Mikrokosms and robots

摘要

This paper presents a computer program that simulates situations (called "MIKROKOSMS") in which severl entities (called "organisms") wander around in an environment that includes one another plus other, apparently simpler, entities (called "objects"). The program has a rather simple set of "laws " of objects, which can be thought of as the laws of physics of the MIKROKOSM. It also has other specifications for organisms- their input and output functions (called "perception" and "response"), their reward functions (called "motivation " or "needs"), and their mechanisms for building up internal memories (called "learning" and "hypothesis formation") that will'help them to recognize objects in the future, and respond appropriately to them (for example, in order to maximize expected rewards). This program lays bare the processes that are needed to handle interactions among simulated organisms and objects, including the learning of hypotheses that will