Predictive Models Are Control Theory

A predictive model is not only a passive description. Once a model is used to choose an action, it becomes part of a control loop.

Control theory makes the relationship explicit: there are actions, observations, state estimates, controllable variables, disturbances, and a desired future state. A model predicts what the system may do after a possible action, and that prediction shapes the next action.

The same structure appears in learned physical models, robotics, reinforcement learning, operations tools, and user-facing software that recommends or automates decisions. The model does not have to be mysterious; it has to expose what it is conditioned on and what effect it is allowed to have.

The useful question is practical: what variables can be changed, what is only observed, what is being optimized, and how does the system detect when its prediction is wrong?