How the Brain Models the World
THE PREDICTIVE MACHINE
How the Brain Models the World — and What That Means for How We Learn, Engineer, and Grow
By Joe McFadden
The Holistic Analyst
www.McFaddenCAE.com · McFadden@snet.net
I. The Fundamental Question
I want to start with a question that sounds simple but is not.
When you look at a coffee cup on a table across the room, what are you actually doing?
The obvious answer is: you are seeing it. Light enters your eye, hits the retina, travels as an electrical signal to the visual cortex, and you perceive the cup. Simple. Passive. Like a camera taking a picture.
But that is not what is happening. Not even close.
What is actually happening is something far stranger and far more interesting. Your brain is not passively receiving an image of that cup. Your brain has already generated a prediction of what is on that table — based on everything it knows about tables, about the room you are in, about coffee cups and the contexts in which they appear — before the light from the cup has even finished traveling to your retina. What you call seeing is not the arrival of information. It is the confirmation or correction of a prediction that was already in place.
You are not observing the world. You are modeling it. Constantly. Automatically. At every moment of your waking life.
This idea sits at the heart of one of the most important frameworks in contemporary neuroscience. It goes by several names: predictive coding, predictive processing, active inference. The scientist most closely associated with its mathematical formalization is Karl Friston at University College London, whose work on what he calls the free energy principle has become one of the most cited frameworks in all of neuroscience and cognitive science. But the intuition behind it is older, and in some ways it is something many of us have sensed without having the language for it.
The brain is not a passive recording device. It is an active prediction machine. And understanding what that means — really understanding it, at the level of mechanisms and implications — changes how you think about perception, learning, expertise, error, and growth. It changes how you think about engineering. And if you will stay with me through this journey, I believe it will change how you think about the work you do and the person you are becoming through doing it.
II. Eleven Million Bits Per Second
The brain receives approximately eleven million bits of information per second through its sensory organs. Eleven million. The eyes alone contribute around ten million of those bits. The other senses — hearing, touch, taste, smell, the internal senses that monitor the body's state — contribute the remainder.
Of those eleven million bits per second, the brain consciously processes approximately forty to fifty. Yes — fifty bits. Out of eleven million.
Pause on that for a moment. The ratio of information available to information consciously processed is roughly two hundred thousand to one. The overwhelming majority of what arrives at the sensory surface never reaches conscious awareness.
So the question is not: how does the brain process information? The question is: how does the brain decide which fifty bits out of eleven million are the ones worth bringing to consciousness? What is the selection principle?
The brain brings to consciousness the information that surprises it — the information that deviates from its predictions.
Everything else — the vast majority of the incoming stream — is simply confirmation of what the brain already predicted. And confirmed predictions do not need to reach consciousness. They are handled automatically, efficiently, without demanding the precious and limited resource of conscious attention.
This is why you do not consciously notice the weight of your clothes on your body right now — until I mention it, and suddenly you do. Your brain predicted that weight. The sensory signals confirming the prediction were suppressed before they reached consciousness. The moment I introduced a reason to attend to that signal, the prediction changed, and the signal came through.
This is why an experienced driver navigates a familiar route without conscious effort — the predictions about the road, the traffic, the mechanics of the vehicle are so well established that almost nothing surprises them. Conscious attention is freed for other things.
This is why a master chess player sees the board differently from a novice. It is not that the master processes information faster. It is that the master has built internal models of board states — through thousands of hours of play and study — that generate extremely accurate predictions about what patterns mean and what moves they imply. The novice sees pieces. The master sees possibilities. The difference is the quality of the internal model.
III. The Architecture of Prediction and Error
In the predictive coding framework, the brain is understood as a hierarchical system of prediction and error. At every level of the hierarchy, from the most basic sensory processing up through the highest levels of abstract thought, the same fundamental process is occurring.
Higher levels of the hierarchy send predictions downward — this is what I expect to receive. Lower levels compare the incoming sensory signal to the prediction and send back only the discrepancy — the prediction error. The part of the signal that matched the prediction is canceled out. It does not need to travel up the hierarchy because it carries no new information. Only the surprise propagates.
The brain is, at its computational core, a difference engine — constantly calculating the gap between what it expects and what it receives.
This is what learning is, at the neural level. Learning is the reduction of prediction error. Every experience that surprises you is an opportunity for the brain to update its internal models. Every surprise that is understood — that is followed by a revised prediction that better fits the world — represents genuine growth in the accuracy of those models.
And every experience that goes entirely as predicted — that confirms what the model already knew — while comfortable, while efficient, teaches the brain relatively little that is new.
IV. The Free Energy Principle
Karl Friston's contribution to this framework is the free energy principle — a mathematical statement that unifies predictive coding, action, and learning under a single principle. The details are technically demanding, but the core intuition is accessible and important.
Friston proposes that all self-organizing biological systems — including the brain, including entire organisms — act in ways that minimize what he calls free energy, which in this context is roughly equivalent to surprise. Systems that minimize surprise are systems that maintain themselves in the states they expect to be in. They resist the tendency toward disorder. They persist.
Two Strategies for Minimizing Surprise
The first is to update your internal model to better predict the world as it is. Learn more. Build more accurate representations. Reduce the gap between prediction and reality by improving the prediction.
The second is to act on the world to make the world conform to your predictions. To change reality to match the model, rather than changing the model to match reality.
Both strategies reduce the discrepancy between prediction and outcome. Both minimize free energy. And the brain uses both, constantly, often simultaneously. The dancer who has internalized the choreography acts to make their body conform to the predicted sequence of movements. The scientist whose experiment produces unexpected results either revises the hypothesis or adjusts the experimental conditions. The engineer who finds a discrepancy between simulation and test either refines the model or modifies the design.
Action and learning are not separate things. They are two expressions of the same underlying drive — the drive to close the gap between the world as modeled and the world as experienced.
This is what Friston calls active inference. The brain is not a passive receiver that occasionally sends motor commands. It is an active agent, constantly engaged in a loop of prediction, sampling, error detection, and revision. Active inference is not something the brain does when it is solving a problem. It is what the brain is doing all the time. Breathing is active inference. Walking is active inference. Conversation is active inference. Falling in love is active inference.
V. The Bayesian Brain
Thomas Bayes was an eighteenth century English minister and mathematician who developed a theorem about how to update beliefs in the light of new evidence. Bayes' theorem — which is simple in its mathematical statement and profound in its implications — describes the optimal way to revise a prior belief given new data.
The Bayesian brain hypothesis proposes that the brain operates as a Bayesian inference machine — that it maintains probabilistic beliefs about the causes of its sensory inputs, and updates those beliefs in a way that is mathematically consistent with Bayes' theorem.
What this means practically is that the brain does not treat all predictions equally. It weights its predictions by its confidence in them. A prediction based on extensive experience — a well-established prior — is treated as more reliable than a prediction based on sparse or ambiguous evidence. When new sensory information arrives, how much the brain updates its model depends on the relative reliability of the prior versus the new evidence.
This is why expertise changes not just what you predict but how confidently you predict it. The novice surgeon and the expert surgeon both make predictions during a procedure. But the expert's predictions are weighted with decades of accumulated evidence. When something unexpected occurs, the expert's brain has a richer prior against which to evaluate the deviation — a more sophisticated model for distinguishing signal from noise, genuine anomaly from expected variation.
And it is why learning under uncertainty is so much harder than learning under clear feedback. If the world gives you ambiguous signals — if the outcome of your action could have been caused by multiple factors — the brain struggles to identify which part of its model needs updating. Unambiguous error signals are the most efficient teachers. The cleaner the feedback, the faster the learning.
VI. All Brains, All the Time
The brain I have been describing — this predictive, Bayesian, active inference machine — is not the brain of a scientist or an engineer or an intellectual. It is every human brain. It is the brain of a three year old learning to pour water without spilling. It is the brain of an athlete developing a feel for the ball. It is the brain of a musician who no longer has to think about where their fingers go. It is your brain, right now, as you read these words — generating predictions about what I am about to say, updating when I say something unexpected, building a model of the argument I am constructing.
We are all, always, doing science. We are all, always, testing hypotheses about the world and revising them in light of evidence.
The difference between informal, unconscious model-building and formal scientific inquiry is not a difference in kind. It is a difference in rigor, in explicitness, in the systematizing of a process the brain already knows how to do.
What formal education — what engineering education specifically — does at its best is take that natural process and make it conscious. Give it vocabulary. Give it methods. Give it tools for generating better predictions and more precise tests of those predictions. Not teach the brain something alien to its nature, but align the explicit practice of the discipline with the implicit practice of the brain.
VII. Engineering, Simulation, and the Prediction-Error Loop
Everything I have described about the brain is a description of what a finite element simulation does. Not by analogy. Not loosely. Structurally.
A simulation is a formalized internal model of a physical system. It encodes the best available knowledge about the geometry, the materials, the boundary conditions, the physics of the event. From that model it generates predictions: this is how the structure will respond. This is where the stress will concentrate. This is the acceleration the circuit board will experience. These are predictions, generated before the physical event occurs, from the best model the analyst can build.
Those predictions are then tested against reality — in the form of physical testing, field observation, correlation studies. Where the predictions match reality, confidence in the model grows. The prior is reinforced. Where the predictions deviate from reality — where the surprise signal arrives — the model is examined and revised. The material characterization is improved. The contact definition is refined. The unmodeled feature is added.
The gap between model and reality is not a failure of the simulation. It is the prediction error — the signal that drives the model's learning. It is the most valuable output the simulation can produce.
A simulation that perfectly predicts every test result would be extraordinary — but it would also have nothing left to teach. It is the gap that teaches. It is the surprise that advances understanding.
When a simulation engineer encounters a discrepancy between model and test, they have two strategies available. They can update the model — refine the material properties, improve the contact definition, add the unmodeled feature. Or they can modify the design — change the product to behave more like what the model predicts it should. Both strategies close the gap. Both minimize the surprise. And in engineering practice, both strategies are used, often simultaneously, in an iterative process that converges toward a product that is both well-understood and well-designed.
This is active inference applied to product development. The engineer is not a passive observer of the simulation's output. The engineer is an active agent in a prediction-error-minimization loop that spans model, test, design revision, and model again.
VIII. Expertise as a Trained Prediction Machine
The experienced engineer does not simply know more facts than the junior engineer. That is a superficial description of the difference. The experienced engineer has built richer, more accurate, more extensively calibrated internal models of how physical systems behave. They have accumulated a vast prior — weighted by decades of predictions confirmed and predictions corrected — that allows them to generate better hypotheses, ask better questions, and interpret evidence with greater precision.
When an experienced structural engineer looks at a stress contour plot and says that result does not look right — before they have run any calculation, before they have checked any number — they are doing something neurologically profound. They are generating a prediction from a rich internal model, comparing it to the simulation output, and detecting a discrepancy. The surprise signal fires. Something does not match the model. That intuition — which looks from the outside like a mysterious gift — is in fact the output of decades of Bayesian updating.
Expertise is a prediction machine that has been extensively trained.
If expertise is a well-calibrated prediction machine, then the goal of engineering education is not the transfer of information. It is the construction of better internal models. And the construction of better internal models requires exposure to prediction error. It requires the learner to generate predictions — even incorrect ones — and to receive clear, unambiguous feedback about where those predictions fall short.
Why Passive Learning Fails
Passive learning — sitting in a lecture, reading a textbook, watching someone else perform the analysis — is so much less effective than active engagement. Passive reception does not generate predictions. Without predictions, there is no prediction error. Without prediction error, there is no update signal. Without an update signal, the model does not change. The information may enter working memory temporarily, but it does not become part of the durable internal model that constitutes genuine understanding.
Active engagement — being asked to predict what will happen before you see the answer, being asked to explain why the result looks the way it does, being asked to propose what change to the design would address the finding — forces the brain to generate predictions. Those predictions, when tested against the reality of the analysis, produce the error signals that actually update the model. That is learning that lasts.
IX. Embodied Learning: Why the Silly Putty Matters
When I hand a student a piece of Silly Putty and ask them to pull it slowly and then snap it fast, I am giving their brain a prediction problem. The first pull probably confirms their prediction — yes, it is soft and stretchy. The snap violates the prediction. The material they just felt stretching now fractures like a solid. Surprise. Error signal. The brain asks: what model do I need to update to account for this?
And the answer — rate dependent material behavior, the idea that the same material can respond completely differently depending on how fast you deform it — is now attached to a sensory experience, not just a verbal description. The model that updates from that experience is richer and more durable than the model that updates from reading the words rate dependent behavior on a page. Because the Bayesian prior is built from evidence, and physical experience is among the richest forms of evidence the brain can receive.
The composition books are not nostalgia. Writing by hand requires the brain to process, synthesize, and encode information rather than simply receive it. The motor act of writing recruits the prediction machinery in a way that typing, and certainly passive listening, does not. The student who writes what they are learning is generating a prediction — this is what I understood — and committing it to a form that can be tested. The act of writing is an act of inference.
X. The Brain Is Always in the Process of Becoming
The brain is not a static machine. It is not a fixed capacity that you either have or do not have. It is a living system that is continuously remodeling itself in response to the prediction errors it encounters. The technical term is synaptic plasticity — the strengthening and weakening of connections between neurons based on experience. But the functional description is simpler and more powerful: the brain is always in the process of becoming.
Every prediction you make and test. Every simulation you run and compare to physical evidence. Every design review where you look at results you do not fully understand and push yourself to ask the question that reveals the mechanism. Every time you step outside the comfortable boundary of your current model and into the territory where your predictions break down — every one of those experiences is an opportunity for the brain to become something it was not before.
The engineer you are right now is the product of every prediction you have made and tested up to this moment. The engineer you will become is being built, right now, by the quality of the predictions you generate and the rigor with which you test them.
And this means that the gap — the uncomfortable space where the model does not fit the reality — is not just professionally important. It is personally important. The gaps in your understanding are where your growth lives. Seeking them out rather than avoiding them. Being curious about the surprise rather than defensive about the prediction that failed. That orientation — toward the gap, toward the error signal, toward the update — is the orientation of the genuine learner. And it is available to anyone, at any stage of their career, at any level of experience.
XI. Engineering as an Act of Prediction
Engineering is an act of prediction. Every design embodies a model — a prediction about how the world will interact with the thing being built. The material will carry this load. The structure will survive this event. The product will function within this environment for this duration. Every specification, every tolerance, every safety factor is a statement about the engineer's model of the world and their uncertainty about that model.
When the design succeeds — when the bridge stands, when the product survives the drop, when the system performs as specified — it is not just a technical achievement. It is a confirmation of a prediction. The model was right, or right enough.
When the design fails — when the product cracks in a location the simulation did not predict, when the glass breaks at a load that should have been safe, when the assumption that was always made turns out to have been wrong — it is not just a setback. It is the most valuable data point the engineer's model will ever receive. It is the surprise that carries the most information. It is the error signal that, if received with curiosity rather than defensiveness, will produce the largest update to the internal model.
The engineer who is afraid to be wrong is the engineer whose model stops growing. The engineer who is curious about being wrong is the engineer who compounds their understanding with every project, every test, every gap between prediction and reality.
Over a career, the difference in the quality of those internal models is the difference between competence and mastery.
XII. Closing: Becoming a Conscious Prediction Machine
We are all prediction machines. The brain sees to that. What we are not, by default, is conscious prediction machines. The default setting is automatic — predictions generated without awareness, errors corrected without reflection, models updated without intention.
What engineering education can offer, at its deepest level, is the shift from unconscious to conscious modeling. The ability to say: here is my prediction. Here is my model. Here is where I expect the prediction to break down, and here is how I will test it. Here is the gap I found. Here is what I believe the gap is telling me. Here is how I am updating the model.
That practice — explicit, intentional, reflective — is what separates the engineer who grows continuously from the engineer who accumulates years without accumulating depth. It is available regardless of how much experience you already have. It is available at the first simulation review of your career and at the thousandth. It is the practice, not the credential, that builds the model.
You are a prediction machine. You always have been. The world has been training your model since the moment you were born — through every reach and fall, every experiment and failure, every surprise and recovery. You did not choose to be a modeler of the world. Your brain made that choice for you, before you had words for it.
What you can choose is how consciously you engage with that process. How deliberately you seek out the gaps. How honestly you receive the error signals. How rigorously you update rather than defend.
The brain updating its models through experience. The simulation updating its models through testing. The engineer growing through engagement. It is all the same process. And it never really ends.
References
The following works underpin the key claims, frameworks, and thinkers discussed in this essay. Readers interested in exploring predictive processing, active inference, Bayesian cognition, or the neuroscience of learning and expertise are encouraged to engage with these primary and secondary sources.
Predictive Processing & Active Inference
1. Friston, K. (2010). The free-energy principle: a unified brain theory?. Nature Reviews Neuroscience, 11(2), 127–138. https://doi.org/10.1038/nrn2787
2. Friston, K. (2009). The free-energy principle: a rough guide to the brain?. Trends in Cognitive Sciences, 13(7), 293–301. https://doi.org/10.1016/j.tics.2009.04.005
3. Clark, A. (2016). Surfing Uncertainty: Prediction, Action, and the Embodied Mind. Oxford University Press.
4. Friston, K., FitzGerald, T., Rigoli, F., Schwartenbeck, P., & Pezzulo, G. (2017). Active inference: a process theory. Neural Computation, 29(1), 1–49. https://doi.org/10.1162/NECO_a_00912
5. Rao, R. P. N., & Ballard, D. H. (1999). Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects. Nature Neuroscience, 2(1), 79–87. https://doi.org/10.1038/4580
6. Parr, T., Pezzulo, G., & Friston, K. J. (2022). Active Inference: The Free Energy Principle in Mind, Brain, and Behavior. MIT Press.
Bayesian Brain & Probabilistic Cognition
7. Knill, D. C., & Pouget, A. (2004). The Bayesian brain: the role of uncertainty in neural coding and computation. Trends in Neurosciences, 27(12), 712–719. https://doi.org/10.1016/j.tins.2004.10.007
8. Weiss, Y., Simoncelli, E. P., & Adelson, E. H. (2002). Motion illusions as optimal percepts. Nature Neuroscience, 5(6), 598–604. https://doi.org/10.1038/nn0602-858
9. Körding, K. P., & Wolpert, D. M. (2004). Bayesian integration in sensorimotor learning. Nature, 427, 244–247. https://doi.org/10.1038/nature02169
10. Dayan, P., Hinton, G. E., Neal, R. M., & Zemel, R. S. (1995). The Helmholtz machine. Neural Computation, 7(5), 889–904. https://doi.org/10.1162/neco.1995.7.5.889
Sensory Filtering & Conscious Bandwidth
11. Nørretranders, T. (1998). The User Illusion: Cutting Consciousness Down to Size. Viking (translated by Jonathan Sydenham).
12. Miller, G. A. (1956). The magical number seven, plus or minus two: some limits on our capacity for processing information. Psychological Review, 63(2), 81–97. https://doi.org/10.1037/h0043158
Expertise & Pattern Recognition
13. Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4(1), 55–81. https://doi.org/10.1016/0010-0285(73)90004-2
14. Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). The role of deliberate practice in the acquisition of expert performance. Psychological Review, 100(3), 363–406. https://doi.org/10.1037/0033-295X.100.3.363
15. Klein, G. (1998). Sources of Power: How People Make Decisions. MIT Press.
Learning, Memory & Synaptic Plasticity
16. Hebb, D. O. (1949). The Organization of Behavior: A Neuropsychological Theory. Wiley.
17. Bliss, T. V. P., & Lømo, T. (1973). Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path. Journal of Physiology, 232(2), 331–356. https://doi.org/10.1113/jphysiol.1973.sp010273
18. Kandel, E. R. (2006). In Search of Memory: The Emergence of a New Science of Mind. W. W. Norton & Company.
Active Learning & Engineering Education
19. Freeman, S., Eddy, S. L., McDonough, M., Smith, M. K., Okoroafor, N., Jordt, H., & Wenderoth, M. P. (2014). Active learning increases student performance in science, engineering, and mathematics. Proceedings of the National Academy of Sciences, 111(23), 8410–8415. https://doi.org/10.1073/pnas.1319030111
20. Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255. https://doi.org/10.1111/j.1467-9280.2006.01693.x
21. Mueller, P. A., & Oppenheimer, D. M. (2014). The pen is mightier than the keyboard: advantages of longhand over laptop note taking. Psychological Science, 25(6), 1159–1168. https://doi.org/10.1177/0956797614524581
Philosophy of Mind & Enactivism
22. Helmholtz, H. von (1867). Handbuch der physiologischen Optik (Treatise on Physiological Optics). Voss. English translation by J. P. C. Southall, 1924, Optical Society of America
23. Clark, A., & Chalmers, D. J. (1998). The extended mind. Analysis, 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7
24. Varela, F. J., Thompson, E., & Rosch, E. (1991). The Embodied Mind: Cognitive Science and Human Experience. MIT Press.
About the Author
Joe McFadden is a simulation engineer, educator, and writer known as The Holistic Analyst. His work bridges the science of how the brain learns with the practice of engineering education. He writes and teaches with the mission of combating engineering mind blindness — one student at a time.
