I used to think AI would make memorizing facts pointless. Building with AI every day changed my mind.
When my kids complain about having to memorize things, I get it. Google knows it, the calculator does the math, AI writes the paragraph. Why keep any of it in your head?
For a long time I didn’t have a good answer. Now I do.
A model in full reasoning mode is great for real analysis and a waste for anything it already knows. It shouldn’t re-derive the Pythagorean theorem every time a geometry problem comes up; it should know the theorem and get on with the problem. Thinking is expensive. For a machine, facts are cheap to store and instant to retrieve. So you cache the facts and save the expensive thinking for where it actually counts.
People work in a similar way. The cognitive scientist Daniel Willingham (who wrote Why Don’t Students Like School) has spent a career on this: the basics you know cold free your working memory for actual thinking, and you can’t think critically about things you don’t know. A chess master doesn’t out-calculate you move by move. He’s seen so many positions that he recognizes them instantly, which frees his mind for the hard part. The facts in your head aren’t the opposite of thinking. They’re what makes good thinking fast.
The storage isn’t free for people, though. Memorizing costs real time and effort, which is exactly why it matters what you spend it on. Schools have made kids memorize plenty that was forgotten by June and never missed. The case isn’t for hoarding every fact. It’s for paying up front for the ones you’ll retrieve for the rest of your life.
The old story was that expertise mattered mainly for catching AI when it makes things up. That part hasn’t gone away. If anything, it got harder because better models make more polished mistakes, and a polished mistake fools the novice, not the expert. But the bigger edge now sits upstream of the answer. It’s asking the right question, and more and more it’s directing the right process. We build systems where different models handle different jobs, and someone has to know which task needs slow reasoning, which needs a quick lookup, and how to string them together. That judgment is a trained eye, a kind of pattern literacy, and it comes from knowing a lot.
Which is why expertise compounds. A good doctor or lawyer or mathematician gets more out of AI than a beginner, not less. They’ve run the problem a thousand times in their own head, so they can smell a plausible wrong answer and spot a dead-end approach before wasting a day on it.
So that’s the answer my kids get now. Knowing things became more valuable in the age of AI, because knowing enough is what lets you ask the right questions and direct the work. AI can help you learn it, but it can’t know it for you. That’s what the multiplication table was for all along