Cognitive Training

What it is

Near transfer is what the trials measure. Far transfer is what the marketing promises. Near transfer means remembering more digits after practising a digit task. Far transfer means remembering where you left your keys, following a conversation in a noisy room, holding a plan together.

The best-designed thing in the field is the ACTIVE trial, a large randomised study in older adults run by Karlene Ball and colleagues, with benefits on the trained ability still detectable ten years later and a lower rate of at-fault crashes in the speed-of-processing arm at that follow-up. That is a real result and it should not be waved away.

The most instructive thing in the field is Adrian Owen's 2010 study, run with a BBC television series. Of more than fifty thousand registrants, 11,430 met the inclusion criteria, split across two training arms and a control arm over six weeks. People got better at the tasks they trained on, and the gains did not transfer to untrained tasks, even closely related ones. Eleven thousand completers is an order of magnitude beyond anything else in the field, and a clean null at that scale is the strongest single result this page carries.

In effect

Both sides have been stated publicly and by name, which makes this easier to report than most disputes.

In October 2014 the Stanford Center on Longevity, with the Max Planck Institute for Human Development, published an open letter signed by more than seventy psychologists and neuroscientists saying the literature does not support claims that brain games improve general cognition in everyday life or prevent decline. In December of the same year a counter-letter signed by 133 scientists and practitioners argued that the literature is full of demonstrated benefits. Both signatory groups contained people with commercial ties to training products, and that should be said in both directions.

The evidence since has moved further towards the sceptical letter rather than less far. Daniel Simons and colleagues reviewed the field exhaustively in 2016 and found the evidence of real-world benefit beyond the trained exercises inadequate, with much of the literature lacking proper controls and failing to account for expectancy. Giovanni Sala and Fernand Gobet ran a second-order meta-analysis in 2019 and found that once placebo effects and publication bias are accounted for, the far-transfer effect and its variance both come out at zero, across populations and programme types; in 2023 the same pair described the area as a field in search of a phenomenon. Cochrane reviews of computerised training in healthy older and midlife adults found little evidence that sustained training maintains cognition, on a low-quality evidence base. Walter Boot and colleagues made the methodological point underneath all of it, that an active control group is not enough to rule out a placebo effect, because participants who know they are being trained expect to improve.

The most useful fact for a general reader is not a meta-analysis. On 5 January 2016 the United States Federal Trade Commission announced that Lumos Labs, maker of Lumosity, would pay two million dollars to settle charges of deceptive advertising. The Commission's stated case was that the company had preyed on consumer fears about age-related cognitive decline, advertising that its games could stave off memory loss, dementia and Alzheimer's disease, without the science to back the advertisements. The order also required the company to notify subscribers and to give them an easy way to cancel auto-renewal. The near and far transfer gap is a scientific distinction; the settlement is the moment a regulator priced it.

What it does not say

It does not say brain training does not work. That is itself an unevidenced blanket claim. Some programmes produce durable near transfer, which is what learning looks like, and adaptive difficulty is a plausible and well-motivated mechanism.

It does not say the evidence is mixed. The evidence is not evenly mixed. It is good for the narrow claim and absent for the broad one, and softening that into mixed misdescribes it.

It does not settle the working memory claim on its own terms either. The adjacent small-study literature has average statistical power below fifty per cent, with typical power in the low teens for cognitive studies and lower still for working memory studies specifically. A literature powered that low produces noise with a significance value attached.

It does not follow that every book in the genre overclaims. Adam Gazzaley and Larry Rosen's chapter on the subject cites its own critics, placing Owen and Boot as consecutive endnotes at the exact point where the case for training is being made, along with a citation about the commercial incentives of the industry. About nine in ten of that chapter's notes are peer-reviewed journal articles, and the authors grade their own category a signal rather than a prescriptive intervention on their own evidence ladder, giving the top rung to physical exercise, in which they have no stake. What that chapter does not do is disclose that its centrepiece is the authors' own game.

This is material about cognition and ageing that readers may act on, and nothing here is a recommendation or a guide to any product.

---