한국어

Is Memorization Still Important in the Age of AI?

At a glance
  • You don't need to memorize everything. But the core terms and ideas of a field need to be in your head.
  • Students who knew baseball remembered a baseball story better, whether they were strong or weak readers.
  • In a meta-analysis, people who knew more before studying also knew more after (r = .53).
  • On average, though, people who knew more did not gain more. Prior knowledge and gains were barely related (r = −.06).
  • Let AI hold what you only need to look up. Keep what you think with in your head, by recalling it with the page closed.

Memorization still matters in the age of AI. You don't need to memorize everything, but the core terms and ideas of a field have to be in your head, because that is what you use to read an AI's answer, understand it, and notice when it is wrong. The research does not say "the more you know, the faster you learn." It says the knowledge in your head changes what you understand and keep from the same text.

If AI knows everything, why memorize anything?

Because what AI knows and what you know do different jobs. An AI's answer is only useful once you read and understand it, and you read and understand with what is already in your head.

An old experiment shows this well. Recht and Leslie (1988) split 64 junior high students into four groups by reading ability (high or low) and by how much they knew about baseball (a lot or a little). Everyone read an account of half an inning of a baseball game. Then they acted it out by moving figures, retold it, and after doing another task in between, summarized the game and sorted sentences by importance.

The result was the same on every measure. The students who knew baseball remembered more and summarized better. And this effect did not interact with reading ability: weak readers who knew baseball got the benefit too.

How much you keep from a text depends not only on how well you read, but on how much you already know about its subject.

An AI's answer is a text too. If you don't know the field, you understand and keep less of the same answer. "I can just ask" is only true up to the moment the answer arrives.

Does knowing more help you learn new things faster?

Not necessarily. People who knew more before studying also knew more after, but on average they did not gain more.

Simonsmeier and colleagues (2022) pooled 8,776 effect sizes in a meta-analysis. The correlation between knowledge before and after studying was high, .534: people who started ahead stayed ahead. But the correlation between prior knowledge and how much people gained was −.059, close to zero. Studies varied so much that the range where a new study's result would likely fall ran from −.688 to .621.

How prior knowledge relates to new gains varied widely between studies

-0.8-0.400.40.8Correlation (r)LearninggainsLearning gains · -0.688~0.621-0.688~0.621average −.059
Correlation (r) between prior knowledge and normalized learning gains. The bar is the 95% prediction interval, the range a single new study's result is expected to fall in. The average was −.059.Source: Simonsmeier et al. (2022), abstract
See the numbers
FromTo
Learning gains-0.6880.621

The authors wrote that this spread rules out both "knowledge is power" and "prior knowledge doesn't matter" as general statements. They called for research on the conditions under which prior knowledge helps, hurts, or makes little difference.

One study found a case where prior knowledge went with more learning. Witherby and Carpenter (2022) tested students' knowledge of football and cooking, then had them learn new facts in both domains by answering questions and getting feedback. To make sure the facts were new to everyone, they used information that was false. Across three experiments, the more students knew about a domain, the more new facts they learned in that domain, but not in the other one. This study measured the knowledge students already had; it did not assign it. In the third experiment, students rated how curious they were about each item, and that curiosity mediated the link. Students were more curious about items in a domain they knew, and that curiosity went with learning more.

On the other side, in an experiment that assigned knowledge at random, Buchin and Mulligan (2025) randomly assigned people to three days of training in one of two academic domains, then had them learn new material. This time, the knowledge created by training did not change how much new material people learned.

So no one can promise that memorizing now will speed up your next round of studying. What is clear is that knowledge accumulates. What you know today is the base of what you will know when you finish.

What do experts remember differently?

Experts remember more of what they glimpse in their field, and some of that advantage survives even when the material is scrambled. The authors read this as a sign that experts hold many small patterns from their field (chunks) in long-term memory.

Sala and Gobet (2017) gathered 45 studies (55 experiments, 1,401 people) that compared experts and novices on "random material": chess pieces placed at random, program lines shuffled, and so on. Experts remembered more in 49 of the 55 experiments. In the 24 studies (903 people) with enough data to compute it, the overall correlation between expertise and recall was .41, and .29 after correcting for publication bias. Music showed the largest effect, which the authors put down to the music material being less random.

Experts remembered more even when the material was scrambled

00.250.50.751Expertise and recall · Music · 0.690.69MusicExpertise and recall · Games · 0.420.42GamesExpertise and recall · Program- ming · 0.360.36Program-mingExpertise and recall · Sports · 0.30.3SportsExpertise and recall · Other · 0.240.24Otheroverall .41
Correlation (r) between expertise and recall of randomly arranged domain material (pieces, notes, code lines). Games are mostly chess, plus bridge. Viewing time ran from seconds for game positions and notes to minutes for programs. Sports and other domains were not statistically significant. Overall .41 (24 studies); .29 after correcting for publication bias.Source: Sala & Gobet (2017), Results, moderator analyses
See the numbers
Expertise and recall
Music0.69
Games0.42
Program- ming0.36
Sports0.3
Other0.24

With meaningful material the gap is bigger. The authors note that effect sizes for real positions or real scores are about 40 to 50% larger than for random material.

More chunks in your head means more you can take in at once. When you read a long AI answer, the terms and ideas you already know go by as chunks, and your attention is free for the new parts. That last point is our reading of this research, not something these studies measured.

Can you tell when AI is wrong?

Not easily, if you don't know the material. On a task where AI was weak, people using AI got it wrong more often.

Dell'Acqua and colleagues (2026) worked with the consulting firm BCG and randomly assigned 758 knowledge workers to do realistic consulting tasks with or without GPT-4. On 18 tasks inside AI's abilities, people using AI completed 12.2% more tasks, finished 25.1% faster, and delivered higher-quality work. But on one complex task chosen to be outside those abilities, people using AI were 19% less likely to produce a correct solution.

Where AI was strong it lifted people up. Where it was weak, people using it were 19% less likely to get the right answer.

This study did not measure what each person knew. But to know when an AI answer is wrong, you need something in your head to check it against. Asking the AI again to check can just mean getting the same answer twice.

So what should you memorize, and what can you look up?

Hand off what only needs to be correct when you find it, and keep in your head what you think with. The research on remembering less of what you save to a device is in our note on the Google effect. Offloading isn't a loss in itself. What you offload is what matters.

The table below is not a rule from the research. It is the split we use, based on the studies above.

KindExamplesWhere it goes
Details that change or you need onceFunction names, versions, dates, lookup tablesLook it up, save it
The core terms and ideas of a fieldOpportunity cost in economics, standard deviationYour head (recall with it closed)
How ideas connectWhy, how it differs, when it failsYour head (explain it yourself)

The test is one question: without this, could you read an AI answer and judge whether it is right? If not, it belongs in your head.

What can you try today?

Pick one thing you asked AI today, close the window, and recall it. You need a sheet of paper, not Aloom.

  1. Pick one question you asked AI today and close the window with the answer.
  2. Write down the core of the answer in three sentences, from memory. Note any terms you got stuck on.
  3. Open the window again, compare, and fix what you missed.
  4. Tomorrow and three days later, recall just the terms you got stuck on, with the answer closed. Why the gaps help is in our note on moving information into long-term memory.

The next time you ask AI something, write your own one-line answer first and then ask. You will see right away where what you know and what the AI says part ways.

In Aloom

The conclusion of this note is: don't memorize everything; choose the core you think with and keep it in your head. In Aloom, that choosing is where things start. You ask AI, keep only the answers worth keeping, and they connect to what you already know. Each day you recall 1, 3 or 5 of them, whichever you choose, as a question before you see the answer. The more you recall a node, the darker its ink. Answers you only need to look up don't have to be kept. See how Aloom works

FAQ

Is memorization still important in the age of Google and AI?

Yes, but what you memorize changes. Look up details that change, and keep the core ideas and terms of your field in your head so you can understand and judge what AI tells you.

Does memorizing more make you learn faster?

The evidence is mixed. In a meta-analysis, the average correlation between prior knowledge and learning gains was −.059, close to zero, and results varied widely between studies. In one set of experiments, students who knew more about a domain learned more new facts in it; in another, knowledge created by training did not change how much people learned. What is clear is that people who know more before studying also know more after.

Is memorization bad for learning?

It depends on how you memorize. Rereading or copying something over and over doesn't last well; recalling it with the page closed and spacing your reviews does. Memorizing and understanding help each other.

References