Working Memory and Sudoku

Working memory is the mental scratchpad you use to hold information while you do something else with it. Phone numbers people read to you. The middle of a long sentence while you wait for the end. The five cells you almost forgot to check in a sudoku scan. It is small, busy and surprisingly fragile.

Sudoku makes demands on working memory the way few daily activities do. Whether that makes you measurably better at working memory in general is the question, and the answer is smaller than you would expect.

What working memory actually does

Cognitive scientists describe working memory as having three or four buckets: a visual one, a verbal one, an executive controller that decides where to look, and (in some models) an "episodic buffer" that holds short-term context. The buckets are small. Most adults can hold four to seven items at a time, depending on the bucket and the items.

Sudoku leans hard on the visual bucket. You hold candidate sets in your head while you scan the board. You hold "which digit am I scanning" while you walk the boxes. You hold a half-finished chain while you look for its other end. Without working memory, you would lose track every few seconds.

What happens when you practise

With practice, much of the load shifts from working memory to procedural memory. The first time you find a hidden single, you consciously hold the digit and the box. The hundredth time, your eye does it without effort. The information is no longer in working memory; it is part of a skill that runs underneath attention.

"You do not get a bigger working memory. You learn to load it more efficiently, then to skip loading it at all."

Why the gains are narrow

Brain training studies consistently find a pattern called "near transfer" and "far transfer". Near transfer is improvement on tasks that resemble what you trained. Far transfer is improvement on unrelated tasks. Sudoku produces clear near transfer (other sudoku-like puzzles, similar scanning tasks) and very little far transfer.

The reason is mechanical. You did not strengthen working memory as a generic resource. You learned a specific set of patterns that no longer demand working memory. The patterns transfer to cousin tasks. They do not transfer to tasks that do not share the underlying structure.

What the controlled studies say

Randomised trials of brain training games over the past two decades have generally found small near-transfer gains and weak, inconsistent far-transfer gains. The largest sceptical meta-analysis (Simons and colleagues, 2016) argued that most commercial brain-training claims were not supported. The US Federal Trade Commission settled with Lumosity in 2016 for unsupported claims; the company paid two million dollars.

Sudoku has the same caveat. It improves you at sudoku. The improvement does not generalise as much as the marketing suggests.

The thing worth saying

If you play sudoku because you enjoy the focused half hour, you are getting exactly what the activity delivers. If you play it hoping to enlarge your working memory, you will probably get a little of that, but not enough to notice in daily life. The first reason is reliable. The second is not.

For the broader picture of what the science actually says, see sudoku and the brain. For the flow side of the same coin, see flow state and puzzles. Or play an medium puzzle and notice what you are holding in your head as you scan.