In 1973, researchers did an experiment on how chess masters think differently from novices. They showed both players a position and asked them to reconstruct it from memory, giving both equal amounts of time. The grandmaster recalled a nearly identical position, while the novice remembered 51%. 

Years of playing chess hadn’t optimized the master for memorization; however, when future 

researchers replicated the experiment and asked the players to memorize a random position that couldn’t have come from a real game, the master struggled just as much as the novice. What happened was that the master and novice saw different things in the initial experiment. The novice saw 20 separate pieces, while the grandmaster could memorize in chunks of concepts, like 4-5 pieces making up the castled king position. Illogically placed pieces took away the thousands of hours of training in pattern recognition. 

Now, many students use artificial intelligence in a way that takes away the trial-and-error 

process. While many believe chatbots like Claude or Gemini were a godsend during cramming, neuroscience reveals that the confusion is learning, and machines are removing it. 

Learning scientists call this “desirable difficulty.” The harder one’s brain works to retrieve or reconstruct the information, the longer and better it’s retained next time. This means: flashcards are better than rereading, writing what you remember is better than highlighting, and problem sets are better than YouTube videos. 

This article is far from a protest against the most important productivity tool created this century, if not ever. What I hope to accomplish with it is to make an argument that it should be used for its intended purpose: as a tool, not a teacher. 

Here are the four methods I use to turn the same responses into actual learning: 

Close and reconstruct

‍Drawing on the earlier principle I mentioned about how writing from memory is better than rereading, learning happens when you reread the AI response and try to write down what you understood in your own words. If you cannot remember or explain a concept, you’ll know exactly which pieces of information were received without processing. Go back to the chatbot and ask about the specific concept you struggled with. Rinse and repeat. 

Flip the conversation 

If you’re not learning the material for the first time, it’s a good idea to ask it to quiz you instead of explaining it to you. You can phrase your prompt like: “I just read about [topic]. Give me five questions, and don’t tell me if I’m right until I’ve answered all of them.” It’s one of the most evidence-backed methods of learning, and something AI almost never does on its own, because no one asks. 

Teach it back 

This draws on the principle used by Richard Feynman, one of the most accomplished theoretical physicists of all time. Pretend like you’re trying to teach the chatbot, and ask “what did I get wrong, oversimplify, or miss?” Through explanations, you’re able to see what you don’t know. That’s half the game, and the AI closes the learning loop by going over all the missed information. After this, though, it’s important that you write the corrected version from scratch using the first method to make sure the AI response wasn’t consumed passively. 

Attempt the problem before you read the answer

This is perhaps the single most important piece of advice: never ask it to solve a question without first trying it yourself. Failing it on purpose and asking for the explanation is called “productive failure” by learning scientists, because those who attempt the question first retain the concept better and show a more nuanced understanding of it. It’s like taking a test and being shown the answer key with justification the next day, only that the justification can explain your specific confusion. 

None of these methods takes longer than the rereading and highlighting most students do right now. Those who use AI will learn faster than anyone around them, and that knowledge will be carried with them when they enter rooms where the laptop is closed and the answer isn’t readily available. Over time, the effects of these methods will compound and supercharge the mastery of any concept. Those who skip these steps and go straight to the answer will find themselves realizing that closing the tab took everything with it. \

The machine can process anything; it can even think for you. What it cannot do is understand on your behalf. That part will always be yours.