Homeostatic models in artificial intelligence (Q1842684)

The authors consider a system \({\mathcal S}\) which can be decomposed into a system \({\mathcal E}\), called the environment, and several subsystems \(s_i\) which interact with \({\mathcal E}\). They give three examples where the interaction is such that global objectives for \({\mathcal E}\) or \({\mathcal S}\) are attained. Regulation in some biological systems is such that the function of the \(s_i\) is to help achieve the global objectives of \({\mathcal E}\). The theory of cooperative behaviour of automata considers the case where the \(s_i\) are automata whose behaviour is dictated by their own goals and the action of \({\mathcal E}\). \({\mathcal E}\) acts on \(s_i\) in a distributed way so that objectives for \({\mathcal S}\) are satisfied. Distributed artificial intelligence (for example networks of computers working in a parallel and decentralized way) considers agents \(s_i\) which learn about \({\mathcal S}\) and adapt their behaviour accordingly. One does not need to consider \({\mathcal E}\) anymore.