An Evolutionary Inference of Optimal Selection in View of Dynamic Programming and Reinforcement

A global utilitarianism of von Neumann–Morgenstern expected utility function is widely accepted. However, players are not used to take their future scope entirely smooth and rather cope with the environment being similar to landscape with hidden land mines. In modern times, this game setting is appropriate for the entertainment industry. The games provided by the industry always impose learning and simulating on players so that they avoid their risks. In an era where machine learning is flourishing, how evolutionary economics can inform this situation is important. Taking a more general case, for instance, if both the observer and the target are moving, then the events faced can only be partially observed. We can see that such a situation can be analyzed in terms of the “Bellman equation” of dynamic programming and “reinforcement learning” (machine learning), although only to the extent that Markov chains are valid. In this chapter, we also focus on the case where the event in question is only partially observable, and identify the optimal selection. Without any particular modeling or approaching function, machine learning uses a common method to perform its tasks independently of specifically human logic/inference. Here, it is noted that some optimal results are not necessarily connected with subjectively intentional/moral modeling, which is preferred in economics. Therefore, there is little implication that the optimum can only be obtained with some special economic modeling. The meaning of optimality also changes because it can be obtained without any specific modeling. This insight will awaken us to reconsider our economic inference.