AI Demands Pattern Thinking That Schools Don’t Teach
I think most people, even those otherwise knowledgeable about AI, haven’t fully absorbed what’s different about this technology.
There are many ways to characterize those differences, but the one that has more impact on knowing when and how to use AI is that prior technology was entirely rational whereas Generative AI is at heart a pattern transforming technology. It’s where I started with the first chapter of my first book describing durable AI knowledge and skills education should focus on (AI Wisdom Volume 1: Meta-Principles of Thinking and Learning).
I began there because it’s the root notion that infuses all the bells and whistles added on top. And yet it’s such an abstract notion that the monumental impact of the reasoning vs. pattern-oriented pivot is not easily absorbed. It is an ongoing process for me, 38 years after building my first artificial neural net. This article tries to articulate the impact and the teaching imperative.
Schooling focuses on our deliberative thinking, and especially rational thinking. The goal is to describe a problem and its solution or paths to one so that reason can be applied. We’ve been taught, often just by emphasis rather than by anyone saying so directly, that careful reasoning is the reliable part of cognition and that everything else is the soft, imprecise stuff we should be trying to get past. That assumption shapes how institutions are run, how students are assessed, and — very visibly right now — how educators and leaders are responding to AI.
But reason is never the neatly divided box where truth lives. Our brains aren’t like prior technology either. They too are pattern transformers. Down underneath our attempts to cordon off thinking to the manageable parts is a pattern-oriented gremlin telling us something isn’t quite right.
The first thing educators often want to show students is how AI gets things wrong. Hallucinations dominate the concern. This is a perfectly rational-world test applied to a system that mostly doesn’t operate in the rational world (though recent AI training emphasis on rationality I think weakens aspects of its natural pattern talents, but that’s a subject for another day). Plus, it’s often with the contrast of humanity as Mother Theresa, Plato, and Michelangelo rolled into one.
Factual accuracy and the rational means to get there is one narrow form of knowledge. The feel of a strong argument, the read of a room, the judgment about which approach is worth pursuing are knowledge too, but they don’t show up on the test. They don’t register as knowledge to evaluators trained to trust only what can be stated and verified. Pattern-based knowledge becomes the weak and unreliable step-sister.
The fuller picture is that knowledge in its most fundamental form is pattern-based. The explicit, logical, expressible version we’ve built education around is a useful subset — sometimes a very powerful one—but it is a subset. AI is forcing that recognition, and whether institutions can catch up to it will have a lot to do with how well they fare.
Reasoning Is a Pattern, Just a Very Constrained One
Reasoning and patterns aren’t distinct. Reasoning is a highly trained, highly constrained form of patterns.
The entire brain—including the parts doing deliberate, careful analysis—is built from neural networks. Not metaphorically. Literally. And a neural network is an associative, distributed, pattern-processing structure by its physical nature. There is no separate logical processor running clean deductive operations somewhere in the prefrontal cortex or a language center that just processes letter symbols. When you reason carefully and explicitly, you are still running pattern operations, ones that have been trained through education and experience to respect logical constraints. The rules of logic are themselves learned patterns.
Plus, once reasoning drifts toward creativity, patterns rise up. Executing a math proof—following the logical steps once the path is known—is something symbolic computers have done since long before generative AI. What mathematicians actually do, the hard part, is figure out how to prove something, or if it can. That requires skill in which direction to explore, which approach has the feel of something that might work, or when a line of attack is quietly dead. Those are judgments, often intuitive, that are driven from natively pattern-based neural nets that didn’t try to put its knowledge into language.
In practice this means reasoning and pattern recognition aren’t two separate processes competing for the wheel. A conclusion often arrives first, assembled from experience, values, and prior pattern recognition, and the reasoning follows to express, test, and refine it. That’s not a character flaw; the best scientist in the world more often than not uses the pattern recognizer before the “thinker.” It’s how the architecture works. The full pattern has more information than the portion of the pattern we manage to articulate and reason about. It’s set up to be better at noticing something’s off, to powering a wise “gut feel.” Or, if left untrained or unleveraged, to delude and oversimplify.
Reasoning lets us check conclusions, communicate them, and build on them across people and time. The problem is treating the reasoning as the whole story, which leaves us blind to what shaped it and where it might have gone astray.
Education’s response to this architecture has been, for understandable institutional reasons, to teach the expressible approximation of things and treat it as the thing itself. A rubric for strong writing captures some of what makes writing strong, but it can never be all of what good writing is. It’s like evaluating somebody on their ability to recognize faces by grading how well they recognize chins, eyes, and noses. It’s why expertise is hard to teach; so much of it lives in the unspeaking pattern analyzer. Ask an expert how they make their best decisions and you get either a retrospective story that sounds more logical than the actual decision was, or “it depends.” That second response isn’t evasion. It’s an accurate report that no set of axioms and advice covers all the situations. The real knowledge lives in the pattern, and the pattern doesn’t compress cleanly into language.
Institutions built around teaching and testing defaulted to the expressible, logical, and rule-governed. Over time that manageability got mistaken for validity. Pattern-based skill is still cultivated to some degree, in literature, in the arts, in the mysteriousness of the “aha.” But it tends not to be recognized as a cognitive mode that transfers or that can be deliberately developed. The result is a long lag before people develop genuine expertise.
Three Pattern Domains
Patterns aren’t a feature of certain kinds of problems. They’re the fundamental structure of how brains work, how AI works, and how the world works.
The brain point I’ve already discussed. Generative AIs, at their core, are pattern-learning systems. They don’t reason from first principles. They detect and respond to patterns at a scale no individual human brain approaches—across language, concept, context, and domain simultaneously. When AI generates a nuanced analysis, an example that illuminates a hard concept, or catches an implied tension in an argument, that’s pattern work that couldn’t be done well by a rational engine.
It also means that working with AI well is itself a pattern skill. Feeding it examples often outperforms feeding it a boatload of text. Showing it the shape of what you want—through cases, contrasts, and context—works better than specifying requirements. That runs counter to how most formal training teaches people to communicate.
Paradoxically, LLMs are trained on language, which means it necessarily throws away aspects of deeper, somewhat nonlingual, patterns. It explains a lot of why the patterns they find seem to not understand the between-the-lines stuff, or is naive to what someone experiencing the world would intuit. A lot might change with training data from world experience (e.g., from robots).
The world is the biggest of the three. The problems that matter most—how organizations actually change, how learning happens, how conflicts escalate, what economic behavior looks like on the ground—are irreducibly pattern-complex. They never fully yield to rational-world analysis. The experts who navigate them best have developed a rich pattern sense through deep, varied experience. They know when a rule applies and when the situation has moved outside the territory the rule was built for. Reasoning is indispensable for these problems. It is not sufficient for them.
Because AI operates on patterns, it creates a specific kind of pressure on how we think. Recognizing where it can genuinely help requires developing a pattern-oriented read of the problems you’re facing. Less “can it follow the rules of this task” and more “is this a problem where pattern recognition at scale adds something?” That’s a different evaluative instinct than most people have had occasion to develop.
Understanding your own pattern vulnerabilities matters too, like the tendency to trust the confident, logical-sounding conclusion, to underweight the vague but signal-rich read, or to impose a reasoning framework on a situation that doesn’t fit one.
It’s also helps in reading others. Someone who doesn’t respond to a logical argument the way you expect isn’t necessarily being irrational; they may be working from different patterns arising from different training. That reframe is useful for anyone who teaches, leads, or tries to change how an organization behaves.
These aren’t skills you get from a workshop on AI. They come from understanding something more fundamental about how thinking works and where it goes wrong. We need an education system willing to take pattern-based knowledge as seriously as it has always taken the kind you can put on a rubric.
Intuitive, pattern thinking about AI is where my AI Wisdom book series focuses if you want an example of how to go about teaching it.
©2026 Dasey Consulting LLC
This article was written with the assistance of AI. The typical way in which I apply AI to the process is detailed in a prior article.