Brains Improve by Deleting Stuff

My favorite myth about intelligence is that it works like a junk drawer.

Just keep adding things. More facts. More worksheets. More apps. More flashcards. More parameters, if you are an AI person wearing the traditional fleece vest of destiny.

Surely a smarter system is a fuller system.

The brain, inconvenient little contrarian that it is, disagrees.

A developing brain does not become more capable by simply piling up connections forever. It builds, tests, strengthens, reroutes, and — this is the part that feels rude — gets rid of what it does not need. Intelligence is not just accumulation. It is editing.

This is where child development and AI research have started whispering the same mildly embarrassing secret: bigness is not the same thing as usefulness.

The brain is not trying to keep everything

Parents often hear about childhood as a time of explosive growth, which is true enough. Young brains are gloriously busy. They form connections, sample environments, learn sounds, faces, routines, rules, and the dark legal loophole by which “bedtime” apparently becomes negotiable if one asks for water with enough sincerity.

But development is not just growth. It is specialization.

Charles Nelson’s review of early intervention through developmental neuroscience makes this point beautifully: different brain systems have different windows of heightened plasticity, and support works best when it fits the system’s developmental timing (Nelson, 2024). Sensory systems, language systems, cognitive control, and social-emotional circuits do not all mature on the same schedule. The brain is not a loaf of bread rising evenly in the oven. It is more like a city under construction, where the roads, power grid, schools, and parks all come online at different moments.

This matters because experience helps decide which pathways become useful highways and which ones become abandoned side streets.

That is not tragic. That is the point.

If every possible connection stayed equally available, the system would be noisy, expensive, and inefficient. A brain that never pruned would not be infinitely creative. It would be the group chat with notifications permanently on.

The “peak brain” story is too tidy

Here is another popular story worth putting gently in a recycling bin: your brain rises to a glorious summit in young adulthood and then begins a slow cinematic decline, possibly with sad violin music.

A lifespan brain-imaging study by Mousley and colleagues complicates that cartoon. The researchers found that structural brain organization shifts through distinct developmental epochs rather than following one smooth “up, then down” curve (Mousley et al., 2025). In plain English: the brain keeps reorganizing across life. Different phases have different wiring priorities.

This is a useful correction for parents, because we often talk as if development is a race to lock in as much as possible before the gate slams shut.

Yes, timing matters. Some windows are especially sensitive. Early support can be powerful. But the brain is not done after preschool, and adolescence is not just a hormonal weather event with sneakers. It is a major remodeling period.

That remodeling can look messy from the outside. Your kid may seem brilliant on Tuesday, feral on Wednesday, and somehow both on Thursday. This is not evidence that their brain has misplaced the instruction manual. It is evidence that flexible systems often pass through unstable-looking phases while they reorganize.

AI researchers know this too, though they use less laundry-based language.

Big models also contain dead weight

The default AI fairy tale says: scale solves everything. Bigger models, more data, more compute. Pour in the internet, stir with GPUs, receive intelligence.

This has been productive. Annoyingly productive, even. I am not pretending scale is useless. Scale has paid rent.

But scale is not the whole story.

A Nature study from the DiCarlo Lab showed that deep neural networks used to model the visual cortex can be compressed substantially while retaining much of their ability to predict neural responses, especially in earlier visual areas (DiCarlo Lab, MIT, 2025). Translation: a lot of the machinery in these models may not be doing the brain-relevant work we care about.

That is deliciously awkward for the “just make it bigger” crowd.

The compressed models were not interesting because they were tiny for the sake of being tiny. They were interesting because compression can reveal structure. When you remove what is redundant and performance holds, you learn something about what mattered.

This is the AI version of developmental pruning. Not identical — analogies are not magical portal guns — but close enough to be useful.

Brains prune connections. Models can be compressed. Both cases suggest that intelligent systems may improve not only by adding capacity, but by finding the right shape.

What parents can actually do with this

Please do not hear “pruning” and panic-buy a brain-training subscription. That would be very on-brand for the internet and very unnecessary.

A better takeaway is simpler: children need rich, varied, emotionally safe experience — and then time for their brains to sort it.

Try this:

• Offer variety, not frenzy. Different kinds of play, movement, language, music, stories, and social interaction give the brain useful material. A packed schedule is not automatically richer than a curious afternoon.

• Repeat without drilling everything flat. Brains strengthen useful pathways through practice. But practice can be playful. Reading the same book again counts. Building the same block tower with a new twist counts. Explaining why the sock puppet is “wrong” absolutely counts.

• Respect developmental timing. If a child is not ready for a skill, more pressure may produce more frustration, not more learning. If you are worried about speech, motor skills, attention, reading, or social development, talk with a pediatrician or qualified developmental specialist rather than trying to troubleshoot it alone.

• Let boredom do some work. Downtime is not wasted time. A brain cannot reorganize if every quiet moment is immediately colonized by content.

• Do not confuse efficiency with absence. When kids grow out of certain sensitivities or interests, that can be refinement, not loss. Development often looks like narrowing because the brain is becoming better tuned to the world it actually inhabits.

The boring middle wins again

So, no: your child’s brain is not a hard drive you need to fill before the warranty expires.

And no: AI systems are not automatically smarter because they are larger, louder, and more expensive to run than a small nation’s toaster fleet.

The better story is more interesting. Minds develop through cycles of growth and constraint, openness and specialization, plasticity and pruning. They learn what to keep. They learn what to ignore. They become powerful partly by becoming selective.

Intelligence is not the junk drawer.

It is the cleanup afterward.