Summary
Allostasis is a brain-centered form of predictive regulation in which the nervous system anticipates the body's needs and attempts to meet them before they arise, maximizing metabolic efficiency. This contrasts with homeostasis, which reacts to deviations from fixed set points within fixed tolerances. Interoception — the brain's sense of the internal physiological state — provides the performance feedback needed to guide allostatic control. The paper "Interoception as modeling, allostasis as control" by Sennesh et al. (2022) formalizes allostasis using control theory, describing how the brain models capacity curves (relationships between physiological disturbances and regulatory responses) and uses them to compute objective functions for stochastic optimal control. This framework integrates physiology, motor control, and decision making, and is concretely illustrated through examples such as the baroreflex, glucose regulation, and a dodgeball scenario.
Key Points
- Allostasis anticipates needs before error occurs; homeostatic set points are replaced by dynamic reference trajectories.
- Interoception solves an inverse problem: inferring body state from noisy, ambiguous visceral signals using an internal generative model.
- Physiological variables are divided into regulated resources (e.g., blood glucose, core temperature) and controlled processes (e.g., heart rate, sweating) that stabilize them.
- Capacity curves describe the relationship between a controlled process and a regulated resource; they have an operating point (optimal responsiveness) and limited range (threshold and saturation).
- The brain estimates capacity curves in terms of quantiles, providing a time-independent performance metric that adapts to shifting parameters.
- Control theory concepts (plant, controller, feedback, disturbance, reference trajectory, internal model) are applied to physiology, motor control (referent control hypothesis), and decision making (stochastic optimal control).
- The proposed Allostatic Path-Integral Control (APIC) model treats action concepts as internal models of potential behaviors, balancing intake and effort over an indefinite horizon.
- Allostasis contains homeostasis as a special case; an allostatic controller can implement homeostatic set points but not vice versa.
- Evidence suggests the brain does not require a dedicated "reward center"; reward prediction errors may be reinterpreted as movements toward or away from physiological operating points.