Answers Are Cheap, Clarity Is Rare
Why the half-second after an AI answer is where leaders now win or lose
A manager asks Copilot to draft a presentation on why sales have fallen. One slide reads: “Marketing leads may be contributing to the decline.” She has suspected marketing for months. In the half-second it takes to read that sentence, “may be contributing” becomes “is responsible.” Her next prompt is: “Make the case for cutting marketing spend.” Twenty minutes later she has polished slides, a confident recommendation, and no idea that the decision had already been made before she had checked anything.
Nothing in that scene is unusual. The AI was not wrong. The manager was not careless. She had plenty of time. And yet the most important moment in the whole exercise, the instant a prediction she was already running met a sentence that fit it and took over the question she asked next, passed without anyone noticing it.
A disclosure before I go further: I am one of the authors of Micro Moments, written with Carrie Tan [1], and I wrote Escape from System 1 [2] before it. This article argues for the ideas in those books, so read it as a case made by an author, not a neutral review.
Here is the case. In the AI era, answers are cheap. Any leader can get a fluent, well-structured answer in seconds. What has become rare is clarity. Clarity means knowing which question deserves to be asked, and noticing when the one you are about to ask was chosen for you by a prediction you never saw being made.
The leaders who thrive now are not the ones with the fastest answers. They are the ones who catch Micro Moments and ask the right next question.
A Micro Moment is the split second in which a prediction your brain has already made begins to feel like fact. Your brain does not record reality; it constructs it from memory, emotion and expectation, and most of the time that construction serves you well. But sometimes a prediction, whether carried for years or formed seconds earlier, turns into certainty before you notice, and quietly decides what you see, what you conclude and what you ask next. The manager had one when “may be contributing” became “is responsible.” That is what the book is about: not the preparation before a decision or the review afterwards, but the instant the decision is forming, and how to catch it while there is still time to change the question.
A Scientific Journey: From Simon to Micro Moments
Micro Moments did not appear from nowhere. It sits at the end of a seventy-year line of inquiry into why capable people make poor judgments, and as Figure 1 illustrates, each step in that line looked at a smaller unit of judgment than the one before.
Herbert Simon started it. In 1955 he argued that human rationality is bounded [3]: we lack the information, time and processing capacity to optimize, so we satisfice, searching until we find an option that is good enough and then stopping. The remedy he proposed was organizational. Build procedures that force deliberation before a decision is reached. His unit of analysis was the decision-maker inside an institution, and the work earned him the Nobel Prize in Economics in 1978 [4].
Daniel Kahneman, building on two decades of work with Amos Tversky [5], moved the unit inside the mind. Thinking, Fast and Slow (2011) [6] showed that Simon’s limits do not produce random error; they produce systematic error, because fast intuitive thinking (System 1) answers an easier question than the one asked and slow deliberate thinking (System 2) rarely notices the substitution. The remedy shifted with the unit: learn the catalogue of biases, and slow down when the stakes are high.
Then came Noise (2021) [7], written with Olivier Sibony and Cass Sunstein. It added a second kind of error. Bias is being systematically off target; noise is unwanted variability, and part of it, which the authors call occasion noise, is the same person judging the same case differently on different days. The remedy, decision hygiene, is preventive and statistical: audit the variability, structure judgments, aggregate independent views.
Noise is where I found the gap. The authors describe noise as a statistical concept rather than a causal one [8]. The book measures that a judge is harsher after lunch; it does not say what happens inside the judge at the moment the sentence forms. Decision hygiene, by the authors’ own analogy, is handwashing against a germ you cannot see [9].
My first attempt at that gap was Escape from System 1 (2022) [2]. It argued that the brain evolved for efficiency, not insight, and that breakthrough thinking requires deliberately leaving System 1. The argument was sound, but it left the hardest question open: escape when, exactly? Telling leaders to think slowly is advice about the hour. The failures I kept seeing lived in the half-second.
Answering that required a second scientific root, one that does not come from the Simon-Kahneman tradition at all. Over the past two decades, neuroscience has come to see the brain as a prediction machine: perception is not a recording of the world; it is the brain’s best guess, built from memory and expectation and corrected by whatever the senses contradict. Anil Seth calls the result a “controlled hallucination” [10]. Lisa Feldman Barrett argues that emotions are predictions too [11].
Micro Moments brings the two lines together. From the decision tradition it takes the problem: judgment fails in patterned ways, and existing remedies arrive before or after the decision. From the predictive brain it takes the mechanism: a guess can be experienced as fact before any checking begins. The micro moment is where they meet. It is the instant the decision is forming, the one point no existing tool reaches.
The Mechanism: Predictive Hallucination
Start with what the brain is doing before you decide anything. It is not waiting for the world to arrive. It is guessing what the world is about to be, and then checking the guess against whatever the senses deliver. Perception is the guess, lightly corrected [12]. This is why you can read a sentence with a missing word without noticing the gap, and why a coat on a chair can be a person for a quarter of a second in a dark room. The brain predicts first because prediction is cheap and waiting is expensive. Most of the time the guess is good, the correction is small, and the system works so smoothly you never see it operate.
The trouble begins when a guess is strong enough, or the correction weak enough, that the guess is experienced as fact. Carrie Tan and I call that event a predictive hallucination [1]. Given its modern tech baggage, however, the term requires an immediate qualification. It is not a clinical term; nobody in this account is unwell. And it must be kept entirely separate from AI hallucination, where a machine fabricates a false output or a fake citation. An AI hallucination is a machine error in generation. The human’s predictive hallucination is an internal, true-feeling perception that was never checked. The second is harder to catch, because it does not look like an error. It looks like seeing.
What makes a prediction strong enough to pass as fact? Usually, it is already loaded. The brain stores predictions the way a field stores mines: out of sight, armed, waiting for the right pressure. Some were laid down over a lifetime. Some were laid down last quarter, like the manager’s suspicion about marketing. Some were laid down thirty seconds ago by the previous paragraph you read. The age does not matter. What matters is that the prediction is in place, untested, and the incoming signal fits it well enough to set it off.
A predictive hallucination is what happens; the micro moment is when: the brief window in which it happens and can still be caught. Two features of that window are easy to miss.
First, it does not need urgency. The manager had all afternoon. A micro moment is fast because the brain is fast, not because the situation is. It happens in the calm review of a slide as readily as in a crisis.
Second, it does not need high stakes. The mine goes off in a trivial email and a billion-dollar decision alike. The stakes only decide how expensive the result is.
The damage is done to the question, not the answer. Once a prediction has become fact, it chooses the next question for you, and it chooses in three predictable ways. It narrows the options you consider, because alternatives to a fact look like noise. It hardens an untested assumption, because every question you now ask is built on it. And it hides emerging opportunities, because the brain stops looking for what the prediction says is not there. The manager did not ask a bad question. She asked a perfectly good question about the wrong fact, and that is the signature of a micro moment that passed unnoticed.
Why AI Makes This Moment More Frequent
Nothing about artificial intelligence changes how the human brain works. What changes is how often the brain is handed an external signal specifically shaped to set off a stored prediction.
To understand the risk, we must maintain a crisp boundary between the two systems meeting on the screen:
· The Machine Side: An AI model does not “know,” “intend,” or “decide.” It simply generates the most probable text continuation of a prompt, drawn from statistical patterns in its training data.
· The Human Side: The human brain is already running its own internal prediction about what that text will say.
When the machine’s output and the human’s internal guess align, the brain registers confirmation and relaxes. When they disagree, it registers threat and resists. Neither response has anything to do with whether the answer is factually correct; both happen before correctness is even considered.
AI output is unusually potent at triggering human bias because of how it is written. It arrives fluent, structured, confident and instant: the exact superficial traits the human brain uses as a shortcut for authority. A hesitant colleague gives the brain a reason to check. A clean machine-generated paragraph does not. The friction that used to force a pause has been engineered out, and the pause went with it.
The Easy Case (When the AI Is Right): Marketing performance is a common, repetitive problem that a model has seen described ten thousand times, so Copilot’s hedge was a fair hypothesis. The error happened in her, where a hedged possibility met a loaded prediction and came out as a fact. When the AI is right, the human is the only place the error can enter, and nothing in the standard toolkit is watching the human.
The Complex Case (The Ambiguous Scenario): a shock with no close precedent, a market regime unseen in a working lifetime. Here the model is at its least reliable, for a reason that matters. It does not know what will happen; it knows what has most often been written about things that look similar. On a rare event, that is the early consensus of commentators, returned in fluent, confident prose, which then feeds the next round of commentary. Read by a leader whose own prediction already points the same way, it does not look like one more opinion. It looks like the verdict. Neither system has seen this event. The model extrapolates from what was written about events that resemble it; the leader extrapolates from experience shaped by the same commentary. When the two guesses line up, the brain reads the match as confirmation, and it does so in half a second. Nothing has been confirmed. Two extrapolations from the same incomplete past have coincided, and the result is a confidently wrong decision, dressed as analysis.
That is where AI parts company with every earlier source of information. With a person, your biased reaction stays in your head. With an AI model, your reaction becomes your next prompt, and the model is built to elaborate whatever bias you feed it. “Make the case for cutting marketing spend” is a prediction that has already hardened into fact, now handed to a model whose job is to make it persuasive. Twenty minutes later it has slides, a chart and a recommendation. AI does not just trigger the micro moment; it industrializes what follows.
Early evidence fits. A 2025 Microsoft Research and Carnegie Mellon survey of 319 knowledge workers across 936 tasks [13] found that the more confident people were in the AI, the less critical thinking they reported. The data are self-reported and correlational, but the signature matches.
Most AI guidance governs only the machine’s side of the equation: better prompts, fact-checking, a human in the loop. Automation bias research adds interface and reliability fixes [14]. Micro Moments addresses the component none of those reach: what happens inside the human the instant the answer lands, and before the next question is typed.
Before, During, After: Where Micro Moments Fits
Almost every tool for better judgment works at one of two points in time. Some work before the decision: Simon’s procedures, Kahneman’s advice to slow down, the decision hygiene of Noise, pre-mortems, checklists, better prompts. Their logic is preparation. Set the conditions right and the judgment will be better. Others work after: post-mortems, audits, fact-checks, the review that catches an error in the finished presentation. Their logic is correction. Find what went wrong and fix it next time.
Both are valuable. Neither is present at the moment the manager read “may be contributing.” Preparation had already happened; review had not yet started. The micro moment fell into the gap between them, and that gap is where Micro Moments works.
The middle column is not empty, and the most important occupant is Gary Klein. His research on recognition-primed decisions [15] studied firefighters, nurses and commanders choosing under time pressure, and showed how experts recognize a situation as a familiar type, simulate one course of action and act. Klein’s question was how expertise works when there is no time to deliberate. Micro Moments asks a different one: whether the situation the expert believes they are recognizing has already been shaped by a prediction they did not notice. Klein’s model is also framed by urgency, which a micro moment does not need; the manager had all afternoon. And it is telling that when Klein wanted a remedy, he built the premortem [16], a Before tool. Even the researcher of the moment reached for preparation.
So the distinctive claim is not that nobody looked at the middle column. It is the combination of three things in one framework: the timing, which is during; the mechanism, which is a prediction experienced as fact; and the scope, which covers urgent and leisurely decisions, high stakes and low, expert and novice alike.
One more map shows why the middle column matters more now.
Ask two questions about any AI answer: is the answer sound or unreliable, and did it set off a prediction you were already carrying? When no prediction fires, the usual before and after checks do their job. But with AI, a prediction usually does fire, because we rarely ask about things we have no view on; we ask about the problems we are already thinking about. If the answer is sound, as Copilot’s hedge was, a fact-check finds nothing wrong, because the error is created in the reading. If the answer is unreliable, as on a rare event, and it agrees with what you already believe, it will not look wrong to you or to anyone who shares your view. Either way, only the reader can catch it, and AI makes the miss more costly: the answer arrives too polished to prompt a second look, and your next prompt turns the mistake into a finished recommendation within minutes.
The Method: OPIJ and the Strategic Use of Doubt
Noticing a micro moment requires knowing what one looks like from the inside. The brain compresses the path from signal to decision into a single felt step: you read the slide and you know. The method’s first job is to decompress that step through four distinct stages abbreviated as OPIJ.
Observation is what arrived: “Marketing leads may be contributing.” Perception is how it registered: a flash of recognition and relief. Interpretation is the story attached: marketing is responsible. Judgment is the commitment: cut the budget. In the manager’s half-second these were one. Written out, the gap becomes obvious. Nothing in the observation says “responsible.” The interpretation came from a stored prediction, and it entered at Perception, as a feeling rather than a thought.
That is not a loose metaphor. Antonio Damasio’s somatic marker hypothesis [17, 18] holds that the body tags options with emotional signals before reasoning begins, that these markers are stored with past experience and reactivated when a similar situation appears, and that they narrow the decision space before we are aware of it. A micro moment is a somatic marker firing on an AI answer. The manager’s months of frustration were stored as a marker; Copilot’s sentence reactivated it; the feeling of certainty did the rest.
Valerie Thompson’s work on the feeling of rightness [19, 20] explains why that feeling is so hard to resist. Across many experiments, the sense that a quick answer is correct was produced not by its accuracy but by how fast it came to mind. A fast answer feels right and ends thinking; a slow one feels doubtful and reopens it. Note where the speed came from in the manager’s case. Not from Copilot’s prose; Thompson found that how easy a text is to read does not change the feeling [21]. It came from her. The answer “marketing” was already loaded, so the sentence retrieved it instantly, and the instant arrival felt like truth.
Damasio and Thompson are describing the same event from two sides. Damasio shows a stored emotional marker firing before reasoning begins. Thompson shows the feeling that marker produces closing the search before reasoning can start. The shared currency is emotion. A micro moment is a stored emotion arriving so fast that it is experienced as knowledge.
Get Andreas Raharso’s stories in your inbox
Join Medium for free to get updates from this writer.
That is why thinking alone cannot intervene. A thought arrives after the search has closed. Emotion alone fails too; an unease with no object goes nowhere. What is needed is a state in which emotion and thinking arrive at the same time, and doubt is that state. Doubt is the one epistemic state that is both thinking and emotion at once, a judgment that something is not yet established, carried by an unease that demands attention [22, 23]. Doubt is the moment when emotion and thinking can work on each other, and so the only moment a micro moment can be managed. Charles Peirce called doubt the irritation that starts inquiry, and belief the calm that ends it [24]. A micro moment is that calm arriving too early. In cognitive terms, doubt is the handle by which emotion can be reached from the inside.
Peter Gollwitzer’s research on implementation intentions [25] shows how: an if–then plan formed in advance lets a cue trigger the response without a conscious decision in the moment. Applied here, the cue is the click of “exactly.” If I feel instantly sure, then I ask what else it could be. Because certainty behaves like a strong habit, and strong habits can override weak plans, the doubt must be practiced until it fires as automatically as the certainty does. That is strategic doubt.
Doubt and OPIJ work as a pair. Doubt brings thinking into the moment; OPIJ gives that thinking a structure, pointing it at the one inference the prediction supplied, so that “is responsible” turns back into “may be.” OPIJ without doubt is a map no one opens. Strategic doubt is not doubt about the AI, or about yourself, but about that interpretation, calibrated to the speed of the certainty and the rarity of the event. Certainty ends search. Doubt, placed exactly there, reopens it.
Case Study: The Copilot Presentation
Before the case, a test. Think of the last time you read a message and knew, instantly, what the sender meant by it. The last time a colleague’s name appeared on a meeting invite and you knew how that meeting would go. The last time an AI answer appeared and you thought, “Exactly.” Each of those was a micro moment: a prediction you already held, touched by something you just saw, arriving so fast that it felt like seeing rather than guessing. You had several today. You noticed none of them, because a micro moment does not feel like anything. It feels like knowing.
Most of them are harmless. The manager’s was not.
You know the setup: two quarters of falling sales, a suspicion about marketing she had carried for months, and a slide that said “may be.” Here is what she did not know, because she had not looked. Qualified leads had been flat for a year. What had changed was her own team’s response time, which doubled after a CRM migration.
Run one: nothing fires
She types her next prompt: “Make the case for cutting marketing spend.” Copilot does exactly that. It finds supporting points, adds a chart, writes a recommendation. Twenty minutes later she has a deck that looks like analysis and is actually her untested hunch in a suit.
Leadership approves a 20 percent cut. Qualified leads fall for the first time in a year. Response time stays broken, because nobody looked at it. The next quarter is worse. No single fact in the deck was false. The damage was done in the half-second before the deck existed.
Run two: doubt fires
Same slide, same instant certainty. But she has trained one if–then habit: if I feel instantly sure, then I ask what else it could be. The click of “exactly” sets off a small unease. Not doubt about Copilot. Not doubt about herself. Doubt about the one leap her mind just made.
Doubt makes her look again, and OPIJ shows her where. The Observation said “may be.” The “is” was her Interpretation, supplied by months of frustration. “Is responsible” turns back into “may be.” Her next prompt is different: “List the competing explanations for the decline and what data would distinguish them.” Four come back. She pulls the two reports she had been avoiding. Leads flat. Response time doubled. The recommendation changes: fix response time, hold marketing spend.
The asymmetry
Had the data confirmed her suspicion, run two would still win: she would present evidence instead of a hunch. Doubt cost forty minutes when she might have been right. It saved a quarter when she was wrong. And because strategic doubt is calibrated, rising with the speed of the certainty and the rarity of the event, it does not turn every email into an investigation. One feeling, placed in one half-second, was the entire difference between the two decks.
Conclusion: Creating Clarity
Return to the manager, half a second after reading “may be contributing.” In the first run, everything that followed, the prompt, the deck, the budget cut, the bad quarter, was decided in that instant, and she never felt it happen. In the second, the same instant was where she undid it. That is a micro moment: a prediction you already hold, touched by something you just saw, arriving so fast it feels like knowing, and the last point at which it can still be caught.
Seventy years of research mapped why capable people misjudge: Simon’s bounded rationality, Tversky and Kahneman’s biases, the noise in our judgments. Its remedies work before a decision or after it. The predictive brain explains what happens in between: perception is a guess, and sometimes the guess is felt as fact. Because that certainty is a feeling, thinking alone arrives too late. Doubt is the one state in which feeling and thinking arrive together. Strategic doubt brings it into the moment, and OPIJ shows it where to look.
AI did not create this problem. It made the instant more frequent and its consequences more polished. Simon helped found the field that now supplies endless answers. The limit he described has moved to the half-second in which we receive them.
Answers are cheap. Clarity is noticing when the question you are about to ask was chosen for you, and asking a better one. Your first thought may be right. Just don’t let it decide before you’ve noticed it.
References
Numbers in square brackets in the text refer to the entries below.
[1] Raharso, A., & Tan, C. (2026). Micro Moments: How Your Mind Sabotages You Everyday. https://www.amazon.co.uk/Micro-Moments-Sabotages-Everyday-Rituals/dp/9819451965
[2] Raharso, A. (2022). Escape from System 1: Unlocking the Science Behind the New Way of Innovation. Marshall Cavendish Business. https://www.marshallcavendish.com/our-books/categories/9789815009316
[3] Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics, 69(1), 99–118. https://doi.org/10.2307/1884852
[4] The Nobel Prize. (1978). The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 1978: Herbert A. Simon. https://www.nobelprize.org/prizes/economic-sciences/1978/simon/facts/
[5] Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131. https://doi.org/10.1126/science.185.4157.1124
[6] Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
[7] Kahneman, D., Sibony, O., & Sunstein, C. R. (2021). Noise: A Flaw in Human Judgment. Little, Brown Spark.
[8] Sunstein, C. R. (2021). Interview on Noise, in which noise is described as a statistical rather than a causal concept. Big Brains podcast, University of Chicago. https://news.uchicago.edu/big-brains-podcast-noise-judgment-cass-sunstein-kahneman-sibony
[9] Sibony, O. (2021). Interview on Noise, in which decision hygiene is compared to handwashing. HEC Paris. https://www.hec.edu/en/faculty-research/news/hec-academic-makes-noise-study-flaw-human-judgment
[10] Seth, A. (2021). Being You: A New Science of Consciousness. Faber & Faber.
[11] Barrett, L. F. (2017). How Emotions Are Made: The Secret Life of the Brain. Houghton Mifflin Harcourt.
[12] Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36(3), 181–204. https://doi.org/10.1017/S0140525X12000477
[13] Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3706598.3713778
[14] Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381–410. https://doi.org/10.1177/0018720810376055
[15] Klein, G. (1998). Sources of Power: How People Make Decisions. MIT Press.
[16] Klein, G. (2007, September). Performing a project premortem. Harvard Business Review. https://hbr.org/2007/09/performing-a-project-premortem
[17] Damasio, A. R. (1994). Descartes’ Error: Emotion, Reason, and the Human Brain. Putnam.
[18] Bechara, A., Damasio, H., & Damasio, A. R. (2000). Emotion, decision making and the orbitofrontal cortex. Cerebral Cortex, 10(3), 295–307. https://doi.org/10.1093/cercor/10.3.295
[19] Thompson, V. A., Prowse Turner, J. A., & Pennycook, G. (2011). Intuition, reason, and metacognition. Cognitive Psychology, 63(3), 107–140. https://doi.org/10.1016/j.cogpsych.2011.06.001
[20] Thompson, V. A., & Markovits, H. (2025). Fast reasoning and metacognition. Psychonomic Bulletin & Review. https://doi.org/10.3758/s13423-025-02662-0
[21] Thompson, V. A., Prowse Turner, J. A., Pennycook, G., Ball, L. J., Brack, H., Ophir, Y., & Ackerman, R. (2013). The role of answer fluency and perceptual fluency as metacognitive cues for initiating analytic thinking. Cognition, 128(2), 237–251. https://doi.org/10.1016/j.cognition.2012.09.012
[22] Vazard, J. (2021). (Un)reasonable doubt as affective experience: Obsessive–compulsive disorder, epistemic anxiety and the feeling of uncertainty. Synthese, 199, 6917–6934. https://pmc.ncbi.nlm.nih.gov/articles/PMC9336137/
[23] Cortial, C., Prado, J., & Caparos, S. (2025). Reconceptualizing metacognitive experience in dual-process reasoning: The role of emotion in triggering deliberation. Cognitive Science, 49, e70084. https://doi.org/10.1111/cogs.70084
[24] Peirce, C. S. (1877). The fixation of belief. Popular Science Monthly, 12, 1–15. https://www.peirce.org/writings/p107.html
[25] Gollwitzer, P. M. (1999). Implementation intentions: Strong effects of simple plans. American Psychologist, 54(7), 493–503. https://doi.org/10.1037/0003-066X.54.7.493
Additional sources for the After column of Figure 2
[26] Argyris, C. (1977). Double loop learning in organizations. Harvard Business Review, 55(5), 115–125. https://hbr.org/1977/09/double-loop-learning-in-organizations
[27] Tetlock, P. E., & Gardner, D. (2015). Superforecasting: The Art and Science of Prediction. Crown.
[28] Edmondson, A. C. (2023). Right Kind of Wrong: The Science of Failing Well. Atria Books.
[29] Duke, A. (2018). Thinking in Bets: Making Smarter Decisions When You Don’t Have All the Facts. Portfolio.
