AI can answer anything. So why can't you learn from it?

Most of the time, AI explains things just fine. But every now and then, you get an answer that’s perfectly logical, perfectly structured — and you still have to read it three times. Then a YouTuber explains the same concept in 30 seconds and you get it immediately.

AI has the knowledge. What it doesn’t have is a model of how you think.

And this problem existed long before AI. Think about how you actually consume information every day — you listen to a 50-minute lecture and walk away remembering almost nothing. You buy a book and months later you’re still on page 12. You watch a tutorial 3 times and get stuck at the same point every time. You ask AI to summarize the key points, and the summary is perfectly fine, but it still doesn’t land.

Most people blame themselves. Not focused enough, or not smart enough.

It’s not you. Everything you’re reading, watching, and listening to was structured for someone else’s mental model.

I’ve been thinking about this a lot recently, and I have a hypothesis: We solved finding information a long time ago, but what we never solved is how it lands.

Algorithms know what to show you, but they don’t care if you understand it.

What you like, what you’ve clicked, how long you stayed, how fast you scrolled — they know which video will make you pause for two extra seconds before you know it yourself. But all they do is select what content to push to you.

Whether you actually understand it after clicking? Not their problem. You get recommended an article you “might be interested in,” but if the writing style, structure, and angle don’t match how your brain receives information — you still can’t get through it. The algorithm’s goal is to get you to click, not to get you to understand.

All the personalization on the internet stops at “what to show you.” Very few is working on “how to show it to you.”

Photo: Courtesy of Tom Brown, via Atlas Obscura

Content creators face the same problem from the other side.

Everyone who makes content has been taught the same lesson: know your target audience. Before you create, figure out who you’re talking to. Speak their language. Match their taste.

But do we actually know them?

The data we have — click-through rates, watch-through rates, time on page, engagement — is all behavioral. You know they clicked, but you don’t know if they understood. You know they shared it, but you can’t tell whether it’s because they genuinely agreed with it or because sharing it makes them look smart.

The brain’s process of receiving and processing information is extraordinarily complex, but we’ve compressed all that feedback into a few simple user interactions — a like, a scroll, a few seconds of dwell time. It’s like judging whether someone understood you based on whether they nodded. When I took cognitive science classes, experiments related to the brain required controlling for environmental variables one by one, because any small factor could affect the outcome. And here we are making content, looking at a handful of crude behavioral metrics and calling that “understanding our users.”

How someone’s brain actually processes what you give them, what their cognitive structure looks like, what kind of framing could break through their comprehension barrier — current data can’t see any of this.

So when we make content, we’re essentially shouting at a blurry silhouette. We kind of know who our audience are, but have no idea how they listen.

Attention is the same story.

Even when the content is right and the topic genuinely interests you, you still might not be able to absorb it. Because everyone sustains focus differently.

Some people can sit with a book for an hour without getting distracted. Others need to switch gears every 15 minutes. Some have to take notes while they listen before they can fully engage. Others only absorb information when they’re walking and listening at the same time. Some need a strong hook to capture their attention. Others need a gradual build-up or they’ll feel overwhelmed.

A lot of this is physiological — ADHD, neurodivergence, differences in dopamine regulation, variations in working memory capacity. The conditions under which your brain enters a flow state are fundamentally different from someone else’s.

But all content assumes you can receive information in the same way, at the same pace. A long article is a long article. A 60-minute podcast is 60 minutes. It doesn’t know you zoned out at minute 8. It doesn’t know that what you need isn’t a shorter version — it’s a different rhythm. And if one piece of content doesn’t hold your attention? The solution is always to find another one, which means more work for the audience.

Cognitive science has been studying this for decades.

David Ausubel said: “The most important single factor influencing learning is what the learner already knows.”

Your brain has an existing architecture of knowledge — cognitive science calls these “schemas.” When new information comes in, it has to dock with an existing schema to be truly understood. If it can’t connect, it slides right past you. That feeling of reading an entire paragraph and retaining nothing? That’s what’s happening. It has nothing to do with how well the information was written.

Dedre Gentner’s research explains why analogies are so effective. She found that analogy isn’t about surface resemblance — it’s about structural alignment. Tell someone who loves cooking that “machine learning is like seasoning — you add some salt, taste it, too much so you back off, too little so you add more, and after hundreds of dishes you know the right amount without tasting” — they get it instantly. Explain gradient descent and loss functions to the same person, and they’ll need to read it five times.

Same information. The first version plugged into a structure their brain already had, whereas the second asked them to build one from scratch.

But shouldn’t learning be hard?

Friction is supposed to challenge the brain. If AI translates everything into a frictionless experience, is the brain even thinking?

Cognitive scientist Robert Bjork studied this in 1994 and found something counterintuitive: the moment you feel like “I’ve got it” is often the moment you actually don’t. A tutorial explains something perfectly clearly, you nod along feeling like you understand completely — three days later you try to do it yourself and remember nothing. But a problem you struggled with for hours, tried three wrong approaches, and finally cracked? You remember that for years.

Bjork called this the fluency trap. Your brain mistakes “I’m reading this smoothly” for “I understand this.” The smoother the learning feels, the less it sticks. He called the productive kind of struggle “desirable difficulties.” But he drew a line: difficulty is only desirable when you have the capacity to engage with it. When you’re using what you already know to reach for a new concept — stretching, failing, adjusting, reaching again — that’s where real understanding happens. That struggle is meaningful.

But if you can’t even connect with the foundational concepts, all your energy goes into fighting the way the material is presented — unfamiliar terminology, unclear structure, an angle that has nothing to do with your background. You’re exhausted before you even get to the actual thinking. That struggle is wasted.

Cognitive translation removes the second kind. Strip away the unproductive friction so all your energy goes into thinking that actually matters. Not to make learning easier, but to make it purer.

So what is AI actually missing?

It has enough knowledge, enough logic and reasoning skills, enough language skills. But it doesn’t know these three things about you:

1. What you already know. Your background, your prior knowledge to the domain, your mental models. So it knows where to start and what to skip.

2. How you process new information. Big picture first or details first? Analogies or data? Do you need to know “why it matters” before “how it works”? What’s your attention rhythm — do you need dense short sentences or expansive long ones? Do you need a concrete example every few minutes to pull you back in, or do you prefer to stay immersed in a single chain of reasoning?

3. Your analogy library. You’re an engineer — map new concepts through engineering language. You’ve worked in restaurants — explain supply chains through kitchen logic. You play basketball — teach game theory through court strategy. Your brain already has an entire system for understanding the world. There’s just never been a tool that uses it.

With these three things, AI could re-encode any material — papers, podcasts, financial reports, technical docs — according to your knowledge structure, attention rhythm, and comprehension style. Remove the unproductive friction. Preserve the valuable challenge.

That’s not simplification, more like cognitive translation.

I started thinking about all of this while building my own product.

I’m building a content writing system that helps everyone find their own voice. Along the way I realized this is only one direction of cognitive translation — the output side. You have ideas in your head, and the tool helps you turn them into content that sounds like you.

But what about the input side? Everything you read, hear, and watch every day is encoded according to someone else’s cognitive patterns. Your brain is constantly running a translation process in the background, and that translation might take more energy than actually understanding the content.

Algorithms solved “what to see.” AI solved “what to find.” But “how to get information into your brain in a way that matches how you actually think” — according to your knowledge structure, your analogy library, your attention rhythm, removing the unproductive resistance and preserving the valuable challenge — we rarely see this kind of solution.

I don’t know what this looks like in the end. But the direction feels right.

We have built AI that knows everything. Now we need AI that knows you.

References: Ausubel, Educational Psychology: A Cognitive View (1968); Gentner, Structure-Mapping Theory (1983); Bjork & Bjork, Desirable Difficulties (1994)