Do Brain Training Games Work?

Researchers split improvement into two kinds. Almost every argument about brain training is really an argument about the second one.

Near transfer is getting better at things that resemble what you practised. Train on the grid above and you improve on the grid above. Move to a version built from letters and you probably carry some of that gain across.

Far transfer is the interesting claim. It means practising a game and then reasoning better, reading better, or coping better with a working day. That is what the adverts promised.

Think of a gym with exactly one machine in it. Use it for a month and you get much stronger at that machine. Whether you can now carry the shopping up three flights is a separate question, and somebody has to go and measure it.

Jaeggi and colleagues reported in 2008 that training on a dual n-back task lifted scores on a reasoning test. The effect looked large, and it looked dose-dependent. More training, more gain.

That one paper launched an industry. It was also the high-water mark.

Redick and colleagues ran the careful version in 2013. Their comparison group trained on something else rather than sitting out. The n-back scores climbed. Nothing else moved.

Melby-Lervåg and Hulme pooled the working memory training studies the same year and found the same shape. The trained task improved. Reading and reasoning did not, once the designs were tight.

Owen and colleagues had already tried it at scale through the BBC in 2010, with more than eleven thousand volunteers online. Everybody improved at the tasks they had practised. Nobody improved on the tasks they had not.

A second-order meta-analysis in 2019 gathered up the reviews themselves, across chess, music, video games and memory training. Near transfer survived, small and real. Far transfer shrank towards zero as study quality rose.

This is the design flaw behind most of the early excitement, and it is worth a minute because it is not obvious.

Train one group on a memory game for a month. Tell a second group to do nothing. Test both at the end. The trained group knows perfectly well which group it was in.

People who believe they have been improved go on to score better on the test that checks. They sit up, they try harder, and they have usually guessed what the researcher is hoping to see.

Boot and colleagues made this point sharply in 2011 about video game research. If a volunteer can work out what the study expects, the study is measuring that expectation alongside the training.

The repair is a comparison group that also does something absorbing and also has reason to think it might help. Apply that repair and far transfer usually vanishes. That pattern, rather than any one failed study, is why the field cooled.

In January 2016 the Federal Trade Commission settled deceptive advertising charges against Lumos Labs, which makes Lumosity. The company agreed to a two million dollar judgment.

The charge was about the claims rather than the games. The FTC said the advertising promised better performance at work and at school, and protection against decline that arrives with age. It said the evidence behind those promises was not there.

So the honest answer to whether Lumosity works depends entirely on what you asked of it. As a set of games you get better at, it works. As the product the adverts described, a regulator decided the science did not back the claim.

Simons and colleagues published a long review later that same year. They read the industry's own cited evidence. It held plenty of studies showing gains on trained tasks, and very little supporting the everyday benefits being sold.

The strongest evidence in this area comes from ACTIVE, a trial of nearly three thousand older adults published in 2002. Three groups trained on memory, reasoning or speed. A fourth group did not train.

Each group improved at the thing it trained, and some of that improvement was still measurable years afterwards. That is a genuine and useful finding about practice in later life.

It is not the advert, though. Effects on everyday functioning were small, largely self-reported, and did not turn up evenly. Training on speed did not turn anybody into a better reasoner.

The fair summary is a narrow one. Structured practice at a specific ability can hold that specific ability up for a while. It does not spread outwards on its own.

Every test here is a measurement, not a course. We do not sell training. We do not claim that any task on this site will make you smarter, enlarge your memory or improve your work.

Your scores here will rise with practice, and now you know exactly what that rise means. You have got better at this task. The methodology page sets out how each number is taken and where the measuring stops.

That is still worth having. A figure you can repeat on your own machine is a real measurement of a real thing. It is simply a smaller thing than the word cognitive tends to suggest.

No score on this site diagnoses anything or rules anything out. Suppose something about your memory or your concentration has genuinely changed and it worries you. That belongs with a doctor rather than with a web page.

None of this makes practice pointless. It makes practice specific, which is a different fact and a more useful one.

• Practise the actual thing. If you want to be quicker at mental arithmetic, do mental arithmetic. Near transfer is dependable, so aim it at something you want.
• Take the load off rather than trying to grow the store. Write things down, group them on purpose, cut the interruptions. How working memory works explains why that helps so much.
• Sleep before anything that matters. Sleep loss shows up in attention and reaction scores quickly and reliably, and that effect is far better established than any training result.
• Compare yourself with yourself. The gap between two of your own runs on one machine is a real measurement. The gap between you and a stranger is mostly a gap between two computers.
• Distrust any single score. What counts as a good digit span shows how wide an ordinary range really is.