Learning in the Age of AI

Our brains are not optimized for comfort — they are optimized for survival through problem-solving. Struggle is not a bug; it is the mechanism. The desirable difficulty of working through hard things is precisely what builds durable, transferable understanding. When you skip the struggle, you skip the wiring.

More importantly, the most consequential ideas in history rarely came from experts staying inside their lane. In an interview I once watched, Leslie Lamport — the architect of distributed systems thinking, Turing Award laureate — explained that he invented the Lamport clock and his foundational work on consensus partly because he came to computer science steeped in physics and the theory of relativity. He saw time differently than his peers. His cross-domain intuition was the innovation. Multidisciplinary learning is not a soft virtue. It is how breakthroughs happen.

There is a profound asymmetry here worth sitting with. Large language models are trained on human-produced text — a vast, rich, and deeply incomplete proxy for reality. Human beings learn from direct experience, from embodied sensation, from watching systems fail and recover in real time.

Ray Dalio, in his book Principles, argues that nature itself is the best model for understanding how complex systems work — how ecosystems balance, how organisms adapt, how evolution selects. He urges us to look at nature not metaphorically but practically, as a source of design patterns for human problems. Your accumulated experience — the books you have read, the failures you have absorbed, the mentors you have watched, the environments you have navigated — forms a model of the world that is richer and stranger than any corpus of text. AI can compress human knowledge; it cannot replicate human experience.

Here is the practical paradox at the heart of the AI era: the less you know, the less you can extract from these tools. A language model will faithfully answer the question you ask it. The problem is that shallow questions produce shallow answers. The ability to ask a precise, probing, well-framed question is not a trivial skill — it is the distillation of expertise. It requires knowing the shape of a problem, understanding its boundaries, recognizing where common wisdom tends to go wrong.

A great engineer does not get value from AI by typing vague prompts and accepting the output. A great engineer uses AI as a thinking partner — pushing back, probing edge cases, synthesizing multiple answers into a coherent design. That capacity for critical engagement comes from years of learning. AI does not diminish the need for expertise. It raises the premium on deep expertise while ruthlessly commoditizing shallow expertise.

This last point is personal, and I share it as such.

For me, learning is not a strategy for career advancement. It is something closer to a biological need — like eating. When I stop learning, I feel stagnant. I feel hungry. I have left jobs not over salary disputes but because the learning had stopped, because I could feel myself going through the motions of work without growing through them. I may not be the wealthiest person in any room I walk into. But I am rarely the most restless, and I am almost never bored.

There is a word for people who have satisfied all their material needs but stopped being curious about the world: they are called finished. I do not want to be finished. I suspect you don't either.