When we talk about the senses, we intuitively imagine a fairly simple process. The world emits or reflects certain physical signals, receptors register them, the nervous system transmits them, and the brain then “decodes” them and assembles a picture of reality. In that model, the brain is a kind of extremely powerful processor that receives input data and reconstructs what is around us.
The problem is that, from what we know today, the brain probably does not work quite like that.
It is not merely a decoder.
It is much closer to the truth to say that the brain continuously maintains an internal model of the world, uses previous experience to predict what will happen next, and then uses new sensory input primarily to correct that model. What we perceive, therefore, is not a complete reconstruction of the world built from scratch at every new moment, but a constant update of an already existing representation.
This idea lies at the heart of what is known as predictive processing.
The World Does Not Have to Be Recomputed from Scratch Every Time
The easiest way to understand this idea is through vision.
Our retina does not send the brain a complete photographic frame. The visual system receives limited, fragmented, and constantly changing information. The eyes move several times per second, retinal images change position, part of the scene disappears every time we blink, and peripheral vision has much lower spatial resolution than central vision. Yet we do not experience the world as a sequence of jerky and incomplete frames.
We experience a stable scene.
One possible explanation is that the brain does not create a new image of the world after every eye movement. It maintains an internal model of the scene and changes only what new data indicate is no longer correct. In other words, the system does not work like a graphics card rendering every frame from scratch. It is more like a system that maintains an existing state and constantly applies an update.
If nothing unexpected has happened, there is no need to process every detail again. If something changes, then that difference becomes informationally important.
This brings us to one of the key concepts of predictive processing: prediction error.
Prediction Error: The Information Is in the Difference
Suppose we are sitting in a room and looking at a door. Based on experience, context, and the previous moments, the brain already “knows” a great deal about the scene: where the door is, what color it is, where the wall is, and where the light is coming from.
If nothing changes, sensory input mostly confirms the existing model. But if the door suddenly opens, a large difference appears between what was expected and what was actually received.
That difference is the prediction error.
In a simplified model, higher levels of the nervous system send predictions downward about what lower levels should register, while lower levels send upward information about the difference between expected and actual input.
The brain therefore does not need to transmit complete information about everything it senses. It is far more efficient to emphasize what does not fit.
For an engineer, this is a very intuitive idea. In many information systems, there is no need to transmit the complete state continuously if we can transmit only the difference from the previous state.
Biology appears to have found a very similar principle.
Prediction Is Not Guessing
When we say that the brain predicts, we do not mean that it is constantly “guessing” what will happen in the sense of conscious speculation. Most of these processes occur automatically.
Prediction emerges from the statistics of previous experience.
If we have seen faces thousands of times, our visual system develops very strong expectations about what a face should look like. If we hear the beginning of a familiar word, the auditory system already has several likely continuations. If we walk on a flat surface, the motor system predicts what proprioceptive signal should arise with the next step.
The brain, in other words, tries to stay one step ahead of its own senses.
That allows it to react faster than it could if every event first had to be fully analyzed and only then answered.
Why We See What We Expect
One of the most interesting consequences of the predictive model is that perception is not determined only by the signal coming from the senses.
It is also influenced by what the brain expects to see or hear.
That is why context has such a powerful effect on perception. An ambiguous symbol may look like one letter inside one word and like another in a different context. Speech that is almost unintelligible in noise can suddenly become clear when we know what the speaker may be saying. A familiar melody can be recognized even when the signal is very weak or partially masked.
The sensory signal has not disappeared. But its meaning is determined by a combination of input and expectation.
In that sense, perception is a compromise between what comes from the outside and what the brain already knows.
A Hierarchy of Predictions
Predictive processing becomes especially interesting when we view it as a hierarchical system.
At lower levels, the brain may predict simple features: edges, contrasts, sound frequencies, or local motion changes. Higher levels may predict objects, words, events, or entire situations.
If we see something four-legged moving through a park, lower levels process lines, colors, and motion, while a higher level may quickly form a hypothesis: a dog.
When additional data confirm a muzzle, a tail, and a characteristic movement pattern, prediction error decreases. If, however, it turns out to be an unusually shaped bag someone is carrying, the system has to correct its model.
Perception thus becomes a continuous cycle:
prediction → comparison with input → prediction error → model correction → new prediction
That looks much more like a closed-loop control system than a simple decoding chain.
The Brain as a Control System
In control engineering, we know systems that have a target value, measure the current state, calculate an error, and then correct their behavior based on that error.
Of course, the brain is not a PID controller.
But the general principle of feedback is strikingly similar.
The internal model provides the expected state. The sensory system provides the current state. The difference between them creates an error signal, and that signal changes the model.
The difference is that the brain is vastly more complex: there is no single controller, but an enormous hierarchy of interconnected loops operating simultaneously across different spatial and temporal scales.
At one level, the brain predicts that with the next step we will feel pressure from the ground beneath the foot. At another, it predicts that the person we are walking toward will remain in the same place. At a third, it predicts that the conversation we are having will continue in a meaningful way.
All of these predictions can be confirmed or corrected at the same time.
When Is an Error Important, and When Is It Not?
Not every prediction error is equally important.
The brain cannot react to every tiny change in every sensory channel, because that would overwhelm it with information.
The system must therefore estimate the reliability of individual signals.
If a sensory input is clear and reliable, it can strongly correct the existing model. If the signal is weak, noisy, or ambiguous, the brain relies more heavily on previous experience.
In mathematical formulations of predictive processing, this is often described as precision weighting — assigning different weights to signals according to their estimated precision.
And this is where the bridge toward attention begins. If one signal receives greater weight than another, it will have a stronger influence on what we perceive and how we respond.
In the next part of the triptych, we will focus precisely on the question of who, or what, determines that neuronal gain.
Phantom Limbs: When the Body Model Outlives the Body
One of the most fascinating examples of the relationship between internal models and sensory input is the phantom limb phenomenon. A person whose arm or leg has been amputated may still very clearly feel the presence of that limb, its position, movement, and even pain.
This does not mean that phantom limbs can be explained exclusively through predictive processing. Peripheral nerve activity, changes in the spinal cord, cortical reorganization, and other mechanisms also play important roles. But the predictive model offers a very interesting framework for understanding part of the phenomenon: the brain’s internal representation of the body does not disappear at the moment when the corresponding peripheral sensory input disappears.
For years, the brain maintained a model in which that arm existed, moved, and generated certain proprioceptive and sensory information. After amputation, the sensory data no longer match that model, but the model itself does not have to vanish immediately.
In a certain sense, the brain continues to expect a signal from a place from which normal input no longer arrives.
It is particularly interesting that some patients experience the phantom limb as stiff, trapped, or locked in an uncomfortable position. Therapies such as mirror therapy attempt to use visual input to modify the internal body model and, in some patients, reduce phantom pain.
Here, the logic of prediction becomes almost tangible: change the information at the input, and perhaps you can change the model the brain maintains.
Placebo and Nocebo: Expectation That Changes the Signal
The placebo effect is often oversimplified as if the patient had merely “imagined” an improvement. That is far too crude. Expectation can genuinely influence neural pain processing, autonomic responses, and other physiological processes.
If a person believes they are receiving an effective analgesic, that expectation itself can activate brain systems involved in pain modulation. This does not mean that placebo can cure every disease, nor that every symptom is caused by expectation. But it does show that the brain does not process bodily signals like a passive instrument.
It interprets them within an existing model.
Nocebo is the other side of the same process. If a person strongly expects pain, a side effect, or worsening of symptoms, the perception of those symptoms can be amplified. The expectation does not necessarily change the peripheral source of the signal, but it can change the way that signal is weighted and interpreted.
This is another reason why the distinction between an “objective signal” and a “subjective experience” is not a simple boundary. Subjective experience emerges within a nervous system that already has a model of what it expects to happen.
Perception as a Controlled Construction
The idea that the brain actively constructs perception is sometimes misunderstood as a claim that the world we see is not real.
That is not the point.
The external world obviously exists and continuously constrains our predictions. If the model becomes too inaccurate, sensory data force it to change.
But there is no simple line between the world and our conscious experience. There is an entire system of filtering, prediction, comparison, and correction.
What we experience is the best model of the world that our nervous system can build from what it already knows and what it is currently registering.
That is why the same ambiguous image may look different to different people, why a familiar word can be recognized even when part of the acoustic signal is missing, why we can anticipate a movement before it is fully executed, and why we sometimes react to an event before the conscious mind has managed to explain what actually happened.
What Happens When There Is Almost No External Signal?
If the brain already possesses a mechanism that continuously generates models and predictions, a natural question follows: what happens when sensory information becomes very weak?
We can close our eyes and imagine the face of someone we know. We can “hear” a melody in our minds that is not physically playing anywhere. We can imagine a room we have never entered.
During sleep, we can experience entire worlds, people, and events without corresponding external sensory stimulation.
The brain, then, clearly has the ability to generate content from within.
The difference between perception and imagination may therefore not be that one uses an internal model and the other does not.
Both do.
The difference may lie more in how strongly that model has to remain constrained by external sensory data.
During normal waking perception, the outside world constantly corrects the internal model. During imagination, that external supervision weakens. During dreaming, it may become weaker still.
This is where the space for creativity, fantasy, and dreams begins — but also the question of what happens when internal predictions gain too much weight relative to sensory input.
That is where we will later arrive at hallucinations.
The Brain Does Not Decode the World — It Negotiates with It
Predictive processing is not a finished and universally accepted “theory of everything” about the brain. There are different formulations of the model, open questions, and serious debates about how broadly the principle can be applied.
But its central idea is extremely powerful.
The brain is not a passive receiver waiting for the senses to deliver a complete description of reality.
It constantly tries to predict what it will receive next.
When the prediction is good, sensory data largely confirm the existing model. When a significant difference appears, the error becomes information and the model has to change.
Perception therefore emerges from a continuous exchange between expectation and correction, between the internal model and the external world.
Perhaps that is why the most accurate statement is not that the brain decodes the world.
It is much closer to say that it continuously negotiates with it.
And in the next part of the triptych, we will come to the question that follows naturally from this story: among the enormous number of signals passing through the nervous system at every moment, why do some receive enough amplification to enter consciousness, while others remain in the background?
Leave a Reply