The Cyber-Physical System Approach Towards Artificial General Intelligence: The Problem of Verification

Cyber-Physical Systems have many components including physical ones with heavy demands on workflow management; a real-time problem. Furthermore, the complexity of the system involves some degree of stochasticity, due to interactions with the environment. We argue that the factored version of the event-learning framework (ELF) being able to exploit robust controllers (RCs) can meet the requirements. We discuss the factored ELF (fELF) as the interplay between episodic and procedural memories, two key components of AGI. Our illustration concerns a fELF with RCs and is a mockup of an explosive device removal task. We argue that (i) the fELF limits the exponent of the state space and provides solutions in polynomial time, (ii) RCs decrease the number of variables and thus decrease the said exponent further, while the solution stays

$$\epsilon $$

ϵ

-optimal, (iii) solutions can be checked/verified by the execution being linear in the number of states visited, and (iv) communication can be restricted to instructions between subcomponents of an AGI system.