Book review: From Animals to Animats 2. Edited by Jean-Arcady Meyer, Herbert L. Roiblat, and Stewart W. Wilson (MIT Press)

Abstract

It is time animals lived side by side with animals. They can live, learn, build maps, evolve, and be our domesticated pets too. In the intersection of AI, Robotics, Philosophy and Psychology, traditions of building ever more complex systems that try to exhibit intelligent behavior or solve puzzles are giving way to systems that live simple lives, without pretending to reason deeply, or to act as if they were real animals. Instead, the system designer builds simple agents and endows them with basic shallow abilities to interact with the world. The point is that the designer doesn't solve all the problems that the system may encounter beforehand. Nor does the designer build a system that acts as if it is trying to talk to itself outloud. Many of the traditional symbolic systems are designed so that they can be made to generate a transcript of human understandable steps they take to decide on performing each and every action. Once deployed in a complex world, the animat (just as a device would) uses its repertoire of interactive behaviors as they become needed (e.g., a microwave oven responds to buttons pushed). The interactions that follow may appear intelligent. Such agents can also be made to learn from these interactions and enrich their range of capabilities.