DEV Community

Cover image for 11,430 People Trained Their Brains for Six Weeks. They Got Better at the Games.
Spartan Rich
Spartan Rich

Posted on Edited on

11,430 People Trained Their Brains for Six Weeks. They Got Better at the Games.

Your brain gets better at exactly what you practise. Almost nothing else.

That sounds like a motivational poster. It's closer to a warning, and in 2010 it had 11,430 people behind it.

The biggest brain-training test ever run

Adrian Owen and his colleagues at the Medical Research Council in Cambridge teamed up with the BBC science show Bang Goes the Theory to settle a simple question. Does brain training work?

Viewers signed up online. They took a set of benchmark tests at the start: reasoning, memory, attention, planning. Then they trained several times a week for six weeks on tasks built to stretch those same abilities. A control group spent the time answering general-knowledge questions and looking up answers online instead.

The paper landed in Nature under the plain title "Putting brain training to the test."

Everyone improved on the games they practised. Every single trained task went up.

Then came the benchmark tests at the end. The trained groups were barely better than the people who'd spent six weeks doing trivia. In the authors' words, there was "no evidence" of transfer to untrained tasks, "even when those tasks were cognitively closely related."

Six weeks of work, and the gains stayed inside the games.

Watercolour neurons with one small, densely connected golden pathway lit up while the surrounding network stays pale and unused

A $2 million footnote

Six years later the US Federal Trade Commission went after Lumosity, the best-known brain-training app. The FTC said the company had advertised that its games would help people do better at school and work and could delay age-related decline, including dementia. Lumos Labs agreed to pay $2 million to settle, without admitting the allegations. "Lumosity simply did not have the science to back up its ads," said Jessica Rich of the FTC.

The same year, Daniel Simons at the University of Illinois led a team that went through every study the brain-training industry itself pointed to as evidence. Their review in Psychological Science in the Public Interest found "extensive evidence" that training improves the trained tasks, less for closely related tasks, and "little evidence" for distant tasks or everyday life. None of the cited studies met all of the team's standards for a clean test.

That's the pattern that keeps showing up. Near transfer is real. Far transfer is mostly a hope.

It isn't only apps

You might think this is about cheap phone games. It goes wider.

People have long argued that chess makes kids smarter, or that music lessons boost maths. Giovanni Sala and Fernand Gobet pulled the experiments together in a 2017 paper covering chess, music and working-memory training. At first glance there were small positive effects. Then they split the studies by quality. When the comparison group did some other engaging activity, instead of nothing, the effects shrank to roughly zero: 0.03 standard deviations for music and 0.05 for working-memory training.

The one thing that held up was near transfer. Working-memory training made people better at other memory tasks.

Monica Melby-Lervåg, Thomas Redick and Charles Hulme found the same shape in a meta-analysis of 87 publications on working-memory training. Reliable short-term gains on working-memory tests. No convincing gains on reasoning, reading or arithmetic once there was a proper control group.

Thorndike said this in 1901

None of this would have surprised Edward Thorndike. In 1901 he and Robert Woodworth trained people to estimate the area of rectangles, checking the true answer after every guess. People got much better at rectangles. When the shapes changed, most of that skill didn't come with them.

Their conclusion became the "identical elements" idea: practice spreads to a new task only as far as the two tasks share the same parts. Sala and Gobet cite it a century later as the theory their data keeps supporting.

So what does this mean if you write code?

I think about this every time I see someone grinding puzzle sites to "get better at programming."

You'll get very good at puzzle sites. That isn't nothing if your next interview looks like one. But reading a messy codebase, naming things, debugging a flaky test at 6 pm, designing a schema that won't hurt in a year? Those share fewer elements with a timed puzzle than we like to admit.

The same goes for watching tutorials, memorising syntax cards, or doing typing drills. Each one trains itself. The parts that overlap with your real work carry over. The rest stays behind.

There's one more twist, and it's the hopeful part.

Transfer is possible, but it doesn't happen on its own

In 1980 Mary Gick and Keith Holyoak gave people a famous medical puzzle: how do you destroy a tumour with radiation without burning the healthy tissue around it? Earlier, they'd read a story about a general who splits his army into small groups that arrive at a fortress from every road at once.

Same solution. Many weak rays, converging.

Only about 20% of people spotted the link by themselves. When they were told the story might help, the figure in one experiment jumped to 92%.

Mid-century illustration of a fortress with many roads converging on it, linked by a dotted line to a figure with many thin beams converging on one glowing point

The knowledge was already in their heads. What was missing was the bridge. People don't automatically see that two problems share a structure. When someone points it out, they can.

Practice still works. It just works locally, which gives you a few practical rules:

Practise the real thing. If you want to get better at reading unfamiliar code, read unfamiliar code. Open a library you use and trace one function to the bottom.

Name the shared structure out loud. When you learn a pattern, a retry with backoff or a cache invalidation trick, write one sentence about where else it would apply. That sentence is the hint Gick and Holyoak gave their subjects.

Mix your examples. The same idea seen in three different codebases is more likely to travel than one idea seen three times in the same tutorial.

Bring it back later. A skill you used once doesn't stay usable. Recalling it a few days later, in your own words, is what keeps it there when you need it.

Where I've landed

That last step is the one I kept skipping, which is part of why I built Recl. When I save a conference talk or a good write-up, it turns the save into short recall questions and brings them back over the following days and weeks. It won't make me smarter in general. Nothing in this research says anything will. It just helps the specific things I chose to learn stay reachable.

Recl screen titled "Your memory, scheduled" showing one save coming back as a recall question on day 1, 6, 15, 38 and 94

Your brain is honest about what you train it for. Train it on the thing you actually want.


I'm Asutosh. I write about how the brain learns and remembers, and I build Recl, a tool for remembering what you watch and read.

Sources

Top comments (0)