Direct interaction among active data structures: A tool for building AI systems

Abstract

A model of computation and a specific language embodying it are presented. Both are called ALM and are inspired by a physical metaphor. In ALM computation is carried on by independent interacting active data structures called active entities. Active entities, like physical particles, interact with “fields” created by other active entities and eventually “collide”. Field interactions and collisions are achieved by means of both an Influence and a Filter; the former is the external display of an active entity internal structure, and the later represents the influences an active entity is sensitive to. These interactions strongly affect the active entities structure. Due to them, an active entity “feels” the world in which it is, and of which it has no knowledge. Active entities evolve following their own prescriptions and then die leaving their sons in their place. Aim entities and their features are proposed as building blocks for constructing reliable artificial intelligence systems whose main characteristics are parallelism, massive distribution of control, maximum availability of distributed knowledge, robustness with respect to changes in their environment, and capability to accommodate unscheduled events. Maximum knowledge availability is achieved by “fields”, while “collisions”, being a powerful synchronization mechanism, allow entities to reach any kind of agreement. Examples are given to show how ALM works with AI problems. Finally, ALM is compared with message-passing and shared memory systems.

Work supported in part by progetto finalizzato 4 C.N.R. 90.00406. PF67.