The Role of constraints in coputation exploring optimality through limited resources and intrinsic motivation

When interacting with the world, agents take actions they consider optimal, that is, best. Typically, this optimality is evaluated over the entire space of possibilities. However, agents are inherently bounded -- for instance by energy, time, or even lifespan -- resulting in behavior that often deviates from global optimality. In this work, we investigate how introducing constraints reshapes the concept of optimality. First, we explore optimality within the framework of the Breadth-Depth (BD) dilemma, a one-shot allocation of limited resources in large decision trees. Next, we rethink utility and reward within the Maximum Occupancy Principle (MOP): in this intrinsic motivation framework, extrinsic rewards are not the goals per se but means to sustain actions in the world. Finally, we introduce NeuroMOP, showing how intrinsic motivation provides a framework to interpret neural variability. Our results highlight the potential role of constraints in shaping computations in the brain, spanning from neural activity to behavior.