Ann Graybiel, a neuroscientist at MIT’s McGovern Institute for Brain Research, spent decades wiring electrodes into the basal ganglia of rats running T-shaped mazes, and what she saw on the readouts changed how neuroscientists talk about habit. When a rat first learned the maze, its striatal neurons fired in a chaotic scatter across the whole run. After enough repetitions, the firing collapsed inward — a sharp burst at the start, near silence in the middle, another burst at the goal. The brain had bracketed the entire sequence into a single packaged unit. Graybiel’s lab called this bracketing chunking, and it is why you can drive the familiar route home from work and arrive with no memory of the drive itself.
The word borrows from the work of psychologist George Miller, who noticed that working memory holds a limited number of items unless those items are grouped — a phone number becomes three chunks, a chess position becomes a single pattern. Graybiel’s insight was that motor routines chunk the same way, and they do it in a specific piece of tissue buried under the cortex.
The structure doing the packaging
The basal ganglia sit deep in the brain, a cluster of nuclei wrapped around the thalamus like a fist around a stone. The striatum is its main input hub, and it is the part Graybiel’s electrodes were reading. For a long time the structure was associated mostly with movement disorders — Parkinson’s disease kills dopamine neurons that feed into it, Huntington’s disease chews through its cells — but the motor framing turned out to be too narrow. The basal ganglia are a general-purpose sequence compressor. They take a string of actions the cortex is laboriously running one at a time and glue them into a routine that can be triggered as a whole.
This is the neural substrate of procedural memory — the kind of memory that lets you tie your shoes, touch-type, ride a bike, or play a scale on the piano without thinking about any of the intermediate steps. It runs on different circuitry from the memory that lets you recall what you had for breakfast, which is why patients with hippocampal damage can still learn new motor skills even when they cannot remember the lessons.
What the rats showed
Graybiel’s mazes were simple. A rat started at the base of a T, ran up the stem, and at a click either turned left for chocolate milk or right for sugar water. Early in training, the rat’s striatal neurons chattered continuously — every whisker twitch, every hesitation, every small correction produced firing. After many trials, the same rat ran the same maze cleanly, and the neural activity had reshaped itself into what Graybiel described as bookends. Fire at the start. Silence. Fire at the end.
The silence in the middle was the tell. It was not that the brain had stopped doing anything — the rat was still running, turning, drinking — it was that the basal ganglia had stopped narrating each step. The sequence had been compressed into a single retrievable unit, initiated by one cue and terminated by another. When the striatum was disrupted, the chunk fell apart. The rat could still run, but it treated every step as new.
Why you forget the drive home
The everyday version of this is the commute. First week at a new job, the drive is exhausting — every turn is a decision, every unfamiliar exit sign gets processed, working memory is fully engaged. Six months later, you pull into the driveway and cannot remember whether the light on Elm Street was red or green. You were not asleep. Your basal ganglia had loaded the whole route as one chunk, and your cortex was free to think about the meeting you just left.
The cost of chunking is exactly what you would expect from a compressed file. The routine runs faster and cheaper, but you lose access to the intermediate frames. This is why habits are hard to interrupt mid-sequence — once the start cue fires, the rest of the packaged unit tends to run to completion. Smokers reaching for a lighter after coffee, nail-biters mid-thought, the way you check your phone before you have decided to check your phone. The chunk starts, and the cortex is a spectator.
The chess player and the pianist
Chunking is also why expertise looks like magic from the outside. When a chess grandmaster glances at a board mid-game and reconstructs it after a five-second look, they are not memorising 32 pieces — they are recognising perhaps five or six chunks, each one a familiar pattern of piece relationships they have seen thousands of times. Randomise the pieces into positions that could never arise in a real game, and grandmaster recall collapses to roughly the same level as a beginner. The pieces are the same. The chunks are gone.
Concert pianists show the same signature. When a concert pianist plays a Chopin ballade, they are not retrieving individual finger movements — they are triggering compressed motor sequences that were built through thousands of hours of slow practice, each chunk running under basal ganglia control while the cortex is free to shape phrasing, dynamics, the emotional arc. Stop mid-phrase and the pianist often cannot describe which finger comes next without mentally restarting the passage from an earlier landmark. The chunk does not have a middle you can query.
Why practice has to be slow
The building of a chunk is a different process from the running of one. Early in learning, the prefrontal cortex is doing most of the work, deliberately sequencing actions, correcting errors, holding the whole plan in working memory. As the sequence repeats, activity gradually shifts inward and downward, from cortex to striatum. Graybiel’s group tracked this migration in the rat brains — as chunks consolidated, cortical involvement decreased and striatal bracketing sharpened.
This is why coaches and music teachers insist on slow, correct practice. A chunk built at speed with sloppy inputs is a chunk that runs sloppily forever. Once packaged, it is stubbornly resistant to editing. Building mastery of a complex skill requires time for the basal ganglia to build a library of clean, deep chunks — and much of that time has to be spent doing the work slowly enough for the cortex to still be watching.
When chunking breaks
The clinical picture makes the mechanism visible in reverse. Patients with Parkinson’s disease, whose dopamine input to the striatum is progressively lost, describe the failure of automaticity in exactly the terms Graybiel’s model predicts. Walking, which for most people is a chunk triggered by the intention to move, becomes a sequence of conscious steps. Buttoning a shirt requires cortical attention. Handwriting, once a fluid packaged skill, has to be built up letter by letter and often shrinks into micrographia as the chunk falls apart.
Huntington’s disease attacks the striatum directly, and its patients show a related but distinct pattern — habits that were learned before onset persist longer than the ability to build new ones, because the packaging machinery is damaged even when older packages remain intact for a while.
Chunking under stress
The system has a well-documented failure mode under acute stress. When stress intensifies, working memory narrows and the brain leans harder on pre-built chunks — which is useful if the chunks are good ones and disastrous if they are not. Emergency-room physicians train drill sequences precisely so that under stress the right chunk fires. Airline pilots run checklists for the same reason. The cognitive-chunking research that has migrated into workplace-performance writing is downstream of the same basal ganglia biology — group the task into packaged units the brain can trigger without deliberation, and stress has less surface to attack.
The flip side is that stress also strengthens whatever chunk gets rehearsed under it. Panic scrolling, stress eating, compulsive email checking — these are chunks that were built in exactly the emotional state that most efficiently deepens them. The routine you reach for when overwhelmed is the routine that becomes hardest to unbuild.
The organoid and the coffee
Newer work is starting to probe how these deep structures develop in the first place. In 2025, a team at Johns Hopkins grew what they called a whole-brain organoid that included multiple interconnected regions, opening a path to studying striatal-cortical circuits in tissue rather than in living rats. And a 2026 43-year cohort study from Mass General Brigham found associations between long-term coffee intake and preserved cognitive function into old age — a finding that matters partly because caffeine acts on the same dopamine pathways that keep the basal ganglia’s chunking machinery humming.
Related reading on the everyday texture of automatic behaviour appears in work on highly perceptive people and in a recent piece on the people who keep showing up when nothing is working, whose persistence often lives in the same procedural circuits.
The bracket you cannot see
Graybiel is now in her mid-eighties and still runs a lab. The bracketed firing pattern she first described in the 1990s has been replicated in mice, monkeys, and, through fMRI, in humans playing sequence-learning games. The basic shape holds. A skill you can do without thinking is a skill your striatum has bracketed. The start cue fires. The end cue fires. In between, the routine runs itself, and you get to think about something else — the argument you had yesterday, the meeting tomorrow, the fact that you have somehow arrived home with no memory of the last twelve minutes of driving.
The road is still there. The turns still happen. Somewhere behind your forehead, a small structure the size of a walnut just ran the whole thing as a single word.
