Simulating artificial life using Boltzmann machines

This work describes a concept neural architecture and an expandable software library we created to simulate autonomous biologically inspired agents (biots) in a changing environment. The concept and the software developed are based on a set of restrictive rules. First, whenever there is a decision to be taken, the method more alike to biological systems is to be chosen. Second, we made it a big point that biots were individual beings capable of learning by their own experiences, not instances of a single algorithm. Third, the biots are supposed to know as little as possible about their location in the testing ground, methods of reaching the goal or any measurement of progress. We call that the "no hints" rule. Finally, a rule that is in a way related to the previous rules; in hope of future hardware implementations, we try to keep the model physically realistic and implementable in hardware. This is not always quite possible due to certain simplifications.