You finish the crossword or sudoku, close Settimana Enigmistica and think you have given your brain a workout. You have certainly exercised some abilities. But how much of that exercise, which took time and concentration, will help you in real life?
Getting better at a game and becoming better at reasoning in general are different outcomes. Research distinguishes three stages.
The first: you get much better at what you do. By practising sudoku, you learn strategies, recognise patterns and eliminate possibilities with greater precision. With crosswords, you accumulate vocabulary, associations and familiarity with clues. You become faster and can tackle increasingly complex crosswords. This is genuine learning, but it is specific. In corporate language: vertical.
The second: perhaps you also get better at related tasks. A strategy you have acquired may help when another exercise shares rules or mental operations. This “near transfer” exists, but it is less reliable and generally more limited than improvement in the activity practised. How far learning travels also depends on how similar the tasks are. And it is fair to ask which real-life tasks resemble sudoku or Settimana Enigmistica: “in the middle of the Caspian Sea”: “sp”.
The third: your general intelligence improves. Here the evidence becomes scarce and much less convincing. In their major 2016 review, Daniel Simons and colleagues find ample evidence of improvement in trained tasks, less in similar tasks and little in distant tasks or everyday cognitive performance. Doing an exercise well does not make you better at understanding problems unrelated to that exercise.
An experiment published in Nature makes the distinction concrete. Adrian Owen and colleagues involved 11,430 participants in six weeks of cognitive training. People improved at the exercises they practised, without showing any transfer to untrained tests, even when these measured closely related abilities.
And chess? Examining chess, music and working-memory training, Giovanni Sala and Fernand Gobet conclude that far transfer occurs rarely, if not merely by chance.
Pattern recognition plays a central role in chess expertise. Gobet and Herbert Simon studied how experts remember meaningful configurations of pieces, organised into units of knowledge. But a sophisticated skill remains a specific skill, and its complexity is no guarantee that it will always, or more often, enable us to solve and understand problems involving other complex skills.
Cases involving old age, memory and dementia have also been studied. Imagine a hundred-year-old woman completing a crossword and, shortly afterwards, failing to remember a visit she has just received. This would not be a contradiction: knowing words and remembering recent events are not the same function. Semantic memory stores meanings and knowledge, while episodic memory concerns events we have experienced.
There is even a documented case. The patient H.M., who suffered very severe amnesia after brain surgery, continued to solve crosswords by drawing on his previous knowledge. Brian Skotko and colleagues studied his abilities in 2004: he showed that a capacity for solving puzzles can coexist with profound memory deficits.
This does not mean that crosswords have no relationship with cognitive decline. On these matters, almost nothing constitutes definitive proof. In 2011, Jagan Pillai and colleagues found, among 101 people who developed dementia, an association between doing crosswords and an onset of accelerated memory decline around two and a half years later. This was an observational study, and here too every expression of caution is necessary: the study did not demonstrate that crosswords had caused that delay in onset, nor that they prevented the disease.
Roger Staff and colleagues, meanwhile, followed 498 people for around fifteen years. Intellectual engagement was associated with better cognitive performance, but not with a slowing of age-related decline. Having a higher level and losing it more slowly are different things.
There are also favourable experimental findings. In 2022, D. P. Devanand and colleagues randomly assigned 107 people with mild cognitive impairment to crosswords or computerised cognitive games. After 78 weeks, crosswords proved superior on the main cognitive measure. There was, however, no group without training: the comparison does not demonstrate the prevention of Alzheimer’s, nor an increase in general intelligence in healthy people.
Sudoku, rebus puzzles and chess engage the mind. The unproven leap lies in attributing a general enhancement to that engagement. Above all, we become more capable at the activities we practise; some benefit may extend to related ones; much less has been demonstrated for distant ones. And skill at a game, on its own, does not tell us how everything else is functioning.
If learning is chiefly specific, it follows that what gains value for general intelligence is variety, not vertical specialisation: every book, every essay, every experience, every conversation with a professional in something we know nothing about, even a practical manual, can add knowledge and tools that give us more patterns to draw on and thus make us more intelligent.
These studies seem to suggest, then, that intelligence increases through many kinds of in-depth, vertical learning, far more than through just one that is extensively practised or taken to the extreme.