Research Institution: University of Michigan
Research title: Improving fluid intelligence with training on working memory
Researchers: Susanne M. Jaeggi, Martin Buschkuel, John Jonides, & Walter J. Perring
Training program used in research: n-back exercise
Published: Proceedings of the National Academy of Sciences, May 2008 Volume 105, pages 6829-6833
Fluid intelligence (Gf) is “…a complex human ability that allows us to adapt our thinking to a new cognitive problem or situation.” It is critical for a wide variety of cognitive tasks and one of the most important factors in learning. Fluid intelligence is highly heritable, i.e., determined in significant ways by genetic factors, and does not typically increase with education. It is generally regarded as a relatively fixed capacity of individuals that cannot be meaningfully enhanced.
The question examined in this study is whether intensive working memory training of working memory would produce gains in fluid intelligence. The theoretical underpinning of this possibility is the idea that fluid IQ and working memory share a common capacity constraint, i.e., the number of elements that can be retained in working memory and the number of elements that can be simultaneously considered during an abstract reasoning task. In addition to this theoretical bridge between working memory and fluid intelligence, recent work suggests that the two abilities rely on similar neural circuits. Thus, training WM could conceivably transfer to gains in fluid intelligence.
To test this hypothesis, 70 healthy young adults attending the University of Bern in Switzerland (mean age 25.6 years) were randomly assigned to a control condition or one of four WM training conditions. Control participants engaged in no training activity; those in the WM training groups received training that lasted 8, 12, 17, or 19 days.
Fluid IQ was measured before and after training; the post-test assessment occurred within 2 days of the final training session. The time interval between fluid IQ assessments for control participants was comparable to that of training participants. Standardized assessments of fluid IQ – the Ravens Advanced Progressive Matrices and the BOMAT, which is a more difficult variant of the RAVENS – were used to measure fluid IQ. In these tests, each problem presents a matrix of patterns from which one pattern is missing. Subjects are required to select the missing pattern from a set of alternatives. Thus, the test is measuring non-verbal reasoning ability.
Working Memory Training
The WM training task employed was complex and I will do my best to explain it clearly. Squares that would appear at one of eight different locations on a computer screen were presented sequentially at a rate of 3 per second. Each stimulus appeared for 2500 milliseconds with 500 milliseconds in between.
At the time each square was presented, subjects heard one of eight consonants that were read through headphones. The task was to decide on each trial whether the square location and consonant presented matched those presented a fixed number of trials earlier.
In the easiest condition, called the one-back condition, the subject had to indicate whether the consonant and square location matched the consonant and location that came right before it. If the consonant matched they would hit one key and if the location matched they hit another key. In the two-back condition, they had to decide whether the consonant and location matched those that were presented two trials earlier.
Thus, to respond correctly, subjects had to retain both auditory and visual spatial information in working memory. This is referred to as an ‘n-back’ task where n represents the number of prior presentations in the sequence that subjects compare new presentations to. The farther back they have to go, i.e., the higher n, the more difficult the task.
Each training session consisted of 20 blocks of 20 trials (400 trials total) and took about 20 minutes to complete. After each block, subjects’ performance was analyzed and ‘n’ either increased, decreased, or stayed the same depending on how well the subject performed. Thus, the difficulty level for each block of trials was adjusted to match the subject’s most recent level of performance.
As you can imagine, this is an extremely challenging task that requires both intense focus and excellent working memory capacity to do well at. Working memory is evoked because subjects have to retain each location and consonant pair in mind so they can be compared to the new stimuli presented.
Gains in Working Memory
The first outcome examined was whether participants’ working memory capacity increased with training. This was computed by plotting the average level of ‘n’ achieved during each training session. Because ‘n’ represents the number of presentations ‘back’ that subjects have to match, increases in ‘n’ reflect increased working memory capacity.
All four training groups showed gains in working memory capacity. The average ‘n’ at the start of training increased steadily over training sessions. For the group that trained 19 days, the average ‘n’ on the final day had increased to roughly 5. The slope of the increase was comparable in all 4 groups and there was no indication that gains had leveled off by the end of training. Thus, if training had persisted additional gains in WM capacity may have been made.
Gains in Fluid Intelligence
The above results demonstrate that participants improved on the task they trained on. Gains on trained tasks are not difficult to demonstrate and the key question is whether training led to gains on the test of fluid intelligence. As is hopefully clear from the above description, the test used to assess fluid IQ bears little similarity to the task used for WM training, so getting better on the WM training task would not automatically result in better scores on the fluid IQ assessment.
Although fluid IQ test performance was identical for training and control participants at baseline, follow-up testing indicated significantly higher scores for trained participants. In addition, gains in fluid IQ showed a linear relationship to training days, such that subjects who trained for more days tended to make greater gains on the fluid IQ test. Furthermore, although fluid IQ improved more for participants whose scores were lower to begin with, improvements were found across the full range of initial ability.
Summary and Implications
Results from this study provide compelling evidence that intensive training of working memory can improve fluid IQ. Thus, an extremely important cognitive ability that was previously believed to be fixed (as, by the way, working memory was also thought to be) can be increased with training. It is easy to imagine that this finding may have important educational implications.
While these are exciting results, additional research is required to better understand the practical applications of these findings. First, it will be necessary to determine how long the gains in fluid IQ persist, as no long-term follow up was conducted in this study.
Second, although the fluid IQ measure used in this study is known to have important real world correlates, it will be important to demonstrate that the benefits of WM training transfer not just to tests of fluid IQ, but also to meaningful real world outcomes, e.g., school performance. In that regard, see a recent study of Cogmed Working Memory Training that was associated with long-term gains in mathematics reasoning –http://www.helpforadd.com/2009/april.htm
Third, it is possible that the external validity of fluid IQ assessments, i.e., the aspects of real-world functioning that such tests predict, may change when scores have been influenced by WM training. For example, imagine two children with fluid IQ scores of 120, which is well above average. One child obtains that score naturally, i.e., without training, and the other child obtains that score after training has led to a 15-point improvement. In this case, would the two children truly have equal capability to succeed at academic tasks that depend on fluid IQ? This seems likely but remains an open question until empirically demonstrated.
Fourth, the control group in this study received no training of any kind. A stronger design would be one where control participants received training on some other task that was not related to either WM or fluid IQ. This would help document that it is WM training specifically, and not spending time on any cognitive demanding task, that is necessary to improve fluid IQ.
Fifth, participants were all Swiss college students and replicating these findings with a more diverse group of participants is essential.
It is likely that this impressive initial demonstration will stimulate further studies to address these issues, as well as others. In the meantime, however, the authors have already made a significant contribution to the literature on cognitive training and intelligence, as their findings modify existing notions of fluid intelligence as a relatively fixed and immutable trait.
Note – You can read an excellent commentary on this study by Dr. Robert Sternberg at pnas.org