There is a tempting way to read a result like this: learn the tango, pick up a guitar or play a strategy game, and the brain becomes younger. The research is more interesting than that slogan, but also much more limited.
The “age” in this study was not the age of brain cells, and no one’s calendar was turned backwards. It was a prediction generated from patterns of communication between brain regions. When those patterns looked more like the model’s younger examples, the researchers described the difference as a younger brain-age gap.
That distinction matters. A brain clock can register experience-dependent change without showing that someone will live longer, avoid dementia or think like a person three years younger. It is a statistical summary of selected signals, not a medical diagnosis.
Even with that caution, the pattern reported in Nature Communications is hard to dismiss casually. It appeared in four very different forms of expertise, grew with skill, and shifted in the same direction after a short training experiment.
A clock built from communication
A normal clock tells time because its mechanism advances predictably. A brain clock works backwards from data. Researchers show a machine-learning model brain measurements from people whose ages are known, and the model learns which combinations tend to accompany ageing.
For this study, the training set contained electroencephalography, or EEG, data from 1,240 healthy people aged 17 to 91. They came from Argentina, Brazil, Chile, Colombia, Cuba, Greece, Ireland, Italy, Turkey and the United Kingdom.
The team converted the recordings into functional-connectivity maps covering 78 cortical regions. In simple terms, they measured how strongly activity in each region rose and fell with activity elsewhere across frequencies from 8 to 40 hertz, then asked a support-vector machine to predict age from that network.
Across cross-validation tests, predicted and chronological age correlated at 0.742. The mean absolute error was 8.7 years. That is useful at group level, but it also makes clear why a three-year movement in one derived score should not be mistaken for a precise biological birthday.
The one-person arithmetic wrinkle
The paper says in its results that 1,473 people were included, which is the figure used in the headline. Its abstract and figure legend, however, describe 1,240 people in the training data and another 232 in the creative-experience data, a total of 1,472.
A methods paragraph gives yet another total after exclusions. The study’s public descriptions sensibly fall back on “more than 1,400.” This small accounting inconsistency does not alter any comparison, because the sizes of the expert and training groups are separately reported, but it is worth making visible.
The international spread was genuinely broad. Adding the creative datasets from Canada, Germany and Poland brought the work to 13 countries. Yet the large headline number mostly belongs to the reference population that trained the clock, not to the expert comparison itself.
The tests of expertise used 196 people: 46 tango dancers in Argentina, 58 musicians in Canada, 30 visual artists in Germany and 62 strategy gamers in Poland. The remaining creative-experience data came from the training study and its active control.
Four kinds of long practice
Within each domain, experts were compared with non-experts matched for age, sex, education and location. The gaming comparison also matched working-memory capacity, while the tango comparison considered several cognitive measures. That is stronger than comparing unrelated volunteers pulled from different places.
Expertise did not mean quite the same thing everywhere. Tango dancers were divided using more than 12 months of formal instruction. Musicians had at least five years of experience. Visual artists had at least three years of university-level art education, and expert gamers met recent play and competitive-ranking requirements.
Across all four domains, the experts’ estimated brain-age gaps were 5.5 years lower on average. Broken apart, the differences were 7.1 years for tango dancers, 6.2 for visual artists, 5.38 for musicians and 4.06 for gamers.
Each estimate came with a wide confidence interval because the groups were small. Still, all four pointed in the same direction after correction for multiple comparisons. When gaming was removed from the pooled analysis, the broader result remained.
A dose of skill, not a magic category
The most persuasive part of the expert comparison may be the gradient. Among 105 participants with a usable measure of expertise, greater skill or experience was associated with a more negative, or younger-looking, brain-age gap.
This was not evidence that tango contains one special anti-ageing movement or that drawing activates a youth switch. The four activities make overlapping demands: sustained attention, prediction, error correction, memory, motor planning and the flexible construction of something that is not fully specified in advance.
That last quality is why the authors treated real-time strategy play as a creative activity. A StarCraft II player must allocate resources, read an opponent, change plans and execute many decisions under time pressure. The product is not a painting, but the play is neither passive nor fixed.
The same principle appears in quite different learning research. As ScienceBlog previously reported, two adult musicians reached perfect-pitch performance after intensive training. Both studies push against the idea that adult expertise is merely the visible residue of childhood talent.
What 30 hours of StarCraft changed
The short experiment gave the researchers a better handle on cause. Twenty-four people with little recent gaming experience trained on StarCraft II for 30 hours over three to four weeks. Their playing was supervised in a laboratory, at five to ten hours per week, and EEG was recorded before and after.
After training, their mean brain-age gap shifted by 3.06 years in the younger direction. A separate group of 12 people spent the same amount of time learning Hearthstone, a slower, turn-based game selected as an active control. Its average change was effectively zero.
The participants also became faster at the game. Among the 20 with complete performance data, people whose actions per minute improved more tended to show larger reductions in estimated brain age. On an attention task, the trained group responded faster to one target and detected a second target more accurately.
This is more informative than a cross-sectional snapshot because each learner served as their own baseline, and the control group helps separate StarCraft II from simply spending time with a game. But 24 learners and 12 controls remain a small experiment. The authors explicitly warn that its effect size may be overestimated or underestimated.
The networks underneath the number
A single brain-age score can conceal more than it reveals, so the team looked at where the signal came from. The strongest expert-related changes tended to occur in frontoparietal hubs and other regions whose connectivity was most vulnerable to age in the reference data.
The maps differed in sensible ways. Expert data were associated with functions involving movement, rhythm, coordination, imagery and visual salience. The short gaming experiment leaned more heavily towards visual perception, object recognition, fixation and attention.
The authors also examined network efficiency. Local efficiency describes how effectively nearby clusters can exchange information, while global efficiency reflects communication across the larger network. Lower brain-age gaps were associated especially strongly with greater local efficiency.
In experts, the model also found stronger global coupling, a measure intended to capture the overall strength of communication between regions. The short experiment did not show that global effect, which fits the possibility that deeply established expertise and a few weeks of training leave different signatures.
Why “younger” needs quotation marks
The model did not inspect tissue damage, plaques, blood vessels or individual neurons. It inferred age from functional connectivity. Even the expert-gaming data required simulated EEG connectivity derived from diffusion MRI, while the other domains used resting EEG or magnetoencephalography recorded under different protocols.
The researchers harmonised those data in a shared brain atlas and tested whether electrode number or signal quality explained the result. Every creative dataset also had its own control group. Those are serious safeguards, but harmonisation cannot make different instruments and old datasets literally identical.
Brain-age estimates are also shaped by the model’s reference population and its assumptions. The brain does not mature along one straight track. Another recent ScienceBlog article described how structural brain networks may keep reorganising until around age 32, illustrating why one summary number cannot capture every developmental process.
A younger predicted age is therefore best read as resemblance: this person’s measured connectivity looks more like patterns the model learned from younger people. It is not proof that all aspects of the brain are younger, much less that ageing has been reversed.
What selection can still explain
The training experiment supports short-term plasticity, but the larger expert result remains observational. People were not randomly assigned years earlier to become dancers, musicians, artists or gamers. Their starting brains may have influenced what they enjoyed and how far they progressed.
Long practice also travels with other experiences. Tango involves exercise, touch, music and social coordination. Performing music may bring social connection and disciplined rehearsal. Art education can track opportunity and income. Competitive gaming is tied to particular ages, routines and technology access.
Matching reduces some obvious differences, but it cannot remove unmeasured factors such as socioeconomic status, sleep, physical activity or personality. The authors acknowledge that future studies need to examine these influences and recruit larger samples in more creative domains.
Nor did the study follow experts into old age or count cases of cognitive decline. It cannot say whether a five-year brain-age gap persists, delays disease or produces a noticeable advantage in daily life. The Trinity College Dublin summary presents creativity as promising for brain health, but the clinical claim still awaits clinical evidence.
Practice may matter more than the label
The study offers no basis for prescribing StarCraft II over piano, or piano over dance. It tested selected groups that happened to have usable brain data. Writing, acting, crafts, languages and many other demanding practices were absent.
It is equally possible that “creativity” is standing in for a bundle of ingredients: novelty, increasing difficulty, feedback, pleasure, agency and repeated coordination across several systems. Those components can keep a practice challenging after the beginner stage, when many routines become automatic.
The authors’ Global Brain Health Institute perspective argues that creative opportunities deserve a place in healthy-ageing policy because they can be affordable, culturally adaptable and enjoyable. The present findings make that proposal testable; they do not settle its effect on health outcomes.
A sensible personal interpretation is modest. Choose a demanding activity worth returning to, accept being clumsy at first, and let skill accumulate. The best reason to dance, draw, make music or play a strategic game remains the experience itself. A younger-looking connectivity pattern is an intriguing possible consequence, not a guarantee.
A result that should stay open
The research earns attention because the signal converged across countries, instruments and activities. It also scaled with expertise, appeared after controlled training and was linked to plausible changes in network organisation.
Its limitations are just as important: small domain groups, reused datasets, heterogeneous measurements, a modest control group and no evidence yet about lasting protection from cognitive decline. The brain clock’s own average prediction error was larger than several reported group differences.
Replication should use preregistered studies, larger active controls and follow-up months or years later. Measuring memory, wellbeing, physical health and everyday function alongside brain age would show whether the computational shift travels with changes people can actually feel.
For now, the study says something narrower and more believable than “creativity reverses ageing.” Long-practised expertise and a short bout of demanding strategy training were associated with brain-activity networks that a model read as younger. That is not youth restored. It is evidence that an adult brain remains responsive to what it repeatedly learns to do.