An artificial organism whose structures and processes are as similar as possible to an animal ones.
The problem is however in the way an animat acquires such structures and processes.
There are three possible models of animats (at least!)
- animats can be made to learn behaviors. This can be done endowing them with a neural network, thus making them adaptive in a range of experimental situations through a set of interacting rules organized as to cope with external changes that may have an impact on the animat
- animats can be made evolutive. Again this can be obtained through an appropiate neural network and a set of rules. But the need for evolution can be installed by introducing a measure of randomness in the rules.
However, none of these animats models seems to tackle all the aspects of modeling animal behavior, specially "intelligent" beings.
E. VACCARI and M. D'AMATO write: "Cognitive systems are composed of multiple subsystems which are simultaneously active and interacting. Their cognitive behaviors are pervaded by both continuities and discreeteness and their kind of structures emerge over time. Further cognitive processes operate over many time scales and events at different time scales interact"(2000, p. 173)
In short, a good animal model should never become an homeostat in the sense of ASHBY.
Or, even the existence of constraints cannot totally prescribe all the modalities of change, and still less, stop change altogether.
Already in the 18th century LEIBNIZ (1646-1716) stated that any change in one element in an entity is conducing to modifications of all the other elements. Such a process takes necessarily place at multiple levels, in different ways, at variable time scales, and can thus be modellized only in a conjectural and imperfect way, be it for an animal, or an animat.