Emergent Properties as Indeterminacy of Dynamical Systems:

Animal Behavior Revisited

Masao Migita and Kanji Ueda

Animal behavior is often referred in AI and AL studies with special care on its evolutionary process. The main issue in those AI discourses is that no animal is exposed to the so called frame problem, while an AI device based on inference processes is destined to be annoyed with the problem; here an animal is considered to act responding to external stimuli, thus, it does not infer at all. And the main issue in AL discourses is emergence of unexpected behavioral properties cumulating known simple ones. For example, the most popular property observed in AL models would be cooperation between 'selfish' agents. Thus, in both AI and AL discourses which refer animal behavior, an agent makes its decision based on so a local viewpoint that it cannot see the 'globality', that means the relevant frame of reference in AI and certain cooperative behaviors beneficial to a group of the agents in AL. Here, the evolutionary processes is important, because it implies what stimulus-response (input-output) relationships are relevant in solving a problem while the agent lacks of global viewpoint. Once the set of stimulus-response relationships to be selected is determined, one can give a fitness value to each relationship according to the problem, and the solution would be brought in the form of the one in a certain dynamical system.

However, if one employs a non-linear dynamical system, and non-linear one is usually the case, one has to cope with every micro perturbation of the value in order to maintain the device as one dynamical system. Any trifle difference of the value has to be identified whether it is mere a perturbation originated from a certain external cause or an intrinsic one which came to be seen by means of the recursive calculation, otherwise, one cannot help admitting that he uses the dynamical system without any specification that enables him to identify every vale in above sense. The authors will discuss that this indeterminacy of the dynamical system that rules the evolution of an agent's behavior would be clear in observations of animal behavior, where one can find positive meanings in such an indeterminacy, for instance, autonomy. Then, the authors will try to manifest that the indeterminacy is also essential in emergence in the context of AI and AL. The main difference between the emergent properties addressed in the present study and those in the conventional AI and AL scheme is that the former implies modification of the problems themselves, but the latter not.