Learning is Memory

This might feel obvious, but many learners don't fully grasp the implications, and as a result, end up not actually learning much.

Want to get notified about new posts? Join the mailing list and follow on X/Twitter.

At the end of the day, learning is memory. Understanding amounts to memory that is well-connected and deeply ingrained.

The difference between “just memorizing” and “deeply understanding” isn’t the substrate of the representation, it’s the depth of the representation. Deep understanding consists of not only declarative facts, but also connections that link facts into related groups or “chunks” (think: concepts), connections that link smaller chunks into bigger chunks, and so on – as well as procedures for operating on chunks (think: skills), connections that chunk sub-procedures into meta-procedures, and so on.

This is all raw mechanical memory. It’s just storage and retrieval of information. The point of building superior representation is to build superior recall abilities, including broadening and fine-tuning the range of stimuli that activate the information. If someone is “just memorizing” as opposed to “deeply understanding,” it really means they haven’t stored enough information in memory.

“Learning is memory” might feel obvious, but many learners don’t fully grasp the implications. If you don’t realize that learning is memory, then you won’t realize that the most effective way to learn is to use memory-supporting training techniques.

It’s easy to get confused, thinking: “Truly understanding something is different from just memorizing it, so learning doesn’t require memory-focused techniques like retrieval practice, spaced review, and interleaving (mixed practice). Those are about memorization, not true understanding.” And if that’s what you think, then you’ll likely shirk the hard work required to build memory, use fun/easy but ineffective training techniques instead, and end up not actually learning much.

I used to think resistance to “learning is memory” was genuine confusion, but now I think it’s mostly laziness. If you accept that learning is memory, then you have to accept that maximizing learning requires memory-supporting training techniques. But those techniques are highly effortful and measurable, which make them unattractive to low-accountability / low-effort folks. The only way to reject the premise is to latch onto the idea that “understanding” is some supernatural thing that can’t arise from raw mechanical memory. Which is problematic because there’s decades of research showing how expertise arises from having lots of domain-specific information encoded into memory that is well-connected and deeply ingrained.

(A response to the most common genuine objection: Even learning to generate new ideas amounts to searching a space of possibilities, combining pieces of memory in ways that haven’t been combined before. Now you might say “aha, the skill of searching/combining is something other than memory,” but let me ask you: when a someone trains the skill of coming up with novel ideas, such as a grad student learning to come up with research ideas that contribute to the cutting edge of knowledge in the field, where is that skill stored for future use? In memory.)

Follow-Up Questions

But isn’t learning a process? It’s not stored anywhere – it’s an effect of understanding, and understanding isn’t memory. If you think learning (understanding) is memory then you’re forced to believe things like ChatGPT “learns.”

I would completely disagree that “learning isn’t stored anywhere,” on the grounds that learning is a physical change that happens inside the brain. It’s not like some spiritual thing that goes beyond the physical world. And processes can be stored in memory too (e.g., procedural memory).

Additionally, I don’t believe ChatGPT learns/understands nearly as well as a human expert, but I wouldn’t chalk this up to usage of memory. I would chalk it up to the brains of human experts having greater storage capacity, superior architecture for representation/inference, and access to way more information (including lots of information that is not publicly available).

Want to get notified about new posts? Join the mailing list and follow on X/Twitter.