Justify representing agent entities as stochastic processes

Establish whether the assumption that the set of entities that contains agents can be represented by interacting stochastic processes, as commonly done in the perception–action loop formalism, is formally justified for agent-like entities within multivariate Markov chains such as cellular automata.

Background

The paper contrasts the standard perception–action loop approach, which models agents and environments as interacting stochastic processes, with an alternative representation of entities as spatiotemporal patterns in multivariate Markov chains. The authors note that while the PA-loop affords convenient definitions of actions and perceptions, the fundamental assumption that agents can be captured by stochastic processes has not been formally established.

Clarifying this foundational assumption would help determine the scope and limits of the PA-loop framework and its compatibility with STP-based entity notions in systems like cellular automata.