Agents are inherently complicated because they are defined as a specific combination of parts; they exist in an environment, they observe the environment and act on it, they have a model of the environment (a “world model”), they have values, and they use this world model to choose actions that they believe will tend to cause the environment to become more valuable. A foundational theory of agents will need to use mathematical tools and frameworks that are capable of robustly capturing all these component parts with enough fidelity to be able to say useful things about the behavior of real implemented agents.
