The Friction Advantage

Here is an uncomfortable claim: the easier a tool makes learning feel, the less you are probably learning. That is not a complaint about technology. It is a finding that cognitive scientists settled decades before anyone typed a question into a chatbot. The feeling of understanding and the fact of understanding are two different things, and modern AI is the most powerful machine ever built for manufacturing the feeling.

I have always been fascinated by how to learn, long before AI entered the picture. So the question I keep turning over is simple. Can AI actually teach us to learn smarter, without burning out? The honest answer is yes, but only if we deliberately use it against our instincts. Left to instinct, we reach for whatever feels fluent and fast. Learning lives in the opposite place.

What the science already settled

The techniques that reliably build durable knowledge are not glamorous, and that is exactly why people avoid them. A landmark review by Dunlosky and colleagues graded dozens of study strategies and found that the two with the strongest evidence were practice testing (quizzing yourself instead of re-reading) and distributed practice (spacing study over time instead of cramming). Re-reading and highlighting, the strategies most students actually use, scored near the bottom.

Two other ideas round out the toolkit. Interleaving means mixing related topics rather than drilling one at a time, which forces your brain to choose the right approach instead of running on autopilot. Elaboration means explaining why something is true, in your own words, until the explanation holds together. Roediger and Karpicke showed that the simple act of being tested, even with no new studying, produces far better long-term retention than additional reading. Testing is not how we measure learning. Testing is how we create it.

The strategies that feel productive (re-reading, highlighting, fluent summaries) are the weakest. The strategies that feel like struggle (self-testing, spacing, explaining) are the strongest. Comfort is a poor signal for learning.

What AI genuinely adds

If self-testing and spacing are the engine, AI is a remarkable supply of fuel. It can generate an endless stream of practice questions on any topic, at any difficulty, in seconds. It can play the role of a confused student you have to teach, which is the Feynman technique with a pulse: you only discover the holes in your understanding when you try to explain it to something that pushes back. And it can calibrate difficulty in real time, keeping you in the zone where the work is hard enough to matter but not so hard that you quit.

That last capability is the real prize. Robert Bjork named this zone with the phrase desirable difficulty: the level of challenge where retrieval is effortful but successful. Too easy and nothing sticks. Too hard and you disengage. A well-prompted AI tutor can hold you at that edge better than a static textbook ever could, because it adapts to your answers. Used this way, AI does not replace the cognitive work. It manufactures more of the right kind.

The trap hiding in plain sight

Now the twist, and it connects to something I have long believed about books. Meta-analyses covering more than 170,000 readers have found a consistent screen inferiority effect: we comprehend informational text worse on screens than on paper. The unsettling part is that the gap has grown over time rather than shrunk, even as digital natives became the majority of readers. Familiarity with screens did not fix it.

The cause is not the pixels. It is the posture. Screens invite shallow, scanning, fast processing, and they breed overconfidence: we feel we understood, but we retained less. Print forces a slower pace and a deeper engagement almost by default. AI carries the exact same risk, amplified. A fluent, well-organized summary feels like understanding. It very often is not. The smoother the answer, the easier it is to mistake reading about something for knowing it.

So the thread connecting print's quiet advantage and AI's hidden danger comes down to one word: effort. Print preserves the friction where learning happens. AI can dissolve it, handing us answers so polished that we skip the productive struggle entirely. The friction is not a flaw in the process. The friction is the process.

And what about burnout?

It is tempting to assume that if AI makes learning more efficient, it must also reduce exhaustion. The evidence points elsewhere. Burnout is driven less by inefficiency and more by sustained workload without recovery. Efficiency tools tend to backfire here, because the time they free up quietly refills with more work. AI helps with burnout only if we treat the hours it returns as recovery and depth, not as room to cram more in.

This reframes the whole question. The goal is not to do more learning faster. It is to do the right learning, protect the friction that makes it stick, and protect the rest that makes it sustainable. Speed without depth is just faster forgetting.

Putting it into practice

The practical pattern is straightforward once you accept the principle. Read deeply, and for anything that truly matters, read it on paper or in a deliberately slow mode. Then use AI as a sparring partner rather than an oracle. Ask it to quiz you before you ask it to explain. Make it grade your explanation, not hand you one. A simple prompt changes everything: instead of summarize this for me, try ask me five questions on this, wait for each answer, then tell me what I got wrong and why.

The principle is to keep the cognitive work on your side of the keyboard. The moment the AI is doing the retrieving, the connecting, and the explaining, it is the one getting smarter, not you. Let it generate the difficulty. Insist on doing the struggle yourself.

What leaders should do

For teams investing in AI to upskill their people, the temptation is to measure adoption and speed. Those metrics reward exactly the shallow processing that erodes real capability. Lead instead toward effort, retention, and recovery.

1. Reward retention, not consumption. Stop measuring training by hours logged or articles read. Measure whether people can apply and explain what they learned weeks later, using short low-stakes quizzes built and graded by AI.
2. Make AI test, not tell. Set the default expectation that learners prompt AI to question them first and explain second. Share the sparring-partner prompts so the habit spreads.
3. Protect deep reading. For material that genuinely matters, preserve a slow, paper or distraction-free mode, and resist replacing primary sources with AI summaries.
4. Spend efficiency on recovery. When AI frees up time, deliberately allocate part of it to rest and depth rather than backfilling it with more tasks. Sustainable learning needs slack.

The skill of this era is not avoiding digital tools, and it is certainly not avoiding AI. It is knowing precisely when to reintroduce the productive struggle that makes learning stick, and having the discipline to keep that struggle for yourself. The organizations that learn this will compound real capability. The ones that chase frictionless fluency will feel faster, and quietly forget more.