What a Memory Actually Is

I forgot a friend's birthday last month. Not just the date — I forgot I had ever known it. We were standing in a coffee shop and she said "so we're still on for the 14th?" and I said yes, instantly, like I had remembered. I had not remembered. I had no idea what the 14th was.

The thing is, my brain did have a memory of her birthday somewhere. It just wasn't the one I could reach. And that small distinction — memory stored but not retrievable — turns out to be the whole ballgame.

Researchers call the physical trace of a memory an engram. It sounds like sci-fi, but it's a real thing: a small, sparse population of neurons that physically changes when you have an experience, and physically fires when you recall it. There are neuroscience papers from the last few years that locate these engrams, activate them with light, suppress them, and watch the memory come and go. The findings are weird, and they line up with something I think we already know intuitively about taking notes.

An engram is small on purpose

When you learn something, only a tiny fraction of the neurons in the relevant brain region get recruited. In the amygdala, that fraction is 10 to 20 percent. In the dentate gyrus of the hippocampus, it's 2 to 6 percent. The number is conserved — the same proportion wins every time, regardless of how strong the stimulus is or what kind of memory it is.

A more vivid memory does not get a bigger engram. A scarier one does not either. The brain picks its small handful of neurons and that is that. It turns out there is an actual mechanism: the most excitable neurons compete to be part of the engram, and once they're in, they suppress the neighbors via local inhibitory circuits. The pool of cells that gets the memory is a winner-takes-all competition, run every time.

Memories are linked by sharing neurons

Here is the part that surprised me. Two memories that are encoded close in time — within about six hours — share a lot of the same engram neurons, because the same excitable cells won both rounds. And if they share neurons, they become functionally linked. Recall one, and the other fires too. That is the physical substrate of association: not a label saying "these two go together," but actual overlapping hardware.

If the two memories are encoded more than 24 hours apart, the pools barely overlap, and you can extinguish one without touching the other. The link is structural. It depends on which neurons won the race on which day.

The same trick works in reverse. Two memories that originally don't overlap can be linked later by repeated co-retrieval — think of studying flashcards and shuffling the deck. Each time you bring the two engrams back to life together, the brain grows a small shared pool of neurons between them, and from then on they fire as a pair. That overlap is what the link actually is.

So what — the brain is doing this anyway, why does it matter for notes?

Two things I think most people get wrong about note-taking.

The first is the idea that more notes is better. A 30-page document on a single topic is not a memory aid. It is a small library. The brain is not running a small library. The brain is running a sparse, distributed system where the unit of storage is one engram, in one small population of neurons, encoding one experience. A long note asks the brain to hold an entire small library in one engram. It can't. It won't.

The second is the idea that the link between two notes is something you write down. "See also: other note." That's a label. The brain doesn't store links as labels. The brain stores links as overlapping hardware. And the only way hardware overlaps is if the two notes came from close-in-time experiences — close in your own attention, close in your own retrieval — that shared enough context to be encoded by overlapping populations of neurons.

What this changes about how I take notes

I now think about each note as one engram. One idea, on its own card, captured close to the moment of the experience. Not when I'm "organizing." Not at the end of the week, in a cleanup pass. At the moment, when the experience is fresh enough that the right neurons are still excitable, when whatever I want to remember is most likely to land in the same pool as the thing it actually relates to.

If I want two notes to be linked, I retrieve them together. Not because I'm building a system. Because the brain only grows the structural link when the two are reactivated at the same time. That's the part no tag can fake.

And if I want a note to stay retrievable, I make it small enough to fire as one unit. One card, one idea. The engram is small on purpose. My notes are small on purpose. It is not a stylistic preference. It is what the hardware does.

A practical version of this

Capture the idea in the moment. Keep each note to one card. Don't merge two ideas into a longer note just because they're both about the same project — they came from different experiences, and the brain allocated them to different engrams anyway. When you want to connect two notes, pull them both up and read them together. Do that a few times over a few days and the link becomes physical.

Index cards were the right idea 100 years ago and they're the right idea now, not because they're charming, but because they fit the unit of memory. One card. One idea. On its own, until you decide to bring it next to another.

What I took from this

My friend had a birthday on the 14th. I didn't forget the date, I forgot the engram was retrievable. The memory was in there. I just hadn't pulled it back out recently enough for the right neurons to fire, and so the link from "she mentioned it" to "the 14th" was gone. That happens. It is not a moral failing, it is hardware.

What I can do is make sure the things I want to keep retrievable get reactivated often. Not by re-reading them. By pulling them out and using them. Index cards let me do that without it looking like studying. It's just shuffling the deck.

If a memory matters, give it its own card and bring it back out. That is the whole game.

Sources: Robins (2023) WIRES Cognitive Science; Roy et al. (2022) Nature Communications; Josselyn & Tonegawa (2020) Science; Chen et al. (2020) Neurobiology of Learning and Memory; Rao-Ruiz et al. (2019) Current Opinion in Neurobiology.