Your Brain Is Guessing: 10 Ways to Build Metacognition in a Predictive Mind

For much of my life I believed the brain reacts to the world. The eyes were windows. The ears were microphones. Reality streamed in, the brain took a look, and then it responded.

That is not what happens.

The brain sits in a dark, silent box. It has never seen light or heard a sound. All it has ever known is a stream of electrical and chemical signals arriving from the sensory surfaces, plus a second stream from inside the body: heart rate, gut, breath, temperature, hormones. It takes those signals, cross-references them against everything it has learned (its priors, beliefs, expectations and attitudes) and generates a best guess about what is out there and what to do next.

Then it acts on the guess, not on the world.

When I first understood this, it changed how I relate to my own thoughts. A thought is not a report. It is a prediction. And predictions can be observed, questioned and revised. That observing capacity has a name: metacognition. This article explains the predictive brain in plain language, then offers ten evidence-based tools for watching your own predictions at work.

What the predictive brain actually does

The dominant model in modern neuroscience is often called predictive processing. Karl Friston's free energy principle (2010) formalised it, Andy Clark's Surfing Uncertainty (2016) made it accessible, and Anil Seth's Being You (2021) gave it the memorable phrase "controlled hallucination". Lisa Feldman Barrett's How Emotions Are Made (2017) extended it to feeling (I've written a plain-language guide to Barrett's work).

The core idea is simple. The brain does not wait for input and then build a picture. It builds the picture first, top down, and uses incoming signals only to check its work. When the signal matches the prediction, nothing much happens. When it doesn't, the mismatch is called prediction error, and error is the only thing the brain really pays attention to.

There are two ways to reduce prediction error:

1. Update the model. Change what you expect so it fits the world. We call this learning, and I've argued elsewhere that the capacity to update is the skill underneath most others.
2. Change the world. Act so that reality comes to match the prediction. We call this behaviour.

A simple example. My brain predicts that a warm bedroom means comfort. If the room is cold, I can change the world: close the window, turn on the heater. Or I can update the model: learn that the window stays shut after May, or that I actually sleep better at sixteen degrees than at twenty-two. Both moves are legitimate. Wisdom is knowing which one the moment calls for.

There is a third possibility, and it is where a lot of suffering lives. The brain generates a prediction, doesn't trust it enough to act, and doesn't have enough error to update it. So it waits. It runs the simulation again. And again. That loop of unresolved prediction is a decent working description of anxiety. Grupe and Nitschke (2013) argued that anxiety is fundamentally a problem of uncertainty and anticipation rather than of threat itself, and Carleton (2016) proposed that intolerance of uncertainty may be the common thread running through most anxiety disorders.

Barrett adds a crucial layer. Many of the brain's most important predictions are about the body, not the world. The brain runs a kind of budget, forecasting how much energy the day will need and adjusting heart rate, blood sugar and cortisol in advance. When the budget is depleted through poor sleep, hunger or illness, the brain's predictions tilt negative, and it interprets neutral events as threatening. The feeling of "something is wrong" is often the brain reporting a budget shortfall, then searching the world for a reason.

Why metacognition is the lever

If predictions run the show, the obvious question is whether we can see them. The answer is partly yes. Metacognition, the ability to monitor and evaluate your own mental processes, is one of the better-studied capacities in cognitive neuroscience. Stephen Fleming's Know Thyself (2021) summarises two decades of work showing that metacognitive accuracy varies between people, is partly separable from raw ability, and can be improved with practice and feedback.

The tools below share one goal: to get data about your own predictive machinery, either by catching predictions before they become actions, or by collecting evidence about the model that produced them.

1. Write the prediction down before the event

The simplest metacognitive tool is to make an implicit prediction explicit. Before a meeting, a difficult conversation, a race, a launch, write one line: "I expect this to go like X." Afterwards, write what actually happened.

Philip Tetlock's forecasting research (Tetlock and Gardner, 2015) found that the people who became measurably better at predicting the future shared one habit: they wrote their forecasts down and scored them against outcomes. Without a written record, the brain rewrites history to protect the model. With one, you get to see your priors on paper, which is often the first time you meet them.

Start small. Three predictions a day, each with a follow-up line. Within a month you will notice patterns: where you are consistently pessimistic, where you overestimate your own influence, where your model of a particular person is out of date.

2. Keep a surprise log

Surprise is prediction error made conscious. Every time you feel it, your brain has just told you that its model was wrong.

Most of us discard surprises within seconds. A surprise log catches them. When something goes differently from how you expected, note it: what happened, what you expected, and what belief the expectation was resting on.

This matters because the brain updates unevenly. Errors that are emotionally costly get processed, while errors that are merely inconvenient often get explained away. A log forces the question "what did I believe that made this surprising?" and that question is the front door to model revision. It is learning, done deliberately rather than by accident.

3. Rate your confidence, then check it

Confidence is a prediction about your predictions. Fleming and colleagues have shown that it can be measured (how well your confidence tracks your accuracy is called metacognitive sensitivity) and, importantly, trained. Carpenter and colleagues (2019) found that giving people feedback on their confidence judgements improved their metacognitive accuracy, and the improvement transferred to a different task.

The practice: when you make a claim or a decision, attach a number. "I'm 70 percent sure this proposal will land." Then track outcomes. If your 70 percent claims come true about 70 percent of the time, your metacognition is calibrated. If they come true 40 percent of the time, you have learned something valuable about how much to trust the feeling of certainty.

I find this particularly useful for beliefs about other people, where confidence tends to run far ahead of evidence.

4. Name the feeling precisely

In Barrett's account, emotions are not reactions but constructions: the brain's best guess about what an internal state means, given the context. That guess is heavily shaped by the concepts you have available.

People with high emotional granularity, who distinguish between disappointed, resentful and deflated rather than lumping everything under "bad", show better outcomes across a range of studies. Kashdan, Barrett and McKnight (2015) reviewed evidence linking granularity to lower anxiety, less binge drinking and less aggression under stress. Lieberman and colleagues (2007) used brain imaging to show that simply labelling an emotion reduced activity in the amygdala and increased activity in prefrontal regions associated with regulation.

The tool is almost embarrassingly simple. When you notice a feeling, name it with as much precision as you can (the interactive emotion wheel exists for exactly this). Then ask what prediction it is carrying. Anxiety before a presentation is a prediction of judgement. Irritation with a colleague is often a prediction of being ignored. Once the prediction is visible, it becomes negotiable.

5. Check the body budget before you believe the thought

Because so many predictions originate in the body, one of the most powerful metacognitive questions is also the most mundane: what state is my body in right now?

Research on interoception (Critchley and Garfinkel, 2017) suggests that people vary widely in how accurately they sense internal signals, and that this accuracy shapes emotional experience and decision-making. A brain running on four hours of sleep generates different predictions from a rested one. Matthew Walker's Why We Sleep (2017) collects evidence that sleep deprivation amplifies amygdala reactivity and weakens prefrontal control, which in predictive terms means more threat predictions and less capacity to question them.

Before acting on a strong negative interpretation, run a quick audit: sleep, food, hydration, movement, illness, alcohol from last night. If two or more are off, treat the thought as a budget report rather than a fact about the world. Fix the budget first. Then see if the thought is still there.

This is also where a slow breath earns its reputation. Extending the exhale sends a signal of safety inward, and the brain revises its predictions accordingly. It does not solve the problem, but it changes which brain is doing the thinking.

6. Run small experiments with uncertainty

If anxiety is an unresolved prediction loop, the way out is not more thinking. It is evidence. The brain updates on error, so the fastest way to revise a fearful model is to give it a chance to be wrong.

Cognitive behavioural therapy calls these behavioural experiments, and the evidence base is strong (Bennett-Levy and colleagues, 2004). The structure: state the prediction ("if I say no to this request, they will be angry"), rate your belief in it, run the experiment, record the result, re-rate.

You do not need a therapist to use the format. Pick one uncertainty you have been avoiding this week, write the catastrophic prediction, and design the smallest possible test. Most of the time the world turns out to be less dangerous than the model. Occasionally it is exactly as dangerous, in which case you have learned that too, and can act rather than loop.

7. Meditate to loosen the grip of your priors

Meditation has a reasonable claim to being the oldest metacognitive technology we have. Predictive processing offers a fresh explanation of why it works. Laukkonen and Slagter (2021) proposed that contemplative practice progressively reduces the weight the brain gives to its own predictions, allowing more raw sensory signal through. The Zen phrase "beginner's mind" describes exactly this: seeing with fewer priors in the way.

The experimental evidence is consistent with that account. Slagter and colleagues (2007) found that three months of intensive practice reduced the "attentional blink", the brain's tendency to miss a second target because it is still processing the first. In predictive terms, meditators were spending fewer resources on what they expected and more on what was arriving.

You don't need three months of retreat. Ten minutes of sitting with the breath, noticing when a thought appears and gently returning, is metacognition in its purest form. You are watching the prediction engine generate content, and declining to act on it.

8. Write freely to find the model you didn't know you had

Not all priors are available to introspection. Some only show up when you give them a way out.

James Pennebaker's expressive writing research, running since the 1980s, has repeatedly found that writing continuously about a difficult experience for fifteen to twenty minutes on several consecutive days improves physical and psychological health outcomes. One plausible mechanism in predictive terms is that unstructured writing lets the brain surface predictions it has been running in the background, translate them into language, and integrate them into a more coherent model.

The practice: set a timer, write without stopping, and do not edit. Pay attention to what appears that you did not plan to write. That material is often a belief the brain has been quietly acting on without ever putting it up for review.

9. Borrow another brain

Your model has blind spots that you cannot see by definition. Other people's models have different blind spots. Comparing notes is one of the few ways to get outside your own predictions.

Kruger and Dunning (1999) showed that the least skilled people in a domain are also the least accurate at judging their own skill, because the capacity to evaluate performance is largely the same capacity needed to perform. The way through is external feedback. Fleming's work supports the same conclusion from a different angle: metacognitive accuracy improves fastest when people receive honest information about how their confidence compared to reality.

Make it structured. Ask a trusted person: "What do you think I'm expecting to happen here?" or "Where do you see me being more certain than the evidence supports?" A good coach, a candid friend or a well-run team debrief all do this. So does reading widely outside your field, which is a slower version of the same move.

10. Choose the move: update the model or change the world

Every tool above generates data. This last one is about what to do with it.

When you catch a prediction and the world disagrees, you have two honest options. You can update the model: accept that the room is cold, the colleague is not going to change, the plan was optimistic. Or you can change the world: close the window, have the conversation, restructure the plan.

Anxiety tends to sit in the gap between these, refusing both. Metacognition lets you name the gap and pick. A useful question is: "Is this a learning moment or an action moment?" Sometimes the model is right and the world needs adjusting. Sometimes the world is right and the model needs to give way. The skill is not in always choosing one. It is in noticing that a choice exists.

Barrett's framing is helpful here too. Because the brain is always working from a budget, the most reliable way to change the world in your favour is often to change the body that is doing the predicting: sleep, move, eat, connect. That is not a soft recommendation. It is a direct intervention on the prior.

The point of all this

The predictive brain is not a flaw to be corrected. It is what allows you to walk into a room and know what to do without processing every photon. But a system that runs on guesses needs a way to check its guesses, and that check is not built in. It has to be built.

Each of these ten tools builds a small piece of it. Together they give you something the brain does not provide on its own: the ability to see a prediction as a prediction, and to decide, rather than just react.

I still catch myself believing the eyes are windows. The difference now is that I notice.

---

References: Barrett, L.F. (2017) How Emotions Are Made; Bennett-Levy, J. et al. (2004) Oxford Guide to Behavioural Experiments in Cognitive Therapy; Carleton, R.N. (2016) Journal of Anxiety Disorders; Carpenter, J. et al. (2019) Journal of Experimental Psychology: General; Clark, A. (2016) Surfing Uncertainty; Critchley, H. & Garfinkel, S. (2017) Current Opinion in Psychology; Fleming, S. (2021) Know Thyself; Friston, K. (2010) Nature Reviews Neuroscience; Grupe, D. & Nitschke, J. (2013) Nature Reviews Neuroscience; Kashdan, T., Barrett, L.F. & McKnight, P. (2015) Current Directions in Psychological Science; Kruger, J. & Dunning, D. (1999) Journal of Personality and Social Psychology; Laukkonen, R. & Slagter, H. (2021) Neuroscience & Biobehavioral Reviews; Lieberman, M. et al. (2007) Psychological Science; Pennebaker, J. & Chung, C. (2011) in the Oxford Handbook of Health Psychology; Seth, A. (2021) Being You; Slagter, H. et al. (2007) PLoS Biology; Tetlock, P. & Gardner, D. (2015) Superforecasting; Walker, M. (2017) Why We Sleep.