What Is Predictive Processing? How Your Brain Predicts the Present

Written by Stephanie Underwood, RSW

Learn how predictive processing works, including priors, sensory evidence, precision and prediction error, and why it matters for trauma and relationships.

What Is Predictive Processing? How Your Brain Uses the Past to Predict the Present

You are not experiencing the world exactly as it is. Actually, none of us are. At every moment, your brain is receiving enormous amounts of information from your senses but sensory information does not arrive with its meaning attached.

For example, you walk into a store and notice the cashier looking angry. A facial expression does not arrive labelled "angry.", just like a delayed text message does not arrive labelled "rejection." Your brain has to determine what those cues probably mean, and it does not start from scratch. It uses what it has learned before to predict what is most likely happening now.

This is the basic idea behind predictive processing, an influential framework in neuroscience and cognitive science that proposes that perception involves an ongoing interaction between predictions generated by the brain and incoming sensory evidence. Rather than functioning like a camera that simply records the outside world, the brain can be understood more like an inference system. It is constantly attempting to answer a question: based on what I already know and the evidence available right now, what is most likely going on? That seemingly simple idea has enormous implications for how we understand perception, emotion, anxiety, trauma, attachment, and human relationships.

Predictive Processing in Simple Terms

Imagine walking into your kitchen late at night. The room is dark. You glance toward the counter and see a long, curved shape. For half a second, you think: snake. Then you turn on the light. It is a phone charger.

What happened? Your eyes provided incomplete sensory information, and your brain had to interpret it. Based on the shape, lighting, context, previous knowledge, and whatever other information was available, one explanation briefly became plausible: there is a snake on the counter. Then new sensory evidence arrived. The lights came on. The new information did not fit the original prediction very well, and your brain revised its interpretation. Phone charger.

That small moment illustrates several of the fundamental ideas behind predictive processing. Prior knowledge influenced perception. The brain generated a prediction about the cause of sensory information. New evidence was compared with that prediction. The mismatch provided information for updating. And your interpretation changed. Most of the time, this happens so quickly that you never notice it.

Your Brain Does Not Simply Receive Reality

The traditional intuition about perception is fairly straightforward: something happens in the environment, your senses detect it, the information travels into the brain, the brain identifies what it is, and you respond. But perception is considerably more complicated than that. Sensory signals are noisy, incomplete, and sometimes ambiguous. Your brain therefore has to infer the hidden causes behind them. The light reaching your retina is not the object itself. Sound waves reaching your ears are not inherently words, danger, or reassurance. Signals provide evidence, but meaning has to be inferred.

Predictive processing proposes that the brain helps solve this problem by using an internal model of the world to generate predictions about the causes of incoming sensory information. Perception emerges from the interaction between what the brain predicts and what the senses provide. The brain is not passively receiving reality. It is actively constructing its best guess about what is happening.

The Brain Is a Prediction Machine, But Not in the Fortune-Telling Sense

The word "prediction" can be misleading. In everyday conversation, prediction usually means consciously anticipating the future: "I think it is going to rain tomorrow." "I think she is going to cancel." "I think my boss is going to be angry." Predictive processing uses the term more technically. A prediction is a model-generated estimate of what sensory input, state, or outcome should occur given a particular explanation of what is happening. These predictions do not have to be conscious. You do not walk around deliberately calculating the expected shape of every coffee cup, the sound of every familiar voice, or the location of your feet before each step. Much of this processing occurs automatically.

So when researchers describe the brain as "predictive," they are not suggesting that your brain is sitting inside your skull making verbal guesses about the future. They are describing a system that uses previously learned regularities to anticipate and interpret incoming information. The prediction is not a thought. It is a computation, and it is happening beneath conscious awareness.

The Core Components of Predictive Processing

The terminology can become intimidating very quickly. Priors. Likelihoods. Precision. Prediction errors. Bayesian updating. It sounds like someone accidentally enrolled your brain in a statistics course. But the core concepts are surprisingly intuitive.

Priors: What Your Brain Already Believes Is Likely

A prior is essentially what the system considers plausible before the newest evidence has been fully incorporated. Your priors are shaped by previous experience, learning, context, and repeated patterns. Suppose someone hears footsteps behind them while walking. If they are walking through a busy grocery store at noon, one interpretation is likely. If they are alone in a parking garage at 2 a.m., another may become more probable. The sound may be similar, but the context changes the prior probability of different explanations.

Previous learning matters too. Someone who has repeatedly experienced aggression in certain situations may assign greater probability to danger than someone whose history has made the same situation seem relatively harmless. The incoming cue does not change. The model interpreting it does.

Sensory Evidence: What Is Happening Right Now

Predictions are constrained by incoming information. This includes external sensory information such as vision, sound, touch, smell, temperature, and pain. It also includes interoceptive information, meaning signals related to the internal condition of the body, including cardiovascular, respiratory, visceral, and other physiological states. Importantly, sensory evidence is not identical to its interpretation. A furrowed brow is visual information. "This person is disappointed in me" is an inference. A faster heartbeat is physiological information. "Something terrible is happening" is an interpretation. Those distinctions matter, because the gap between evidence and interpretation is where our predictive models do their work.

Likelihood: How Well Does the Evidence Fit an Explanation?

The brain must evaluate how compatible incoming information is with different possible explanations. Imagine receiving a one-word reply from someone: "Fine." What does it mean? They could genuinely be fine. They could be irritated. They could be busy. They could be distracted. They could be ending the conversation. The sensory evidence alone does not resolve the ambiguity. Your brain evaluates possible explanations according to how well the available evidence fits each one, and your previous experience may strongly influence which explanation seems most plausible.

Precision: How Much Should This Information Be Trusted?

Not all information receives equal weight. Predictive-processing accounts use the concept of precision to describe something roughly analogous to confidence or reliability. Imagine hearing a faint noise outside your window. You are unsure whether you actually heard something, so that sensory evidence has relatively low reliability. Now imagine hearing glass shatter three feet away. Much stronger evidence. But confidence can also exist in prior beliefs. If a person has an extremely strong prior belief that rejection is imminent, ambiguous evidence may have difficulty changing that prediction. This becomes particularly important when thinking about persistent anxiety or trauma-related expectations. Sometimes the issue is not simply what a person predicts. It is how much confidence the system places in that prediction relative to contradictory information.

What Is Prediction Error?

Prediction error is one of the most important concepts in predictive processing, and it is also one of the easiest to misunderstand. A prediction error occurs when incoming evidence differs from what was predicted. In simple terms: you predicted X, you observed Y, and the mismatch between them is the prediction error.

Suppose your friend normally replies within ten minutes. Today you message them, and three hours pass. That outcome differs from the pattern you expected, which is a prediction error. Now imagine they eventually reply: "Sorry! My phone died." You have acquired new information that may explain the discrepancy without requiring you to fundamentally revise your model of the relationship. But imagine instead that delayed responses begin happening repeatedly. Your previous model may eventually need updating. Prediction error provides information that allows learning to occur.

Prediction Error Does Not Mean Danger

This distinction is extremely important. Prediction error is not the same thing as threat. Imagine you are absolutely certain that someone is going to criticize you, and then they criticize you. Your prediction was accurate, so prediction error may actually be low, but the situation is still threatening. Now imagine you expect criticism and instead they respond warmly. That may produce a large prediction error, but the unexpected event is positive. So it would be inaccurate to say that prediction error itself "activates threat." Threat responding depends on what the brain infers about the probability, significance, cost, and consequences of what is happening. Prediction error is primarily information about a mismatch between prediction and outcome, and that mismatch can contribute to learning.

Bayesian Updating: How Predictions Change

Predictive processing is closely related to Bayesian inference, but you do not need to understand the mathematics to understand the basic idea. Your brain begins with prior beliefs about what is likely. It receives new evidence. The evidence is evaluated according to its reliability. The prior and the evidence are combined, and the resulting belief becomes the basis for future predictions. In simplified form: prior belief plus new evidence leads to updated belief, and that updated belief becomes the future prediction. This process happens continuously.

The important part is that updating depends on the relative strength and reliability of both the previous model and the new evidence. One contradictory experience may not be enough to overturn years of learning, and from an adaptive perspective, that makes sense. If you have seen the sun rise thousands of mornings in a row and one morning somebody tells you, "The sun probably isn't coming up tomorrow," your brain should not immediately throw away the model. Stable models are useful. The problem emerges when a previously useful model remains highly influential after the environment has changed.

Why Old Predictions Can Survive New Evidence

This may be one of the most clinically important implications of predictive processing. People often ask: "If I know something isn't true anymore, why do I still react like it is?" The answer is that knowing something consciously and revising a deeply learned predictive model are not necessarily the same process.

Imagine growing up with a caregiver whose anger was unpredictable. Small changes in tone might have preceded conflict. Silence might have meant something was wrong. A particular facial expression might have reliably predicted criticism. Over hundreds or thousands of interactions, those patterns provided data, and your brain learned them. Years later, you may intellectually understand that your partner, friend, or therapist is not your caregiver, but similar interpersonal information may still increase the inferred probability of an old outcome. A quiet partner becomes "something is wrong." A delayed response becomes "I am being rejected." Someone asking for space becomes "I am going to be abandoned." Someone expressing disappointment becomes "I have failed."

These interpretations are not necessarily random irrationalities. They may reflect predictions generated from a model that once fit the available evidence extremely well. The problem is that yesterday's model is being used to interpret today's environment.

Predictive Processing and Trauma

Predictive processing offers a very different way of thinking about trauma. Popular trauma language has often suggested that traumatic experiences become physically "stored" somewhere in the body and later need to be released. But contemporary predictive accounts offer another possibility.

The body unquestionably participates in trauma. Heart rate changes. Breathing changes. Muscle tension changes. Hormonal and autonomic processes change. Interoceptive sensations can subsequently influence how situations are interpreted. But this does not require the body to function as an archive containing traumatic experiences. Instead, trauma-related reactions can be understood as recurrent brain-body loops involving prediction, perception, memory, attention, physiological regulation, action, and new sensory evidence.

In a 2026 paper, Steven Kotler, Michael Mannino, Glenn Fox, and Karl Friston argued explicitly for understanding trauma through predictive coding rather than bodily storage. From this perspective, what persists is not a "score" physically embedded in tissue. What persists is a predictive architecture that continues to assign substantial probability to danger. The distinction is profound. Instead of asking, "Where is the trauma stored?" we can ask, "What is the system predicting?" and "What evidence would be necessary for that prediction to change?"

Your Body Still Matters

Rejecting literal bodily storage does not mean pretending that trauma is "all in your head." That would simply recreate the same mind-body split in the opposite direction. The brain and body continuously exchange information. Changes in heart rate, respiration, pain, temperature, gastrointestinal state, and other physiological processes provide interoceptive evidence that can influence ongoing inference.

Consider anxiety. Your brain predicts possible danger, and physiological mobilization occurs. Your heart begins beating harder, and now your brain receives information from the changed body. That information becomes additional evidence. Depending on context and prior learning, the increased heartbeat might be interpreted as "I am excited" or "I just exercised" or "I drank too much coffee" or "I am having a panic attack" or "Something is wrong." Same heartbeat. Different inference. The body provides information, and the brain interprets its significance.

Why Context Changes Everything

One of the most powerful implications of predictive processing is that signals do not have universal meanings. A raised voice is not inherently dangerous. A soft voice is not inherently safe. Eye contact is not inherently reassuring. A racing heart is not inherently evidence of threat. Silence is not inherently rejection. Their significance depends partly on context, learning history, and expected consequences.

This becomes particularly important in relationships. A particular facial expression may mean nothing to one person and carry enormous significance for another, because people do not enter relationships with identical predictive histories. Each person brings a model built through previous relational experience, and those models shape what the same cue means to each of them.

Predictive Processing and Relationships

Human beings are unusually dependent on relationships. We spend an exceptionally long developmental period relying on caregivers for protection, nourishment, regulation, and access to the social world. That means early relationships provide an extraordinary amount of information. Children repeatedly encounter questions such as: What happens when I need someone? What happens when I cry? What happens when someone becomes angry? Does distress lead to comfort? Does closeness remain available? Does conflict lead to repair? Are other people predictable? Can I influence what happens? What does someone's silence usually mean? What happens when I make a mistake?

The child does not need to formulate these questions consciously. The patterns themselves provide information, and repeated relational experiences gradually make some outcomes more probable than others. Those learned models can then influence later relational inference. This is where predictive processing becomes especially relevant to attachment and relational trauma.

Schemas Can Be Understood as Models of What Relationships Mean

Psychology has long used the concept of schemas to describe organized structures that shape how people interpret themselves, other people, and the world. Predictive processing provides another way of thinking about this process. A relational schema may help organize predictions about what is likely to happen under particular interpersonal conditions. For example: "If I depend on someone, they will eventually let me down." "If someone sees my flaws, they will reject me." "If somebody is upset, I need to fix it immediately." "If I become vulnerable, I will lose control." "If someone pulls away, abandonment is coming."

These conscious statements are not themselves the entire predictive mechanism. They are verbal representations of deeper learned relational models. The important question becomes not simply "What does this person believe?" but "What outcomes has this person's relational system learned to treat as probable?"

The Same Relational Cue Can Produce Different Predictions

Imagine two people receive the same message from their romantic partner: "Can we talk tonight?" Person A thinks, "Sure. I wonder what they want to discuss." Person B immediately thinks, "They're leaving me." Person C thinks, "Oh God. What did I do wrong?" Person D thinks, "I don't want to deal with this." Same message. Different inference. Those responses are not contained inside the words "Can we talk tonight?" They arise from the interaction between current evidence and existing models. This is why relational safety cannot be reduced to a universal catalogue of supposedly "safe" cues. Human beings learn what interpersonal signals mean.

Predictive Processing Helps Explain Attachment Patterns

Attachment strategies can also be understood within this broader predictive framework. If closeness has repeatedly predicted comfort and responsiveness, proximity may remain an attractive strategy. If closeness has repeatedly predicted intrusion, criticism, rejection, or loss of autonomy, distancing may become more adaptive. If availability has been inconsistent, increasing proximity-seeking may become useful. These patterns do not require us to assume that someone is consciously deciding, "My attachment strategy today will be deactivation." Their behaviour may instead represent an action policy shaped by learned expectations about what different interpersonal actions are likely to produce.

What Is Active Inference?

Predictive processing is not only about perception. Organisms act, and active inference extends the predictive framework into action selection. Sometimes the system updates its beliefs to fit the world. Sometimes it acts on the world and changes what happens next.

Suppose you are uncertain whether someone is upset with you. You could ask them, withdraw, send another message, seek reassurance, watch their facial expression, make a joke, apologize, become defensive, or end the conversation. Each behaviour changes the information you encounter next. Your actions therefore help generate the evidence that your brain subsequently uses for inference. This creates a potentially powerful feedback loop.

How Our Predictions Can Accidentally Confirm Themselves

Imagine someone strongly predicts rejection. They notice ambiguity in a relationship, and they withdraw before the other person can reject them. The other person experiences the withdrawal as disinterest and eventually stops reaching out. The original person observes: "See? People always leave." The prediction helped shape the behaviour. The behaviour changed the environment. The changed environment produced new evidence. And the new evidence appeared to confirm the original prediction.

This does not mean people simply "manifest" their trauma. It means perception and action are connected. We do not passively observe relationships. We participate in them, and that is one reason entrenched relational predictions can become remarkably persistent.

How Predictive Models Change

If old learning contributes to current predictions, then an obvious question follows: how do we update them? Not by simply telling ourselves that the old prediction is wrong. Information matters, but deeply learned models usually require evidence.

If someone predicts, "When I express a need, people become angry," then understanding the origin of that prediction can be useful. But the model begins receiving different evidence when the person repeatedly expresses needs and encounters responses that contradict the predicted outcome. The person predicts anger, and the other person remains engaged. The person predicts rejection, and the relationship survives disagreement. The person predicts that setting a boundary will destroy the connection, and the boundary is respected. The person predicts that conflict means abandonment, and conflict is followed by repair. Now the predictive system has something new to work with. Not reassurance. Data.

Healing May Require More Than One Corrective Experience

One safe interaction usually does not erase years of learning, and that is not evidence that the brain is defective. It is evidence that stable models are difficult to change. A brain that rewrote its model of reality every time something unexpected happened would be useless. Updating therefore depends on factors such as how established the previous model is, how reliable the new evidence appears, how often contradictory evidence occurs, whether the new experience generalizes across contexts, how much confidence is assigned to the original prediction, and whether behaviour allows contradictory evidence to actually be encountered.

This is why repetition matters. A reliable relationship may gradually provide enough evidence for an old relational model to become less dominant. The brain learns: that prediction was once useful, but it is not always the best explanation anymore.

Predictive Processing Does Not Mean "Everything Is in Your Head"

There is another misunderstanding worth addressing. If the brain interprets sensory information, that does not mean reality is imaginary. There is a real external world, and other people actually behave in ways that are caring, inconsistent, coercive, trustworthy, abusive, predictable, or dangerous. Predictive processing does not mean "your perception created everything." It means the brain must infer the causes and significance of the information it receives from that world.

Good inference depends on both sides of the equation: previous learning and current evidence. Sometimes prior beliefs distort interpretation. Sometimes the environment really is dangerous. Sometimes anxiety generates a false alarm. Sometimes the alarm is accurate. A scientifically useful framework must leave room for both.

Is Predictive Processing Proven?

Predictive processing is enormously influential, but it should not be presented as a completely settled description of brain function. There is substantial experimental evidence for prediction, prediction-error processing, and experience-dependent updating across perceptual and learning systems. Researchers have also proposed neural implementations of predictive coding and developed increasingly sophisticated active-inference models.

However, scientists continue to debate how broadly predictive-processing explanations should be applied, how particular computational concepts map onto neural mechanisms, and whether some findings are explained equally well by alternative models involving attention, feed-forward processing, or reinforcement learning. That scientific debate is healthy. Predictive processing is powerful precisely because it generates mechanisms that can be tested rather than functioning as a metaphor that explains everything after the fact.

Why Predictive Processing Matters

Predictive processing gives us a different lens through which to understand human experience. We are not simply reacting to what is happening. We are continuously interpreting what is happening through models built from what happened before. That helps explain why two people can interpret the same situation differently, why old relational experiences can influence new relationships, why ambiguous cues can trigger powerful reactions, why conscious knowledge does not always immediately change emotional responses, why avoidance can maintain old predictions, why corrective experiences can gradually alter expectations, why bodily sensations can amplify threat without literally storing trauma, and why relationships can become powerful sources of new evidence.

Most importantly, predictive processing shifts the question. Instead of asking, "Why am I reacting like this when nothing is happening?" we might ask, "What does my brain think is happening?" And then: "What experiences taught it to make that prediction?" And eventually: "What evidence would allow it to learn something different?"

The Past Does Not Have to Be Present to Influence the Present

Your childhood does not need to be physically stored in your body for it to influence you decades later. The past can persist through learning. Through memory. Through expectations. Through attention. Through interpretation. Through prediction. Through the actions those predictions generate. A model built in one environment can continue operating in another, and because the brain is capable of updating, those models are not necessarily permanent.

The past supplies the data. The brain builds the model. The present supplies new evidence. And under the right conditions, the model can change. That may be one of the most important things predictive processing can teach us about trauma, relationships, and healing. The goal is not to erase the past. It is to help the brain discover that the past is no longer the only available prediction.

Frequently Asked Questions About Predictive Processing

What is predictive processing in simple terms?

Predictive processing is a framework proposing that the brain uses previous experience to generate predictions about the causes of sensory information, compares those predictions with incoming evidence, and updates its model when important discrepancies occur.

What is predictive coding?

Predictive coding generally refers to computational models in which higher levels of a processing hierarchy generate predictions while differences between predicted and incoming signals contribute prediction-error information. Predictive coding is closely related to the broader predictive-processing framework, although the terms are not always used identically.

What is a prediction error?

A prediction error is a mismatch between what a model predicted and what was actually observed. Prediction errors can provide information for learning and updating. They do not automatically mean that something is dangerous.

What is a prior in predictive processing?

A prior represents the probability assigned to possible states or explanations before the newest evidence has been incorporated. Priors can be shaped by past experience, learning, and context.

What does precision mean in predictive processing?

Precision refers broadly to the confidence or reliability assigned to information. The relative precision of prior beliefs and incoming evidence influences how strongly new information changes an existing model.

What is active inference?

Active inference extends predictive accounts into perception-action loops. Instead of only updating beliefs in response to sensory information, an organism can act to change the world or gather new information. Behaviour therefore influences the evidence available for future inference.

How does predictive processing relate to trauma?

Predictive approaches suggest that trauma-related difficulties may involve persistent threat expectations and altered interpretation of sensory or interoceptive evidence. Past danger can therefore influence present inference without requiring trauma to be literally stored in bodily tissue.

Does the body store trauma?

There is no established biological mechanism demonstrating that traumatic experiences are stored independently in muscles, fascia, or other non-neural bodily tissues. The body participates extensively in trauma through brain-body regulation, physiological responses, and interoceptive feedback, but memory, inference, and the representation of experience depend on neural processes.

Can old predictions change?

Yes. Predictive models can be revised when sufficiently meaningful and reliable new evidence contradicts previous expectations. However, long-established models may require repeated experiences across time and context before substantial updating occurs.

---

References and Further Reading

Bastos, A. M., et al. (2012). Canonical microcircuits for predictive coding. *Neuron, 76*(4), 695–711.

Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. *Behavioral and Brain Sciences, 36*(3), 181–204.

Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11*, 127–138.

Friston, K., et al. (2015). Active inference and epistemic value. *Cognitive Neuroscience*.

Friston, K., et al. (2017). Active inference: A process theory. Neural Computation.

Hodson, R., Mehta, M., & Smith, R. (2024). The empirical status of predictive coding and active inference. Neuroscience & Biobehavioral Reviews.

Kotler, S., Mannino, M., Fox, G., & Friston, K. (2026). The body does not keep the score: Trauma, predictive coding, and the restoration of metastability. Frontiers in Systems Neuroscience, 20, 1812957.

Rao, R. P. N., & Ballard, D. H. (1999). Predictive coding in the visual cortex. Nature Neuroscience, 2, 79–87.

Seth, A. K., & Friston, K. J. (2016). Active interoceptive inference and the emotional brain. Philosophical Transactions of the Royal Society B, 371, 20160007.

Sladky, R., Kargl, D., Haubensak, W., & Lamm, C. (2024). An active inference perspective for the amygdala complex. Trends in Cognitive Sciences, 28(3), 223–236.