Biofeedback from the Free Energy Principle Perspective: Some Psychoeducational and Clinical Implications

Why do some people stay calm under pressure while others feel overwhelmed? What enables humans to adapt to constant stressors, changing environments, and unpredictable life events? These everyday challenges point to a deeper scientific question that sits at the heart of neuroscience and cognitive science: How do biological agents maintain stability and resilience despite continuous environmental fluctuations? The Free Energy Principle (FEP) (Friston, 2010) offers a theoretical framework to understand these mechanisms of adaptation with far-reaching implications for mental health, behavioral therapy, and personal well-being—a quantity that measures the mismatch between what the brain expects and what it actually senses. It serves as a proxy for surprise or uncertainty: The greater the gap between the brain’s predictions and the real sensory input, the higher the free energy (Friston, 2010; Friston et al., 2006).

When an organism is surprised—facing unexpected changes or ambiguous situations—its free energy increases. Conversely, when it successfully predicts or influences its environment, free energy decreases. According to the FEP, all perception and action aim to reduce this mismatch either by updating internal beliefs to better match the world or by acting to make the world more predictable (Friston et al., 2017; Parr et al., 2022). In this way, minimizing free energy becomes a universal strategy for adaptive behavior and maintaining physiological and psychological balance. The process of beliefs updating is called active inference.

In simple terms, active inference is the process by which organisms try to make sense of the world and act in it by minimizing surprise. Because they can’t directly calculate surprise, which would require knowing all possible outcomes, they use a practical shortcut: minimizing a mathematical approximation of surprise known as free energy (Buckley et al., 2017). This allows the brain to constantly compare its expectations to incoming sensory information and update its internal model accordingly. At the same time, it can also act on the world to make sensory input more predictable. Through this dual process—updating beliefs and guiding actions—organisms keep their internal states (such as temperature, heart rate, or glucose levels) within healthy bounds. In this way, homeostasis is not passively maintained but actively regulated by continuously reducing prediction errors through perception and action. The FEP brings together ideas from physics, probability theory, and cognitive science to explain how the brain supports intelligent, adaptive behavior.

To illustrate the principles of active inference, consider a person navigating through a dimly lit room. One must rely on an internal model of the room’s layout to predict obstacles and guide movement. If unexpected sensory prediction errors occur—such as bumping into furniture—the internal representation (perceptual inference) can be updated or one can modify behavior (active inference). This continuous interplay between predictive modeling and action exemplifies free energy minimization.

FEP is deeply rooted in predictive coding, a hierarchical model of brain function in which cortical areas generate predictions about sensory input with error signals propagating upward when mismatches occur (Friston et al., 2006; Rao & Ballard, 1999). These error signals prompt adjustments in internal models, allowing for refined predictions and adaptive responses. The brain, therefore, is not a passive processor of sensory input but an active constructor of perceptual experience, minimizing uncertainty about the external world.

Beyond perception, FEP governs biological regulation from cellular homeostasis to higher cognitive functions. For example, thermoregulation illustrates free energy minimization in physiology: When an individual experiences cold, the person can either adjust perception (reinterpreting cold temperature as a tolerable mild chill) or engage in corrective behavior (putting on a sweater). This hierarchical control ensures organisms not only react to stimuli but proactively shape their experiences through predictive regulation.

Active inference extends this view by treating action as an inferential process rather than a response to external stimuli (Friston et al., 2012). In this framework, behavior is driven by the imperative to resolve uncertainty rather than simply by reward maximization. Agents select actions that optimize evidence for their generative model, reducing uncertainty about their own sensory states. This perspective aligns Bayesian inference with decision making, demonstrating that behavior emerges from the need to minimize both epistemic and pragmatic uncertainty.

This framework positions the brain as an anticipatory system, continuously generating and adjusting predictions based on sensory evidence. This concept is supported by research on perception, motor control, and psychiatric disorders, showing that aberrant inference—such as excessive or insufficient precision of prediction errors—can underlie conditions ranging from hallucinations in schizophrenia (Adams et al., 2013) to motor impairments in Parkinson’s disease (Friston et al., 2016). Similarly, decision making under uncertainty can be understood as the process of selecting actions that minimize expected free energy, balancing exploration (epistemic actions) with goal-directed behavior (pragmatic actions) (Parr & Friston, 2017).

Taken together, the FEP presents a unifying theory of cognition, perception, and action, suggesting that all adaptive behavior arises from minimizing variational free energy. By extending Bayesian inference beyond perception to action selection, FEP bridges the gap between sensory processing and motor control. This theoretical framework provides a foundation for understanding how individuals learn to interpret and regulate their own bodily states through interoception.

Interoception, Homeostasis, and Stress: An Active Inference Perspective

The ability to regulate physiological states is an extension of the predictive mechanisms governing perception and action. Just as the brain continuously updates its internal generative model to minimize surprise in exteroceptive perception, it applies the same inferential process to interoception: the modeling and regulation of internal bodily states. This predictive framework provides a theoretical foundation for understanding homeostasis, stress regulation, and biofeedback interventions, which are explored in the following sections.

Interoception refers to the brain’s ongoing inference about the body’s physiological condition (Craig, 2002; Critchley & Harrison, 2013). It governs the monitoring of autonomic variables such as heart rate, respiration, and hormonal levels. Within the FEP, interoception is conceptualized as an inferential process in which the brain generates and refines predictions about bodily states to ensure stability (Barrett & Simmons, 2015; Seth, 2013). This process enables homeostasis, maintaining internal equilibrium in response to environmental and physiological fluctuations (Stephan et al., 2016).

Continuing with the example of feeling cold, the brain uses two complementary strategies to regulate internal bodily states. The first is interoceptive inference, by which it updates its internal model based on signals from within the body, such as chills or tension in the skin. The second is interoceptive control, which involves triggering physiological or behavioral responses to restore balance (Pezzulo et al., 2015). So, after stepping into the cold, the brain might downplay the discomfort by reinterpreting the sensation as tolerable (adjusting perception), or it might initiate an automatic response such as shivering to generate heat (adjusting physiology). This dynamic interaction between updating predictions and taking action illustrates how minimizing interoceptive free energy supports flexible, adaptive regulation of the body.

Stress regulation is a critical domain of interoceptive inference. According to predictive coding, stress arises when a sustained discrepancy exists between expected and actual interoceptive states, signaling a breakdown in homeostasis (Paulus et al., 2019). The autonomic nervous system responds by mobilizing resources, such as increasing heart rate and cortisol secretion, to restore equilibrium. However, when stress-related prediction errors persist, maladaptive responses may develop, contributing to anxiety and affective disorders (Barrett et al., 2016).

Consider public speaking as an example. An individual with accurate interoceptive predictions may interpret mild arousal as normal and regulate the response adaptively. These predictions rely on interoceptive priors: the brain’s preexisting expectations about how bodily signals should feel and what they typically mean. In effect, these priors act as mental filters, shaping how internal sensations are interpreted before they even reach conscious awareness. In contrast, someone with maladaptive priors—such as an exaggerated expectation of failure—may misinterpret the same bodily sensations as signs of imminent danger, escalating stress responses and reinforcing maladaptive predictive loops (Seth & Friston, 2016). Cognitive strategies such as reappraisal and mindfulness function as mechanisms for updating interoceptive priors, supporting more flexible physiological regulation.

The ability to modulate stress responses depends on precision weighting, which determines the balance between sensory evidence and prior beliefs (Clark, 2013). Excessive precision assigned to interoceptive prediction errors can lead to hypervigilance and heightened stress sensitivity as seen in anxiety disorders. Conversely, rigid priors may result in an inability to respond appropriately to physiological changes, a phenomenon observed in alexithymia and hypoarousal states (Ondobaka et al., 2017). Effective stress regulation requires adaptive precision weighting, allowing individuals to appropriately recalibrate bodily predictions.

Interoception also plays a foundational role in emotional regulation as emotions emerge from the brain’s attempts to interpret internal physiological signals in a predictive framework (Seth, 2013). The constructed emotion hypothesis suggests that emotions are not direct responses to external events but active inferences integrating interoceptive and exteroceptive cues (Barrett, 2017). For example, fear is not simply a reaction to an external threat but a prediction linking autonomic arousal to a fear-related context. When interoceptive signals are misattributed, emotional dysregulation may occur as seen in panic disorder when benign bodily fluctuations (e.g., heart palpitations) are misinterpreted as signs of imminent catastrophe (Barrett & Simmons, 2015).

By understanding interoception as an inferential process, we can reframe autonomic and affective dysregulation as failures in predictive updating rather than purely pathological responses. This perspective has direct implications for therapeutic interventions: if maladaptive physiological states arise from faulty interoceptive priors, treatments should focus on retraining interoceptive inference. In this context, biofeedback emerges as a valuable tool for recalibrating bodily predictions, providing structured sensory evidence to guide more adaptive regulation of physiological states.

Biofeedback and the Free Energy Principle: An Active Inference Perspective

Biofeedback therapy provides individuals with real-time physiological data, enabling them to regulate autonomic processes that are typically outside conscious awareness. Biofeedback techniques have been widely applied in stress management, pain regulation, and cognitive enhancement by allowing patients to modulate variables such as heart rate variability (HRV), skin conductance, and muscle tension (Schwartz et al., 2013). From the perspective of the FEP, biofeedback can be understood as a structured means of reducing interoceptive prediction errors by providing precise sensory evidence that updates an individual’s generative model of bodily states (Stephan et al., 2016). In essence, biofeedback enhances the precision of interoceptive signals, facilitating the recalibration of priors related to autonomic regulation.

The active inference model, an extension of the FEP to action selection and control, provides a mechanistic explanation for how biofeedback influences physiological regulation. Within this framework, physiological self-regulation is not simply a conditioned response to reinforcement but an inferential process wherein the agent actively minimizes free energy by aligning interoceptive signals with expected bodily states (Friston et al., 2017). The biofeedback loop serves as an augmented sensory channel that refines interoceptive inference by offering external confirmation of internal physiological states. For example, a patient learning to regulate breathing through HRV biofeedback does not simply learn an association between breathing patterns and rewards but rather develops an improved generative model of cardiorespiratory dynamics, reducing interoceptive surprise and optimizing autonomic function.

A key advantage of the active inference framework over traditional reinforcement models lies in its ability to account for complex, continuous feedback mechanisms. In conventional paradigms of biofeedback training, physiological processes are modified through reinforcement mechanisms, which shape and strengthen desirable responses (Hardt & Kamiya, 1978). However, this view struggles to explain why biofeedback modalities that use continuous, nondiscrete feedback—such as smooth video progressions or auditory modulations—can be equally or more effective than simple binary cues, such as beeping or flashing lights. Under active inference, this discrepancy is resolved by considering biofeedback as an epistemic resource that enhances the precision of interoceptive signals rather than merely reinforcing motor outputs. This means that a moving visual stimulus, such as a progressively clearer image or a smoothly advancing video, provides a richer probabilistic structure for inferring bodily states compared with discrete reinforcement cues (Witte et al., 2019).

An illustrative example is found in thermal biofeedback, a biofeedback-based relaxation technique that facilitates voluntary control over physiological states, such as peripheral blood flow and temperature regulation (Fahrion et al., 1986). From an active inference perspective, autogenic training serves as an intentional modulation of interoceptive priors, reducing prediction errors associated with autonomic regulation. Consider an individual practicing hand-warming through autogenic training: Instead of merely responding to external reinforcement, the individual constructs a generative model in which warmth is expected in the extremities. By repeatedly engaging in this practice, interoceptive precision weighting is adjusted, and descending control over vasodilation is facilitated through the autonomic nervous system. This aligns with the broader active inference view that bodily regulation emerges through hierarchical predictive coding rather than simple stimulus–response mechanisms (Pezzulo et al., 2015).

The role of beliefs in active inference is central to understanding the cognitive mechanisms underlying biofeedback. In the FEP framework, a belief is not a static propositional attitude but a probabilistic model that encodes expectations about sensory states (Hohwy, 2013). Biofeedback can modify these beliefs by providing an explicit mapping between interoceptive sensations and external feedback, thereby reducing uncertainty about the regulation of physiological states. For instance, a patient suffering from chronic stress might implicitly hold a maladaptive belief that the autonomic arousal is uncontrollable. Through biofeedback training, the patient can receive empirical evidence—via HRV modulation for example—that demonstrates the capacity to exert some voluntary control over autonomic function, thus revising the patient’s generative model toward a more adaptive state.

This perspective also suggests that thoughts themselves can serve as an intervention technique within biofeedback paradigms. If beliefs about bodily states influence interoceptive inference, then cognitive reframing—such as mental imagery, self-suggestion, or attentional shifts—can act as an endogenous form of biofeedback, shaping physiological responses in a top-down manner (Seth & Friston, 2016). For example, research on placebo effects suggests that expectancy-driven modulation of physiological processes follows the same computational principles as active inference: The brain constructs priors about anticipated bodily states, which, in turn, influence autonomic and somatic outcomes (Büchel et al., 2014). This implies that mental strategies employed during biofeedback, such as visualization or verbal self-instruction, can be conceptualized as precision-weighted priors that guide physiological regulation.

Taken together, biofeedback provides a structured method for recalibrating interoceptive inference by enhancing the precision of sensory signals and updating maladaptive generative models. Unlike reinforcement-based approaches, which frame physiological regulation as a conditioned response to external cues, the FEP and active inference offer a more nuanced explanation in which biofeedback serves as a tool for epistemic control, reducing uncertainty about bodily states through structured sensory evidence. This perspective advances our understanding of biofeedback’s efficacy and underscores the broader role of interoceptive inference in self-regulation and health.

Psychoeducation Strategies for Biofeedback Patients

Effective psychoeducation helps patients understand the mechanisms behind their experiences and provides them with practical tools for self-regulation. Below, each theoretical concept is presented alongside a corresponding therapeutic message designed to convey it in a clear and supportive manner.

Reframing Biofeedback Through Predictive Processing

Biofeedback, when viewed through the lens of predictive processing, reveals itself as more than a method for altering bodily signals—it becomes a structured process for recalibrating the brain’s generative models of internal states. Grounded in the FEP and active inference, biofeedback helps individuals correct misaligned interoceptive predictions, reduce persistent prediction errors, and reestablish physiological balance. This framing shifts our understanding of biofeedback from reactive symptom control to proactive model refinement.

Key mechanisms involved in this process include the following:

• The modulation of precision weighting

• The restructuring of interoceptive priors

• The enhancement of interoceptive awareness

By addressing how individuals assign confidence to sensory signals versus internal expectations, biofeedback trains more adaptive stress responses. When combined with cognitive techniques—such as reframing or attentional redirection—it supports the development of more flexible interpretations of bodily states. In this way, biofeedback acts as both a physiological regulator and a cognitive tool for emotional resilience.

A promising direction for future interventions lies in explicitly incorporating interoceptive training, helping individuals better detect and interpret subtle internal cues, such as heartbeats or breath rhythms. Research suggests that improving interoceptive accuracy supports better discrimination between stress signals and neutral sensations, thus reducing maladaptive responses (Khalsa et al., 2018). Integrating this layer into biofeedback protocols could strengthen predictive updating and foster deeper self-awareness. Ultimately, by understanding biofeedback as a predictive recalibration tool, we open the door to more personalized, precise, and psychologically grounded approaches to stress regulation and emotional well-being.

This reconceptualization also positions biofeedback within the broader domain of psychophysiological psychotherapy rather than as a collection of isolated technical procedures. By emphasizing its role in retraining interoceptive inference, biofeedback becomes a process of therapeutic change that is both physiological and psychological. It engages the patient’s beliefs, expectations, attentional patterns, and emotional responses, making it deeply experiential and personally meaningful. In this view, the intervention does not simply teach someone to lower one’s heart rate, but instead invites them into an embodied process of self-discovery where physiological change arises from cognitive insight and vice versa. This aligns with contemporary models of psychotherapy that emphasize integrative, brain–body approaches to regulation and well-being.