A finite net of interconnected elements which operate on basis of an on-off logic.
This is a very general model that has been used in genetics (St. KAUFFMAN, 1969), in models of the cerebral cortex (D. DUBOIS, 1991), in connexion machines (T. TOFFOLI & N. MARGULIS, 1987) and quite recently in distributed artificial intelligence (R. BROOKS (1989); C. LANGTON; L. STEELS, 1990)
Automata nets present characteristic emergent properties and behaviors, which depend on the nature of their interconnections or "wiring".
The net must have constraints, i.e. some connections being permitted and other ones suppressed. Flows through the connections can be permanent or more or less intermittent. Furthermore, such nets tend to subdivide into local and specialized subnets and, as a result, simultaneous events in different parts of the net do not propagate their effects instantaneously and generate very complex nonlinear (and sometimes chaotic) behavior.
Automata nets present very numerous alternative possible steady states and in some cases, undergo giant fluctuations that may lead to bifurcations and dissipative structuration.
Once the set of internal states can be specified through a finite state transition matrix, the automata net becomes a finite automaton or, if it produces outputs, can be assimilated to a sequential machine. This does not however make it wholly predictable.