Where Minds Move

A Rough Map of Prediction Space

I was reading Andy Clark’s Surfing Uncertainty when one of those dangerous little doors opened in my head. Clark was discussing predictive processing, especially the way differences in precision weighting might help us think about conditions like schizophrenia and autism. I do not mean “explain them away,” and I definitely do not mean “reduce them to one broken knob in the brain.” I mean something more modest and more interesting: predictive processing gives us a language for thinking about how minds lean into the world, how the world leans back, and how much confidence gets assigned to different parts of that exchange.

That thought stayed with me because precision is such a strange and powerful idea. A brain is not merely receiving signals. It is deciding, moment by moment, which signals deserve trust. The sound in the hallway, the look on someone’s face, the feeling in the stomach, the memory of how this kind of situation went last time, the expectation that a task will be boring, the expectation that an adult is about to be unfair, the expectation that the next step in a math problem will make sense. All of these arrive with some degree of confidence attached to them, though we rarely experience that confidence directly. We usually just experience reality as reality.

That is the part that hooked me. If the brain is constantly balancing prediction and error, and if different signals can be given different levels of confidence, then cognitive states might be understood as movement through a kind of space. The technical version of this idea would probably be called a state space. I am not going to pretend I can build that model. But as a metaphor, imagine a vast prediction space where each point represents a possible operating state of a mind.

At one point in the space, sensory signals are loud and hard to ignore. At another, prior expectations are heavy and hard to update. Somewhere else, the body is noisy, the future is dim, social cues feel dangerous, or action feels impossible. A diagnosis may describe regions someone often visits, but it does not tell us exactly where they are right now.

In prediction space, what moves is not the person as a whole, and not the diagnosis as a label. What moves is the person’s current operating state: the shifting configuration of prediction, error, confidence, attention, body-state, social expectation, and action-readiness. That is a clunky sentence, but I think the clunkiness is useful. It keeps me from making the idea too magical. A “location” in prediction space is not a tiny glowing dot inside the brain. It is a temporary pattern in how a living system is meeting the world.

The phrase “prediction space” is intentionally loose. I am not proposing a finished theory, and I am certainly not proposing a new diagnostic system from the comfort of a blog post. I am trying to describe a way of seeing. The idea is that minds do not simply belong to categories like ADHD, autism, anxiety, depression, schizophrenia, trauma, and so on. Those categories may still matter. They may be useful, compassionate, clarifying, and necessary in many contexts. But they are not the whole terrain. They are named regions on a map that is still being drawn.

This is where ADHD first pulled the idea into focus for me. The more I read and listen, the less ADHD looks like one thing. That is not a complaint. It may be a sign that the label is pointing toward a family of related patterns rather than a single clean mechanism. Some people with ADHD seem primarily pulled by novelty. Some seem trapped by poor task initiation. Some are crushed by emotional volatility. Some can focus with terrifying intensity under the right conditions and then cannot make themselves answer a perfectly ordinary email. Some seem driven by the promise of movement, urgency, conflict, interest, risk, or social accountability. Some seem less like they lack attention and more like they cannot regulate the auction in which attention is sold.

Recent research appears to be moving in a similar direction, though far more carefully than I am in this post. Neuroimaging work has continued to emphasize heterogeneity in ADHD, including studies that identify possible neurobiological subtypes or biotypes. That does not mean there are now three simple kinds of ADHD, as if the brain finally handed us a color-coded filing cabinet. It means that symptom-based labels may be too coarse to capture the underlying variation. The same outward behavior might be reached by different internal routes, and similar internal differences might produce different outward behaviors depending on the person’s environment.

Predictive processing feels useful here because it gives us a way to think about those routes. Imagine one person whose system assigns too much importance to distracting sensory input. The flickering light, the chair noise, the tag in the shirt, the whisper behind them, the possible judgment in someone’s tone. Imagine another person whose system calibration is less sensory intensity and more the weak gravitational pull of distant rewards. The assignment due next Friday does not generate enough useful predictive force today. Imagine another person whose internal body signals are noisy or hard to interpret, so boredom, hunger, fatigue, anxiety, and irritation blur together into one unpleasant fog. All three people might be described with the same label in a school or clinical setting. All three might need support. But the supports may need to work on different parts of the space.

The metaphor that helps me most is a changing terrain. A person’s current cognitive state is a location in prediction space, but that location is never just a dot on a map. It is a location under conditions. Those conditions emerge from forces operating at different time scales.

Some forces operate slowly. Genes, development, temperament, early experience, trauma, security, and long practice help form the deep terrain. They carve valleys, raise hills, create fault lines, and make some paths easier than others. This deeper terrain does not determine everything, but it shapes what kinds of movement are easy, difficult, likely, or costly.

Other forces operate at a regional scale. Family routines, classroom norms, medication, exercise, sleep patterns, peer dynamics, repeated success, repeated embarrassment, institutional expectations, and stable relationships act more like currents or prevailing winds. They do not determine exactly where a person goes, but they bias movement. They make certain regions easier to drift toward and others harder to reach.

Then there are local conditions: hunger, fatigue, noise, shame, novelty, pain, uncertainty, time pressure, social threat, or one adult’s tone of voice at the wrong moment. These can change what is possible right now. The same hill is not the same hill when you are rested, fed, trusted, and guided. The same hallway is not the same hallway when you are exhausted, overstimulated, ashamed, and expecting punishment.

This is the version of the metaphor that feels least sloppy to me. Deep terrain, regional currents, local conditions. I do not want to overbuild it. The reader should not need a legend, a compass, a barometer, and a minor in geology to get through a blog post. But the three scales help me avoid a mistake I kept making in earlier versions of the idea. I kept wanting the landscape metaphor to handle everything, and then I wanted weather to handle everything the landscape could not. That quickly became a metaphor pileup. A small conceptual traffic accident. Possibly with injuries.

The three scales make the image more disciplined. The current state is the location. Long-term biology and development help shape the deep terrain. Routines, institutions, relationships, and repeated patterns create regional pressures. Immediate bodily and social conditions affect what the next step feels like. Behavior is movement within all of that.

A student may enter the room with deep terrain shaped by ADHD, anxiety, trauma, temperament, or years of academic failure. The regional currents may include a school culture where mistakes are embarrassing, math feels like a status threat, and adults are expected to become irritated. Then the local conditions arrive: little sleep, no breakfast, fluorescent lights, a confusing warm-up problem, and a classmate laughing nearby. By the time the student refuses to work, the refusal is not floating in empty space. It is movement in a shaped terrain under strong conditions.

That does not excuse the behavior. I do not want to make this sound soft. People still need boundaries, expectations, practice, and consequences. In fact, predictive processing may help explain why those things matter. A boundary is a prediction-shaping structure. A routine is a prediction-shaping structure. A clear assignment, a calm voice, a visible timer, a worked example, a stable classroom norm, a recovery break, a chance to move, and immediate feedback are all ways of shaping the space in which a mind is trying to act.

The point is not that context magically explains away responsibility. The point is that responsibility becomes more useful when it is attached to mechanisms. If a person is stuck, we can ask what keeps returning them to the same region. Is the path too well worn? Are the local conditions overwhelming? Are the regional currents pushing them there every day? Is the deep terrain making some transitions unusually expensive? Those questions are more useful than simply deciding the person has a bad attitude and calling it a day.

This matters in classrooms. Teaching is often described as delivering instruction, but that phrase is too thin. A teacher is also designing a prediction environment. The room teaches students what to expect from effort, from mistakes, from adults, from peers, from confusion, from boredom, and from themselves. A chaotic room trains one set of predictions. A rigid and humiliating room trains another. A room with clear routines, low social threat, meaningful challenge, and fast feedback trains something else. The curriculum matters, obviously, but the nervous system has to survive the room before it can use the lesson.

I keep coming back to precision weighting because it offers a useful way to think about attention. Attention is often talked about like a flashlight, as if the main question is whether the beam is pointed at the right thing. That metaphor works sometimes, but it hides too much. In predictive processing, attention can be thought of partly as changing the gain on certain prediction errors. It is closer to adjusting the volume on the signals the brain treats as important. That means attention problems may not always be failures to point the beam. They may be failures to tune the system.

This makes ordinary situations look different. A student tapping a pencil may be regulating arousal. A student staring out the window may be escaping prediction-error overload. A student arguing about a minor instruction may be responding to social threat rather than the instruction itself. A student who cannot start may not lack the desire to succeed; the task may not yet have enough shape for action. A student who works beautifully during a crisis may not be mysterious at all. The crisis supplies urgency, clarity, stakes, and a short enough time horizon to make the next move obvious.

Adults are not exempt from this. Anyone who has opened a laptop to complete one important task and then emerged forty minutes later from a swamp of tabs has experienced a small failure of predictive governance. The world offered little handles of possible relevance, and the brain kept taking them. Each click promised a reduction in uncertainty. Each one created more. This is why “just focus” is such comically bad advice. It is like telling someone in a canoe to “just river better.”

Prediction space also gives me a way to think about states that we usually discuss separately. Anxiety might involve prediction-error systems that are too ready to treat uncertainty as threat. Depression might involve priors about futility, effort, and future reward becoming too heavy. Trauma might involve a system whose threat predictions were calibrated by real danger and now fire in safer environments that resemble the old one in partial ways. Flow might be a temporary region where prediction, error, skill, and feedback are beautifully tuned. Creativity might involve looser priors, wider search, or a willingness to let unusual errors matter for longer than usual. These are sketches, not conclusions. They are invitations to be more precise later.

The schizophrenia and autism examples from Clark are especially powerful because they show both the promise and the danger of this kind of thinking. It is tempting to make tidy claims. Autism is this kind of precision difference. Schizophrenia is that kind. ADHD is another. The temptation should be resisted. Minds are not simple machines with three labeled dials. A change in one part of a predictive system can ripple through perception, action, social inference, bodily feeling, memory, and belief. Similar symptoms may arise from different mechanisms, and similar mechanisms may look different in different environments.

So the point is not to replace old boxes with shinier boxes. The point is to become less box-drunk.

A space lets us think about relations, movement, and transformation. It provides a geometry that lets us ask how one state becomes another. How does curiosity become overwhelm? How does uncertainty become panic? How does boredom become misbehavior? How does structure become safety for one person and suffocation for another? How does social ambiguity become creativity in one context and paranoia in another? How does a student move from “I cannot do this” to “I can try the next step”? Those questions are not answered simply by naming a diagnosis, though a diagnosis may help us ask better questions.

There is a research version of this idea that I am not equipped to build but can vaguely imagine. It would involve trying to identify dimensions of predictive processing that can be measured or approximated across individuals and contexts. Some dimensions might come from cognitive tasks, some from sensory processing measures, some from interoceptive measures, some from behavioral data, some from reports of lived experience, and some from environmental features. Then we might look for clusters, trajectories, attractors, and transition points. Not just who belongs to which group, but what moves people between states.

The word “attractor” is useful here, though I want to handle it carefully. In complex systems, an attractor is a pattern or region a system tends to settle into. That does not mean the system is doomed to stay there. It means the system has tendencies. A person might have an anxiety attractor, an avoidance attractor, a hyperfocus attractor, a shutdown attractor, or a conflict attractor. Those attractors may be deepened by biology, history, and environment. Attractors may also be softened by sleep, trust, medication, practice, relationship, movement, meaning, and better-designed surroundings.

I like the attractor idea because it respects both pattern and change. It lets us say that a person may repeatedly end up in a certain kind of state without claiming that state is their essence. A student who shuts down during multi-step math problems may not be choosing shutdown in any simple sense. They may be sliding into a well-worn basin where confusion predicts humiliation, effort predicts failure, and escape predicts relief. If that is true, then the intervention is not merely “care more.” The intervention has to change the expected path.

If minds move through prediction space, then we are always participating in one another’s terrain. Parents, teachers, managers, friends, spouses, algorithms, architecture, schedules, grading systems, economic pressure, and cultural norms all shape prediction. Some ecologies make good predictions easier. Others make bad predictions rational. If a person repeatedly learns that effort leads to humiliation, confusion leads to punishment, honesty leads to danger, or stillness leads to unbearable internal noise, then the resulting behavior is not random. It may be a reasonable adaptation to an unreasonable space.

This is the part of the idea that feels most useful to me as a teacher and parent. The goal is not to remove difficulty. Difficulty is where learning happens. The goal is to tune difficulty so that the next move remains available. Too little error and nothing changes. Too much error and the system protects itself. A good learning environment probably lives in the middle, where prediction fails just enough to update but not so much that the person abandons the game.

I realize this could all become a fancy way to say “context matters.” Maybe that is one of the risks. But I think predictive processing adds something sharper than that. It does not merely say the environment matters. It suggests some possible mechanisms by which the environment matters. The world changes the signals available to the organism. The organism weights those signals according to its history, body, goals, and expectations. Action then changes the world, which changes the next round of prediction. Mind and world are not two separate actors passing notes across a desk. They are coupled systems.

This is why the image of movement through prediction space feels alive to me. A mind is not sitting inside the skull calmly viewing a model of reality on a screen. It is actively trying to stay viable in a changing world. It predicts, tests, moves, corrects, ignores, amplifies, suppresses, and sometimes gets stuck. It uses the body. It uses tools. It uses other people. It uses routines, language, stories, classrooms, calendars, medications, songs, caffeine, exercise, and little rituals that would look ridiculous if they did not work.

There is humility built into this view, or at least there should be. If predictive processing is right in even a broad sense, then each person we meet is the visible surface of an enormous hidden negotiation. We see the behavior. We do not see all the predictions. We do not see which errors are screaming and which ones are silent. We do not see the body’s background budget, the old memories shaping today’s threat detection, or the social model the person is carrying into the room. We rarely see the full space. We see someone appear at one point in the journey and start making judgments about the whole map.

I am not sure where this idea goes yet. That is part of the attraction. It feels like a rough map worth sketching, not because it is already right, but because it may help me notice better questions. What dimensions matter most? Which ones are measurable? Which ones are useful only as metaphors? How do different diagnostic categories overlap in this space? How do sleep, stress, medication, exercise, trauma, and social safety move someone through it? What would it mean to design a classroom, a home, or a day with prediction space in mind?

I keep thinking that the best use of this idea may not be classification. It may be compassion with teeth. Compassion, because it reminds us that behavior emerges from systems rather than from isolated acts of will. Teeth, because systems can be studied, shaped, constrained, practiced, and improved. We do not have to choose between blame and helplessness. There is a middle path where we take mechanisms seriously enough to change conditions and take people seriously enough to expect growth.

That is the place I want to keep exploring. Somewhere between neuroscience and teaching, between philosophy and parenting, between the elegance of a theory and the mess of a Tuesday morning classroom. I am tiptoeing, yes. But sometimes the minefield is where the interesting maps are buried.

Source notes and starting points

Recent ADHD research continues to emphasize heterogeneity rather than one clean underlying profile. Examples include work on ADHD neurobiological subtypes and biotypes using neuroimaging and normative modeling:

Nan Pan et al., “Mapping ADHD Heterogeneity and Biotypes by Topological Properties of Morphometric Similarity Networks,” JAMA Psychiatry, 2026.

Xiangyu Bu et al., “Toward individual heterogeneity and neurobiological subtypes in attention-deficit/hyperactivity disorder: A systematic review of neuroimaging studies,” Australian & New Zealand Journal of Psychiatry, 2025.

Yan Chen et al., “Distinct neuroimaging subtypes of ADHD among adolescents,” Translational Psychiatry, 2025.

For broader predictive-processing context, see work by Andy Clark, Karl Friston, Anil Seth, Jakob Hohwy, and Lisa Feldman Barrett. The essay above uses these ideas as inspiration, not as a settled technical model.

I have also created this quasi bibliography for the works that are shaping my ideas.