The narrow transfer problem

If you've been doing Sudoku for a while, you might catch yourself wondering whether the brain you bring to a Tuesday meeting is any sharper for it. The honest, slightly deflating answer is: probably not in any way you'd notice. The thing that gets better when you practise Sudoku is your Sudoku.

This is what cognitive psychologists call the narrow transfer problem, and it's one of the most consistently replicated findings in the field. Skills are stubbornly domain-specific, and the brain doesn't generalise the way the marketing for any given training app would prefer.

The broader research piece walks through the population-level evidence. This one zooms in on the specific question: what does cognitive psychology actually mean by transfer, why is it so narrow, and what's the small handful of cases where it isn't?

Why transfer is rare

When you learn a skill, what you're learning is a constellation of small components: pattern recognition for shapes that show up in this domain, working-memory routines that hold this domain's information in useful chunks, motor patterns specific to this domain's interface, decision rules for this domain's typical tradeoffs. None of those components generalises automatically. They're built for the domain you trained them in, and outside it they have nothing much to grip on.

This is why the chess grandmaster has world-class memory for chess positions and average memory for shopping lists, and why the violinist has extraordinary fine motor control in their fretting hand and ordinary fine motor control everywhere else. The skill is real, but it's anchored to a particular set of cues, and away from those cues there's nothing for it to do.

A 1993 review by Detterman and Sternberg titled Transfer on Trial surveyed decades of laboratory studies and concluded, to widespread psychologist agreement, that transfer is rare, effortful, and usually requires explicit teaching of the transfer target itself¹. The thirty years since haven't moved the picture much, and if anything the disappointing track record of "brain training" apps has tightened the consensus rather than loosened it.

Three studies that show what narrow looks like

A Dutch psychologist named Adriaan de Groot ran a famous experiment in 1946: he showed chess positions to grandmasters and to club-level players for five seconds, then asked them to reconstruct the positions on an empty board. Grandmasters were dramatically better than club players at this, until the experimenters started showing them random arrangements of pieces that didn't correspond to a real game. On random positions the grandmasters did about as well as the club players, and not much better than someone who'd never played at all². The grandmaster's memory wasn't a general "I have a great memory" capacity; it was an enormous library of valid game patterns, and outside that library there was nothing to fall back on. The finding has been replicated in dozens of variants across the half-century since, and it holds.

A famous 2000 study by Eleanor Maguire scanned the brains of London taxi drivers, who train for years to memorise the city's twenty-five thousand streets and the rules-of-thumb for navigating between them. The drivers had larger posterior hippocampi than non-drivers, a real structural change from a real intensive training³. They were not, however, better than the rest of us at general spatial reasoning, at route-planning in unfamiliar cities, or at memory for non-route information. The brain change was specific to the trained skill.

Mental abacus calculators in East Asia, who train from childhood to perform astonishing arithmetic by manipulating a virtual abacus in their heads, have similarly precise gains. They can multiply six-digit numbers in seconds, and they do it by visualising bead movements rather than by anything we'd call ordinary calculation⁴. They are not, however, generally better at mathematics — at geometry, at algebra, at applied problem-solving — than peers who didn't go through abacus training. The skill is the calculation, and the calculation is what they're better at.

When transfer does happen

The list of conditions under which transfer reliably occurs is short and precise. Transfer between two skills tends to happen when the skills share underlying cognitive components, for instance two tasks that both make heavy use of the same kind of working-memory chunking. It happens more when the trainee is explicitly taught to look for the transfer (a phenomenon cognitive psychologists call metacognitive instruction). And it happens when training varies across many surface forms of the same underlying problem, rather than drilling one form repeatedly.

None of these conditions is met by ordinary daily Sudoku-doing. The Sudoku you do at breakfast is not surrounded by metacognitive instruction. The components you're sharpening — visual scanning of a constrained grid, holding pencil-marked candidates in working memory, applying a small library of deduction rules — are tightly bound to the Sudoku context, and everyday life doesn't often look like a 9x9 grid with three constraints.

What this means for Sudoku

A few things follow, plainly. Doing Sudoku will sharpen your Sudoku-specific pattern recognition, your working memory for candidate digits, your discipline about scanning before placing, and your tolerance for mid-puzzle ambiguity. Those gains are real and measurable, and none of them is going to noticeably help you with the unrelated cognitive demands of your work, your relationships, or your life — not because Sudoku is a bad puzzle, but because that's just not how transfer works.

This isn't a reason to quit. It's a reason to stop expecting Sudoku to do something it can't, and start enjoying it for what it actually is: a daily fifteen minutes of focused, satisfying logic that's pleasant on its own terms.

If you've been doing Sudoku for a while and want to think about what does change with practice over years and decades, and what changes more generally about cognitive function in middle and later life, what changes in puzzling after 60 is the more direct take.

So: the narrow transfer problem is a problem only if you came in expecting Sudoku to be medicine. If you came in expecting it to be a puzzle that's interesting on its own terms — that gets satisfyingly harder as you climb its difficulty ladder, that fills fifteen minutes with focused logic instead of fifteen minutes of doomscrolling, that feels like a small good shape in the day — it isn't a problem at all. The puzzle is doing exactly what puzzles do, and that, on its own, is plenty.