On cognition, memory and learning

Mechanical sympathy is understanding a system’s underlying structure and properties, and using it in ways that align with those realities to get better outcomes.

The same idea applies to your brain.

Your brain is not a file system. It does not save memories in one place and retrieve them unchanged later. A memory is rebuilt from patterns of activity across the brain, and learning changes the physical connections that make those patterns easier to rebuild.

This is useful to understand because most learning advice makes more sense once you understand the hardware.

The hardware

This part is a little dense, but hang in there. Once the pieces are in place, the rest of the article starts to make a lot more sense.

The brain is made of neurons, which are cells that pass signals to each other. A neuron has a cell body in the middle, thin extensions that receive signals, and one longer extension that sends signals outward. The receiving extensions are called dendrites. They look branch-like because they split into many small paths, giving the neuron many places to receive input from other neurons. The cell body adds those signals up. If enough input arrives, the neuron fires, meaning it sends a brief electrical pulse down its axon. The longer sending extension is called an axon, and it carries that outgoing signal away from the neuron.

The tiny gap where one neuron influences another is called a synapse. A signal reaches the end of one neuron, chemicals cross the gap, and the next neuron becomes more or less likely to fire.

One important detail is that these connections are weighted. A neuron is not taking a simple vote from all its inputs. Some synapses matter more than others. Some signals barely count. Some become important only in combination with many others. Learning changes these weights. This is why a concept can be vague, flexible, and still reliable: it is supported by many weighted signals rather than one exact rule.

Synapses are where a lot of learning happens.

When one neuron repeatedly helps activate another neuron, the connection between them can get stronger. The receiving neuron can become more responsive. More receptors can be added. Synapses can become easier to trigger. In some cases, structure changes too.

This is the physical substrate of learning. Not the whole story, but the important part for mechanical sympathy.

A single neuron is simple. The interesting thing is what happens at scale. Billions of neurons, each responding to a narrow slice of activity, form patterns large enough to recognize faces, drawings, words, routes, proofs, songs, and memories. The concept is not sitting in one place. It emerges from the structure of the connections.

Your cortex is the wrinkly outer layer of the brain. It handles a lot of perception, movement, language, planning, and long-term knowledge. Different parts specialize in different things, but they are heavily connected.

Your hippocampus sits in the medial temporal lobe, tucked inward on each side of the brain near the center. It is especially important for forming new episodic memories. It helps bind together the pieces of an experience, including where you were, what you saw, what you felt, what was happening, and what it meant.

Your prefrontal cortex helps with attention, planning, inhibition, and holding things in mind. When you are thinking through a problem, it helps keep the relevant pieces active long enough to use them.

Your thalamus routes sensory information. Your amygdala helps tag emotionally important things. Your basal ganglia are important for action, habits, and reinforcement. None of these regions work alone, but the distinction matters because different learning problems use different machinery.

This is why "the brain" is too vague. Learning a face, learning a route, learning a piano movement, learning a proof, and learning that a stove is hot are not identical operations. They overlap, but they lean on different systems.

What a thought is

A thought is coordinated activity across neurons.

When you think of a word, picture a room, remember a friend, or decide what to say next, a distributed pattern of neurons becomes active. Some of that pattern is sensory. Some is language. Some is emotional. Some is motor. Some is executive control.

The thought is not stored in one place. It is the active pattern.

This is why a thought can pull in related thoughts. Activating one pattern makes nearby or associated patterns easier to activate. Think of a song and you may remember a person. Think of the person and you may remember a city. Think of the city and you may remember a restaurant.

Attention is what happens when the brain biases some patterns over others.

Your prefrontal cortex helps keep the important pattern active and suppress irrelevant ones. This is working memory. It is small, expensive, and easy to disrupt. If you keep switching tabs, checking your phone, or half-reading, you are forcing the system to constantly reload context.

What a memory is

A memory is not a perfect recording.

It is a pattern that can be reconstructed.

When something happens, different parts of the brain represent different pieces of the event. Visual cortex handles visual detail. Auditory cortex handles sound. Language areas handle words. Emotional systems handle salience. The hippocampus helps bind those pieces together so the event can be reactivated later.

Early on, the hippocampus is doing a lot of the indexing. It is not storing the whole thing like a video file. It is helping coordinate the pattern so the pieces can be brought back together.

Over time, especially with sleep and repeated recall, the cortex carries more of the structure itself. The memory becomes less dependent on the hippocampus. This is consolidation.

When you remember something, your brain partially reactivates the old pattern. It reconstructs the event from the pieces that are available now. This is why memory is useful but not perfectly reliable. Each recall is influenced by context, emotion, later knowledge, and what you are trying to retrieve.

This also explains why you can suddenly remember something you did not know you remembered. The memory was not gone in the simple sense. The retrieval path was weak or unavailable. A smell, place, question, or adjacent idea can activate enough of the network to make the rest come back.

This is your memory, your sense of self, every fact you know, and every thought you have ever had. Not as neat objects in storage, but as circuits that can be activated again.

What learning is

Learning is repeatedly activating the circuits you want to strengthen.

That is the whole game.

Practice, recall, testing, sleep, spacing, attention, and teaching all matter because they change which circuits activate, how strongly they activate, and how likely they are to activate again later.

If you read something once and never use it, the circuit is weak. It may feel familiar later, but familiarity is cheap. It is not the same as being able to recall the idea, explain it, or use it.

If you recall it from memory, connect it to other ideas, apply it to a problem, sleep, and then come back to it again, the circuit gets stronger.

This is why rereading is usually overrated. It feels good because the page becomes familiar. But the real test is whether you can recreate the idea without the page in front of you.

Sleep matters

Sleep is not just recovery. Sleep is part of learning.

The hippocampus replays recent experience. The cortex slowly absorbs patterns. Some connections stabilize. Some noise gets cleaned up.

This is why sleep after hard learning works so well. You are not just resting. The system is doing background work.

This also explains why cramming has such bad shape. You can push a lot into short-term activation, but the brain still needs time to consolidate it. At some point, more input is not the bottleneck. The bottleneck is stabilization.

After a hard session, stop. Take a walk. Nap. Sleep. Let the machinery do the part you cannot consciously force.

Learning on purpose

The practical move is not to collect tricks. It is to make the right circuits fire, make them fire in useful contexts, and give the brain enough time to stabilize them.

Attention comes first because unattended input is weak input. If you are reading while checking your phone, the brain is not just getting less information. It is repeatedly practicing context switching. The prefrontal cortex has to reload the task, rebuild the relevant pattern, and suppress whatever just interrupted it. That is expensive.

Good learning starts by protecting the active pattern. One thing open. One problem in front of you. Enough time for the circuit to stay warm.

Chunking matters because working memory is small. You cannot hold every detail active at once, so the brain has to compress. A beginner sees scattered facts. An expert sees larger units. This is why a chess master sees structure where a beginner sees pieces, and why a good engineer sees a system where a beginner sees files and functions.

The goal is not to memorize less. The goal is to bind smaller facts into larger shapes that can be moved around in your head.

Recall matters because it makes the circuit do the work. Rereading gives you recognition, which is usually too easy. Recognition says you have seen the idea before. Recall says you can rebuild it without the page.

After reading a section, close it and reconstruct the point. Write the argument from memory. Explain the mechanism without looking. Then reopen the source and correct what you missed. That correction is not a failure. It is the learning signal.

Spacing matters because circuits need repeated activation separated by time. If you hit the same idea again after a delay, the brain has to retrieve and rebuild it. That strengthens the retrieval path. Cramming can create short-term familiarity, but it often fails at making the path durable.

Interleaving matters because real use requires selection. If you do ten identical problems in a row, part of the work has already been done for you. You know what kind of problem it is. Mixing related problem types forces the brain to choose the method, not just execute it.

Prediction matters because the brain is constantly comparing expectation to input. Before checking an answer, commit to one. Before reading the next paragraph, guess where the argument is going. When you are wrong, the difference between your prediction and reality becomes easier to notice, and easier to remember.

Reading better

Reading is not one skill. It is several different operations that often get collapsed into one word.

Sometimes you are surveying. You want the shape of the piece, the claims, the structure, and the parts that deserve attention. Skim the headings, diagrams, opening, conclusion, and topic sentences. This is not fake reading. It builds a map so the detailed pass has somewhere to attach.

Sometimes you are parsing. You need to slow down, follow the exact argument, and keep definitions stable. This is where subvocalizing can help. Hearing the sentence in your head is slower, but sometimes the sentence is doing real work and speed is the wrong goal.

Sometimes you are extracting. You are looking for the mechanism, the example, the claim, or the contradiction. Underline only what you might use later. If everything is underlined, nothing is.

Sometimes you are integrating. This is the part people skip. Stop after a section and rebuild the idea in your own words. Ask what changed in your model, what it connects to, what would be different if it were false, and where you could use it.

A pen or finger can be useful, but not because it is a magic reading hack. It gives your eyes a stable path and reduces regressions, which are the little backward jumps your eyes make when attention drifts or the sentence gets hard. For easy material, this can keep you moving. For dense material, it can reveal where you actually need to slow down.

Reading faster is only good when comprehension survives. The point is control. You should be able to skim when mapping, move quickly when the material is easy, slow down when the argument is dense, and stop entirely when you need to reconstruct.

If something is important, write about it without the source open. Writing forces compression, sequencing, and recall at the same time. It is one of the fastest ways to find the difference between recognizing an idea and actually having it available.

The practical version

Protect attention so the right pattern can stay active.

Read first for structure, then for detail.

Use recall to solidify newly formed paths.

Compress facts into chunks you can reason with.

Space repetitions so retrieval has to work.

Mix related problems so selection is part of practice.

Predict before checking so errors become informative.

Stop after hard learning and let sleep do its part.

The brain is physical. Learning is physical. Treat it like a system with properties and it starts behaving less mysteriously.