parameter (here utilizing computer resources). The surprising result was that after a relatively small number of generations, programs emerged that solved certain programming problems more efficiently than those written by human programmers did. It is interesting that the performance improvements were not simple monotone functions of time but showed similar behavior as other evolutionary systems sometimes referred to as "punctuated equilibrium".