Controlled Hallucination: Three Theories Explaining How the Brain and Life Work
In the past decade, several theories have emerged that distill generations of interdisciplinary scientific experience into accessible frameworks. Perception, cognitive biases, adaptive strategies—all share a common principle.
Consider the famous painting by Belgian artist René Magritte. It depicts a smoking pipe, with the French caption “This is not a pipe.” When you compare the image and the text, you might experience a subtle mental conflict between expectation and perception. This internal “clash with reality” was described by American social psychologist Leon Festinger in 1957 as the “Theory of Cognitive Dissonance.” The musical origin of the term (from Latin dissonantia—disagreement, discord, inconsistency) intuitively hints at the core idea: a sharp, jarring note disrupts the smooth, harmonious process of perceiving reality.
Cognitive dissonance isn’t always a psychological conflict that leads to frustration; it’s more of a spectrum of sensations, from confusion and uncertainty about what to do next, to mild puzzlement—like the riddle that the brave soldier Schweik used to baffle forensic doctors: “There’s a four-story building, each floor has eight windows, the roof has two dormer windows and two chimneys, each floor has two tenants. Now tell me, gentlemen, in what year did the doorman’s grandmother die?”
In popular culture, cognitive dissonance is often seen only as psychological conflict, overlooking the second part—the mechanism for resolving this conflict, or reconciling expectations with reality. Festinger’s theory includes not just the stress and discomfort from new, contradictory information, but also the ways we reduce this dissonance.
Where Do Expectations Come From?
Recall the dualistic model of visual perception: sensory stimulation activates processes in the brain. Imagine the complex chain of events, from a photon hitting the light-sensitive cells (rods and cones) in the retina, to the assembly of a complex visual image in the higher brain regions, placed in a specific context. Now scale this up to all available visual stimuli. And that’s just vision—one of many “channels” of incoming information about the world. The mind can’t possibly process the overwhelming flood of sensory signals we receive every second. If perception worked according to outdated views, life simply wouldn’t exist—it’s impossible to keep up with the ever-changing train of reality. If you can’t keep up, you have to anticipate.
Imagine the brain, locked inside the skull—it sees and hears nothing directly. It just receives a stream of signals and must guess what’s happening outside. In fact, it doesn’t just guess, it must predict—so the body can prepare and react in time.
Predictive Processing Theory
The brain functions as a multi-level prediction machine, where a top-down stream of predictions (what we expect from the world) is constantly compared and adjusted against a bottom-up stream of sensory data (what our senses perceive). The top-down stream is everything we know about the world—our best heuristics (quick, simplified reasoning for efficiency), our prior beliefs and expectations, all our previous experience—from E = mc² to “London is the capital of Great Britain.” The bottom-up stream consists of three parts: exteroception (what’s happening outside the body), interoception (what’s happening inside the body), and proprioception (the position and movement of the body), all combined into a multimodal model. All our knowledge forms the foundation for constructing predictions about what we should feel.
How It Works
The brain generates mental models (called generative models) that predict what the sensory apparatus should receive as input. These predictions are called prior beliefs. Predictive models are layered in a hierarchy, reflecting the brain’s organization from lower to higher, from simple to complex—higher levels send predictions downward, and lower levels send incoming sensory data upward.
If top-down predictions don’t match bottom-up sensory data, a prediction error occurs, and the model either updates its priors or ignores the incoming data as noise, keeping its previous assumptions.
Example
Think about vision. We never see the world as it appears on the retina. First, the image on the retina is inverted (the eye is a camera obscura, and the brain flips the image). Second, it’s blurry at the periphery due to uneven distribution of visual cells. Third, there’s a layer of blood vessels over the retina (inverted retina). Fourth, there’s a blind spot where the optic nerve exits. Plus, our eyes make countless tiny, rapid movements (saccades), “scanning” the space. Yet we enjoy a full-color, three-dimensional, stabilized image, already interpreted. Our brain even predicts light and shadow, as in visual illusions.
How the Brain Uses Bayesian Statistics
A critical parameter for both streams is the level of precision. We care not just about the data, but also its accuracy or probabilistic “weight.” A bottom-up signal like “there’s an elephant in front of you” has high weight; a vague silhouette in the fog has low weight. A top-down prediction that water is probably wet has very high weight; “the Dow Jones Index should drop a couple of points because of rising diaper prices” has very low weight.
Both streams—bottom-up and top-down—constantly interact at every level, and this ongoing probability adjustment can be described using Bayesian statistics. Bayes’ theorem is about determining the probability of an event based on prior events. In simple terms—if the shot glass you drank from last night with some shady people smells like acetone, you’ll probably feel bad in the morning.
In a graph of Bayesian inference with a Gaussian distribution, Expectation is our prior, Reality is the actual data, and Estimate is our perception—a compromise between the two. The X-axis is any parameter we’re trying to predict; the Y-axis is the probability of each value. Uncertainty is the variability of expectations; Noise is the variability of precision. The process:
- There’s an expectation (prior), whose precision depends on uncertainty.
- There’s sensory input (likelihood), or reality, whose precision depends on noise.
- Between expectation and reality is what we perceive—the posterior. We adjust our prior based on the new signal and get a posterior probability.
A Simple Example
You decide to skip work, assuming your boss is on a business trip and won’t notice. That’s your prior. The accuracy of your prediction depends on uncertainty—are you sure he left? Did anything change? Where’s the info from? The less you know, the higher the uncertainty, the less accurate the prediction.
You start gathering information—ask colleagues, managers, even check his flight online. The accuracy of this data depends on noise—did you hear it in the break room (low precision), from your project manager (medium precision), or from his assistant who bought his tickets and saw him off (high precision)?
Your final decision is the posterior—a balance between your prediction and what you learned. If your predictions and data were accurate, your unsanctioned day off goes unnoticed. Prediction error is small. But if you relied on vague assumptions and random data, your prediction fails, the boss just stepped out, your “trusted sources” rat you out, and you get in trouble. Next time, you’ll analyze more carefully and update your beliefs accordingly.
Surfing Uncertainty
Now that we’ve covered the “Bayesian brain hypothesis,” let’s expand our understanding of how predictions interact with incoming data. There are three scenarios:
- If predictions roughly match sensory data, everything is calm—predictions come true, and all is well.
- If low-precision sensory data contradicts high-level predictions, Bayesian math may decide the predictions are correct and the data is faulty. Lower levels “fit the data” to the prediction, and higher levels stick to their expectations.
- If high-precision sensory data conflicts with predictions, Bayesian math concludes the predictions are wrong. The involved neurons signal “Alert! Something’s off!” The greater the mismatch and the higher the data’s precision, the bigger the surprise—the louder the internal alarm.
Each level’s main task is to minimize surprise. Ideally, the brain predicts the world so well that surprises are rare, because each surprise triggers a flurry of activity to update the generative model until calm is restored. All this happens in fractions of a second. Lower levels bombard higher levels with data, which adjust their hypotheses and send predictions back down. After countless cycles, everything is more or less predicted and expected—until the next crisis.
Andy Clark, in his book Surfing Uncertainty, compared this predictive process to surfing: “To act quickly and flexibly in an unstable and noisy world, the brain must become a master of prediction—riding the waves of noisy and ambiguous sensory stimulation, trying to stay ahead. An experienced surfer stays in the ‘pocket’: close, but just ahead of where the wave breaks. The wave carries you, but doesn’t catch you. The brain’s task is the same. By constantly trying to predict incoming sensory signals, we can learn about the world, think, and act in it.”
The result is perception, which predictive processing theory calls a “controlled hallucination.” We don’t perceive the world as it is, but our predictions about it, corrected by incoming data. As Anil Seth said in his TED talk, it’s “our brain’s best guess.”
Active Inference
We’ve explored the leading theory of brain function—predictive processing—to understand where our expectations come from. Now we can grasp what’s meant by the “Bayesian brain.” After the work-skipping and factory fire examples, the following diagram should be clear. It “packs” the predictive processing process to show which processes happen in the brain and which outside. The brain builds an internal model of the world, makes predictions, compares them to incoming information, updates its worldview, and the cycle repeats. Note the background color: everything on beige is the external environment, everything on white is internal. Sensory data and actions are at the boundary.
Let’s look at another diagram. It’s almost the same: world model, expectations/forecast, prediction, prediction error, model update. Forecasting is just another word for “prediction.” Here, we add a boundary between the system (internal) and the external world, shown as a dashed line. All the processes we’ve discussed happen inside the system, while actions and sensory data are at the boundary with the outside world.
To simplify further: Sensory states are sensations, our sensory input. Active states are actions or behavior. Internal states are our internal states, the result of all these processes. External states are the states of the surrounding world, our environment.
The states of the world (S) determine our sensory states (o), which, after internal processing, become our internal states (s), which determine our active states (a), which change the world, closing the causal loop. This is called active inference and is essentially how autonomous agents function in a dynamic environment.
We Are All Markov Blankets
The term “Markov blanket” was coined by Israeli-American scientist and philosopher Judea Pearl, who works on probabilistic approaches to AI and Bayesian networks. Andrey Markov (senior), whose name it bears, was a pioneer in the study of stochastic (random) processes and probability theory. His son, also Andrey Markov (junior), was an equally outstanding mathematician, giving us Markov chains and Markov processes.
The “blanket” or “boundary” of Markov is a concept that goes far beyond consciousness and neuroscience—it’s even more fundamental. Absolutely anything exists as a Markov blanket. Without it, you couldn’t draw a boundary between something and everything else. If something doesn’t have a Markov blanket, it simply doesn’t exist. Everything in our world is a Markov blanket, “nested” within other Markov blankets, as far as scaling allows.
“If the Markov blanket is minimal, meaning it can’t drop any variable without losing information, it’s called a Markov boundary.” This is the very boundary where we end and the world begins, and vice versa.
Without any of these components—sensory, internal, or active states—we wouldn’t exist as autonomous subjects. Our Markov boundary protects us from the causal complexity of the world.
The Free Energy Principle
What are all living organisms doing in this chaotic, unpredictable, and, most importantly, non-equilibrium world? First and foremost—simply existing, maintaining their boundaries that separate them from the environment and preserve some internal structure and processes. To do this, they must perceive the world (Bayesian math), represent it internally (a generative model), predict (hierarchical predictive processing), and act (active inference) to update their internal model.
We “probe” the world through active inference, create its internal model via predictive processing, and update this model (learn) using Bayes’ theorem. The final, perhaps key, element: all these processes can be reduced to optimizing a single parameter—the difference between expectation and reality. All our complex adaptive strategies boil down to reducing uncertainty. This parameter is called variational free energy.
In essence, we’ve just encountered the Free Energy Principle, which is now considered as explanatory as the theory of evolution by natural selection.
Karl Friston, the author of the Free Energy Principle and predictive processing theory, has a citation index higher than Einstein, with 1,200+ scientific publications. Anyone even slightly familiar with his work is left with the impression that he’s incredibly brilliant.
It’s hard not to appreciate the elegance of the idea that all living things are generators of predictions about the states of the world, engaged in self-maintenance and self-organization by separating themselves from the environment and minimizing their prediction errors.
Drawing Parallels and Conclusions
Festinger’s theory of cognitive dissonance, describing the conflict between expectations and reality and the mechanisms for resolving it, was a precursor to newer, more complex, and far-reaching theories. Starting with explanations of mental processes, they evolved to address the very essence of adaptive strategies in all living things. A good theory is like a prism—it lets us see what’s hidden from the naked eye. The world we perceive is a generative model, built on our brain’s guesses about what’s happening outside—a controlled hallucination. We can’t escape this fact, but we can listen more closely to what our senses tell us and not be afraid to update and complicate our worldview. Only those who never learn or try new things avoid making mistakes.