Active inference, a computational framework derived from the free energy principle, provides a unified account of perception, action, and learning in biological systems. At its core, subpersonal priors play a crucial role in shaping how organisms interact with their environment. These unconscious probabilistic beliefs, encoded at the neural level rather than in conscious awareness, serve multiple essential functions within the active inference framework.
Encoding of Incentive Value Within Prior Beliefs
One of the most distinctive features of active inference is how it handles motivation and value. Unlike traditional reinforcement learning approaches that represent rewards separately from beliefs, active inference absorbs incentive value directly into subpersonal priors.
“On the active inference view, the incentive value of an outcome corresponds to its prior (log) probability, so that preferred outcomes (or goals) have high prior probability. Active inference therefore eludes a separate representation of incentive value, which is absorbed into (subpersonal) prior beliefs.”1
This elegant formulation allows active inference to unify motivational and control processes within a single framework. By encoding goals as high-probability outcomes in the agent’s generative model, the system naturally pursues actions that lead to these expected states without requiring a separate value function.
Integration of Control and Motivational Processes
Subpersonal priors facilitate the crucial integration of control and motivational processes in the brain, which have “partially orthogonal demands and can be factorized; yet at some point they need to be functionally integrated.”1 This integration is essential for motivated control of action.
Within this framework, control and motivation (implemented mainly in dorsal and ventral neural streams, respectively) work together to propagate and prioritize goals. The control hierarchy propagates goals through structured plans or policies, while motivation processes (encoded in priors) prioritize certain goals over others based on their probability within the agent’s generative model.
Hierarchical Goal Processing
Subpersonal priors operate within a hierarchical structure that enables increasingly complex and abstract goal-directed behavior:
“In a control hierarchy, higher hierarchical levels regulate lower levels by setting their preferred or predicted outcomes (or set points), which lower levels realize.”1
This hierarchical organization allows for nested goals and contextual modulation of behavior. Higher-level priors provide context for lower-level inferences, “finessing outcome prediction based on additional (semantic or episodic) information as well as on long-term action consequences and future affordances.”1 For example, choosing a restaurant in anticipation of satiating hunger represents a higher-level prior influencing lower-level sensorimotor processes.
Multimodal Integration and Prediction
Active inference necessarily generates and predicts sensory outcomes across multiple domains. Subpersonal priors integrate predictions across “exteroceptive, proprioceptive and interoceptive signals,”1 allowing for unified multimodal processing.
This multimodal integration is particularly evident in emotional inference, where active inference provides “a formal account of emotional inference and stress-related behaviour, using the notion of Bayesian belief-updating and subsequent policy selection.”2 The model “generates predictions in multiple (exteroceptive, proprioceptive and interoceptive) modalities, to provide an integrated account of evidence accumulation and multimodal integration that has consequences for both motor and autonomic responses.”2
Self-Evidencing and Existential Imperatives
At a fundamental level, subpersonal priors encode basic expectations about continued existence. According to active inference, “agents are fashioned by natural selection, development, and learning to expect to sense the consequences of their continued existence; this is sometimes called self-evidencing.”3
This principle suggests that organisms inherently expect to remain within their characteristic states. Subpersonal priors thus encode the most basic imperative of biological systems—to maintain homeostasis and persist as bounded, separable entities rather than dissipating into their environments.
Driving Epistemic Exploration
Beyond maintaining homeostasis, subpersonal priors also guide epistemic behavior—the active exploration of the environment to reduce uncertainty:
“We use simulations of abstract rule learning and approximate Bayesian inference to show that minimizing (expected) variational free energy leads to active sampling of novel contingencies. This epistemic behavior closes explanatory gaps in generative models of the world, thereby reducing uncertainty and satisfying curiosity.”5
This facet of active inference explains how organisms balance exploitation of known resources with exploration of novel possibilities. Subpersonal priors about expected information gain drive curiosity-based behaviors that ultimately improve the agent’s model of the world.
Solving the Inverse Problem
Subpersonal priors help address the fundamental “inverse problem” faced by any perceptual system—inferring the causes of sensations when there is no direct access to those causes:
“The brain does not have direct access to causes of sensations, nor is there a stable one-to-one mapping between causes and sensations… the brain cannot access the true posterior probability over the causes of its sensations because this requires evaluating an intractable marginal likelihood.”4
By providing structured expectations about the world, subpersonal priors make this otherwise intractable inverse problem solvable through approximate Bayesian inference. They constrain the space of possible interpretations of sensory data, making perception and action possible despite inherent computational limitations.
Conclusion
Subpersonal priors serve as the essential probabilistic backbone of active inference, unifying perception, action, motivation, and learning within a single theoretical framework. By encoding incentive values, facilitating hierarchical control, integrating multimodal predictions, driving epistemic exploration, and solving the inverse problem of perception, these unconscious beliefs enable organisms to efficiently navigate their environments despite computational constraints and sensory limitations.
The elegance of active inference lies in its ability to absorb traditionally separate constructs like value and reward into prior beliefs, creating a unified framework that bridges control and motivation while explaining how organisms persist in uncertain environments through continuous self-evidencing.