Wow...after having done several posts uncritically passing on studies by Jaeggi and others claiming that games to improve working memory, such as the n-Back game, increase cognitive skills in other areas, a number of studies have failed to replicate these phenomena. Gareth Cook has done an interesting
article on this in The New Yorker, suggesting that claims made by commercial software sites like Cogmen, Lumosity, and CogniFit are bogus.
Over the last year, however, the idea that working-memory training has broad benefits has crumbled. One group of psychologists, lead by a team at Georgia Tech, set out to replicate the Jaeggi findings, but with more careful controls and seventeen different cognitive-skills tests. Their subjects showed no evidence whatsoever for improvement in intelligence. They also identified a pattern of methodological problems with experiments showing positive results, like poor controls and a reliance on a single measure of cognitive improvement. This failed replication was recently published in one of psychology’s top journals, and another, by a group at Case Western Reserve University, has been published since.
The recent meta-analysis, led by Monica Melby-Lervåg, of the University of Oslo, and also published in a top journal, is even more damning. Some studies are more convincing than others, because they include more subjects and show a larger effect. Melby-Lervåg’s paper laboriously accounts for this, incorporating what Jaeggi, Klingberg, and everyone else had reported. The meta-analysis found that the training isn’t doing anyone much good. If anything, the scientific literature tends to overstate effects, because teams that find nothing tend not to publish their papers. (This is known as the “filedrawer” effect.) A null result from meta-analysis, published in a top journal, sends a shudder through the spine of all but the truest of believers. In the meantime, a separate paper by some of the Georgia Tech scientists looked specifically at Cogmed’s training, which has been subjected to more scientific scrutiny than any other program. “The claims made by Cogmed,” they wrote, “are largely unsubstantiated.”
The controversy is becoming an interesting story! I refer to the following snippet from KurzweilAi:
ReplyDeleteTraining increases connectivity
Bornali Kundu of the University of Wisconsin-Madison, who works in Postle’s laboratory, used transcranial magnetic stimulation (TMS) with electroencephalography (EEG) to measure activity in specific brain circuits before and after training with an n-back task.
“Our main finding was that training on the n-back task increased the number of items an individual could remember over a short period of time,” said Kundu. “This increase … was associated with enhanced communication between distant brain areas, in particular between the parietal and frontal brain areas.”
Michael Pohlmann
In the n-back task, Kundu’s team presented stimuli one-at-a-time on a computer screen and asked participants to decide if the current stimulus matched both the color and location of the stimulus presented a certain number of presentations previously. The color varied among seven primary colors, and the location varied among eight possible positions arranged in a square formation.
The control task was playing the video game Tetris, which involves moving colored shapes to different locations, but does not require participants to remember anything. Before and after the training, researchers administered a range of cognitive tasks on which subjects did not receive training, and simultaneously delivered TMS while recording EEG, to measure communication between brain areas during task performance.
After practicing the n-back task for 1 hour a day and 5 days per week over 5 weeks, subjects were able to remember more items over short periods of time. Importantly, for those whose working memory improved, communication between the dorsolateral prefrontal cortex (DLPFC) and parietal cortex also improved. “This is in comparison to the control group, who showed no such differences in neural communication after practicing Tetris for 5 weeks,” Kundu says.
Working-memory training also produced improvement on cognitive tasks for which participants were not trained that are also believed to rely on communication between the parietal cortex and DLPFC.
For two of these tasks — the ability to detect a change in a briefly presented array of squares, and the ability to detect a red letter “C” embedded in a field of distracting stimuli of rotated red “C”s and blue “C”s — those who had trained in the n-back test also showed a decrease in task-related EEG. The training exercise had registered a similar decrease.
“The overall picture seems to be that the extent of transfer of training to untrained tasks depends on the overlap of neural circuits recruited by the two,” Kundu says.