TL;DR: The cognitive literature does not say that adults over 40 cannot learn. It says different abilities peak at very different ages: in one large study, processing speed peaked in the late teens, short-term memory span in the early twenties, working memory around 30, emotion recognition showed a broad plateau between 40 and 60, and vocabulary peaked around 50 in standardised test norms and around 65 in the same team’s web sample. The claim that general decline begins in the twenties is real, published and openly contested in the same journal issue for relying on cross-sectional data that cannot separate ageing from cohort differences. Meanwhile the official European statistics point somewhere else entirely. Between 2021 and 2025 the share of European Union adults with at least basic digital skills rose 13.1 percent among 45 to 54 year-olds and 20.1 percent among 55 to 64 year-olds, against 8.6 percent among 25 to 34 year-olds and 4.7 percent among 16 to 24 year-olds. Adult learning survey data shows that when older adults do want training, they convert that intention into participation at almost the same rate as younger adults. And the reason most often given by those who wanted training and did not get it is scheduling, at every single age band. The binding constraint after 40 is a calendar problem and an intent problem, not a capability problem.
“Am I too old to learn this?” is one of the most searched career questions of the AI era, and it is almost always answered with encouragement rather than evidence. The evidence is more interesting than the encouragement, and in one respect considerably more uncomfortable.
There are two separate bodies of data here that are rarely read together. One is the psychology of cognitive ageing, which is genuinely contested. The other is official European statistics on adult skills and training participation, which are not contested at all and which almost nobody consults when thinking about this question. Put side by side, they point to a conclusion neither one reaches alone.
What the cognitive research actually found
The single most useful study for this question is Hartshorne and Germine’s 2015 paper in Psychological Science, which combined 48,537 web participants with a comprehensive analysis of normative data from standardised IQ and memory tests. Its central finding was not that cognition declines. It was that the notion of a single peak is wrong.
In their words, the results reveal considerable heterogeneity in when cognitive abilities peak: some peak and begin to decline around high school graduation, some plateau in early adulthood and begin to decline in the thirties, and still others do not peak until the forties or later.
The specific task-level results are worth stating precisely, because the popular summary of this literature usually flattens them. Digit symbol coding, a processing-speed measure, peaked in the late teens. Digit span peaked in the early twenties. Working memory tasks peaked at around 30. Emotion recognition showed a much broader peak than any other task, reflecting a long period of relative stability in performance between the ages of 40 and 60. Vocabulary peaked around 50 in the standardised test norms and around 65 in the authors’ own web-based sample, a difference the authors attributed at least partly to generational effects, with later peaks in more recent cohorts.
Two implications follow directly.
The first is that the abilities peaking latest are the ones most professional work actually runs on. Vocabulary, accumulated knowledge and the reading of other people are not incidental to knowledge work. They are most of it. The abilities peaking earliest are raw speed and short-term span, which are exactly the abilities that tools have been substituting for since the invention of the notebook.
The second is that the vocabulary result includes a cohort effect the authors observed in their own data: peaks arrived later in more recent generations. That is not a small technical footnote. It is the crack running through the entire field.
The disagreement that should change how you read all of this
In 2009, Timothy Salthouse published a paper in Neurobiology of Aging arguing that some aspects of age-related cognitive decline begin in healthy educated adults when they are in their twenties and thirties. It is a careful paper, and it is widely cited.
In the same journal, K. Warner Schaie published a direct response arguing that this reifies what he called the cross-sectional fallacy. His argument is precise. Age changes and age differences can only be identical if there was a perfectly stable environment over time and no differences in performance between successive birth cohorts at the same age, and, as he put it, nobody has ever provided empirical data meeting those conditions. Comparing a 30 year-old to a 55 year-old today does not measure what happens to a person as they age. It measures the difference between two groups who were born, educated and employed in different decades.
That objection matters enormously for the practical question, and it matters equally for the statistics in the rest of this piece, which are also cross-sectional. Any table showing skills by age group is showing cohorts, not trajectories. That caveat is stated here rather than buried, and it applies to the tables below as much as to Salthouse.
What the official statistics actually show
Two European Union datasets speak to this directly. Both are official, both are current, and both are almost entirely absent from the popular discussion of reskilling after 40.
Table: Digital skills and adult learning participation by age, European Union, 2025 (source data as published by the European statistical office)
| Age group | Adults with basic or above basic overall digital skills, percent | Participation in education and training in the last four weeks, percent |
|---|---|---|
| 16 to 24 | 74.55 | 63.7 (ages 18 to 24) |
| 25 to 34 | 74.47 | 21.2 |
| 35 to 44 | 70.14 | 14.0 |
| 45 to 54 | 62.28 | 12.2 |
| 55 to 64 | 50.49 | 8.5 |
| 65 to 74 | 33.00 | 4.5 |
The two columns come from different surveys with different reference periods and are not a like-for-like ratio. What they can be used for is a comparison of gradients, and the gradients are strikingly different.
Table: Capability and access indexed against the 25 to 34 age group (CEOtudent editorial framework, derived from the two series above)
| Age group | Digital skills index | Training participation index | Access relative to capability |
|---|---|---|---|
| 25 to 34 | 100.0 | 100.0 | 1.00 |
| 35 to 44 | 94.2 | 66.0 | 0.70 |
| 45 to 54 | 83.6 | 57.5 | 0.69 |
| 55 to 64 | 67.8 | 40.1 | 0.59 |
| 65 to 74 | 44.3 | 21.2 | 0.48 |
Read the 45 to 54 row slowly. That cohort retains 83.6 percent of the young-adult digital skill level while participating in training at 57.5 percent of the young-adult rate. Whatever is happening to European 45 year-olds, it is happening roughly twice as fast to their training as to their measured skills. The calculation is CEOtudent’s; both underlying series are the statistical office’s.
The finding that reverses the usual story
Now hold the age groups constant and look at change over time. The same two series were published in 2021, before the current wave of AI tools, and again in 2025.
Table: Four-year change 2021 to 2025 by age group (CEOtudent editorial framework, derived from the published series)
| Age group | Digital skills 2021 | Digital skills 2025 | Relative gain | Training participation 2021 | Training participation 2025 | Relative gain |
|---|---|---|---|---|---|---|
| 16 to 24 | 71.17 | 74.55 | +4.7 percent | 61.8 (18 to 24) | 63.7 (18 to 24) | +3.1 percent |
| 25 to 34 | 68.56 | 74.47 | +8.6 percent | 18.3 | 21.2 | +15.8 percent |
| 35 to 44 | 64.31 | 70.14 | +9.1 percent | 11.2 | 14.0 | +25.0 percent |
| 45 to 54 | 55.07 | 62.28 | +13.1 percent | 9.2 | 12.2 | +32.6 percent |
| 55 to 64 | 42.04 | 50.49 | +20.1 percent | 5.7 | 8.5 | +49.1 percent |
| 65 to 74 | 25.45 | 33.00 | +29.7 percent | 2.3 | 4.5 | +95.7 percent |
Both gradients run in the same direction, and it is the opposite of the direction the cultural story predicts. In relative terms, the older the cohort, the larger the four-year improvement, in digital skills and in training participation alike, monotonically across every band. The 55 to 64 group improved its digital skills 4.3 times as fast as the 16 to 24 group and increased its training participation more than fifteen times as fast.
The honest reading of that requires the Schaie caveat again. These are repeated cross-sections, not a panel. Someone who was 45 in 2021 is 49 in 2025, so part of what moves each band is cohort replacement rather than individual learning. The result does not prove that any specific 50 year-old learned anything. What it does establish is that the population aged over 45 in Europe is measurably more digitally skilled than the population aged over 45 was four years ago, by a larger margin than any younger group, and that no plausible story about fixed adult incapacity survives that fact.
Intent, not capacity, is the bottleneck
The most decision-changing number in this whole area comes from the European adult education survey, which asks not only whether people trained but whether they wanted to.
In 2022, among European Union adults aged 30 to 54, 48.4 percent participated in education or training, and a further 11.1 percent did not participate but wanted to. Among adults aged 55 to 69, 29.0 percent participated and 7.5 percent wanted to but did not.
Convert those into a conversion rate: of everyone who was interested in training, what share actually got it?
Table: Interest converted into participation, European Union, 2022 (CEOtudent editorial framework, derived from the published adult education survey)
| Age group | Participated, percent of population | Wanted to but did not, percent of population | Total interested | Interest converted into participation |
|---|---|---|---|---|
| 18 to 24 | 78.7 | 5.8 | 84.5 | 93.1 percent |
| 30 to 54 | 48.4 | 11.1 | 59.5 | 81.3 percent |
| 55 to 69 | 29.0 | 7.5 | 36.5 | 79.5 percent |
The conversion rates for the 30 to 54 and 55 to 69 groups differ by 1.8 percentage points. Participation itself differs by 19.4 points. Almost the entire gap between a middle-aged adult and an older adult in actual training is a gap in how many of them wanted training in the first place, not in how many of those who wanted it managed to get it.
That reframes the whole problem. If older adults who wanted training were being systematically shut out, you would see the conversion rate collapse. It does not collapse. What collapses is the interested population, from 59.5 percent to 36.5 percent.
What actually blocks the ones who do want it
For the minority who wanted training and did not get it, the survey records the main reason. The pattern is the same at every age, and it is not what the cultural script expects.
Table: Main reason given by adults who wanted education or training but did not participate, European Union, 2022, percent within each age group
| Main reason | 25 to 34 | 35 to 44 | 45 to 54 | 55 to 64 |
|---|---|---|---|---|
| Schedule | 24.6 | 21.6 | 24.0 | 20.1 |
| Costs | 17.0 | 13.9 | 12.8 | 9.3 |
| Family reasons | 13.2 | 20.4 | 12.7 | 9.4 |
| Other personal reasons | 9.4 | 7.8 | 7.7 | 8.6 |
| Health or age reasons | 3.6 | 3.6 | 6.7 | 12.2 |
| No suitable offer available | 6.1 | 6.0 | 7.7 | 9.0 |
| Lack of support from employer or public services | 6.7 | 8.6 | 7.8 | 7.2 |
| Distance | 2.5 | 2.1 | 2.4 | 2.8 |
| No response | 8.3 | 8.3 | 10.4 | 13.8 |
Schedule is the single largest obstacle in every age band, and it barely moves with age: 24.6 percent at 25 to 34 and 20.1 percent at 55 to 64. Cost falls steadily with age rather than rising. Health or age reasons do rise, from 3.6 percent to 12.2 percent, but even in the oldest band 87.8 percent of people who wanted training and did not get it named something other than health or age. At 45 to 54 the figure is 93.3 percent.
That is the practical heart of this piece. For a 47 year-old who wants to reskill, the most statistically likely obstacle is not the brain and not the money. It is the week.
What to actually do with this
If the constraint is intent and calendar rather than capability, the interventions change completely.
Design for the schedule constraint first, not the cognition. The most common blocker is time-shaped, so the first move is a time-shaped fix: a fixed, protected, recurring block rather than a resolution to study more. Scheduling is the top-cited obstacle at every age, which means it is the single highest-yield thing to fix regardless of how old you are. Anyone building a learning plan should start there, and a structured 90-day personal curriculum is largely an exercise in calendar design.
Choose skills that sit on the late-peaking abilities. The cognitive evidence is a targeting instrument, not a verdict. Skills that lean on accumulated knowledge, judgement, vocabulary and reading people are where the research says a 50 year-old holds an advantage that a 25 year-old cannot manufacture. Skills that lean on raw processing speed and short-term span are where the advantage runs the other way, and they are also the ones most exposed to automation. Choosing accordingly is a strategic decision rather than a consolation, and it is the same logic behind distinguishing which parts of career capital compound and which decay.
Treat the intent gap as your actual competitive advantage. The conversion data says something unusual: among people over 55, only about one in ten non-participants even wanted to train. If you are over 40 and you want to, you are already in a small minority of your cohort, and the scarcity is on the wanting rather than on the doing. That is an unusually cheap edge.
Set a realistic time budget from evidence rather than folklore. The single most common failure in later-career reskilling is not slow learning. It is a wrong estimate of how long competence takes, followed by quitting at the point where the estimate was exceeded. Working from documented time-to-competence figures for in-demand skills removes the most common reason people conclude, wrongly, that age was the problem.
Start from an inventory, not from a course. Before choosing what to learn, establish what is already there. Late-career learners usually underestimate transferable knowledge and overestimate the novelty of the new field, which is the exact opposite of the error that younger learners make. Mapping what you know and what is expiring is the cheapest step in the whole sequence.
The CEO and the student in the same person
The CEO reading of this data is about resource allocation under a constraint that has been misdiagnosed. If you believe the constraint is cognitive, you invest in motivation and hope. If you accept that the constraint is schedule and intent, you invest in calendar architecture and in the decision to start, which are both things you actually control. Misdiagnosing a constraint is the most expensive error an allocator can make, and this one is misdiagnosed at population scale.
The student reading is more demanding. The most uncomfortable number in this piece is not any of the skill measures. It is that among European adults aged 55 to 69 who did not participate in learning, only about one in ten wanted to. The research offers no support for the belief that the door closes at 40. It offers a great deal of evidence that most people stop trying to open it, and that those who keep trying get through at very nearly the same rate as everyone else.
Age is not the variable that stopped most people. Wanting to was.
Frequently asked questions
Does cognitive decline after 40 make reskilling harder?
Some abilities do decline, and processing speed is the clearest case, peaking in the late teens on standardised measures. But the abilities that most professional learning depends on peak far later, with vocabulary peaking around 50 in test norms and emotion recognition holding a broad plateau between 40 and 60. Whether general decline begins as early as the twenties is genuinely disputed in the literature, on the grounds that the evidence is cross-sectional and cannot separate ageing from generational differences.
Is it harder to get training at 50 than at 30?
Participation is much lower, but the survey data suggests it is not mainly a gate-keeping problem. Conditional on wanting training, adults aged 55 to 69 converted intent into participation at 79.5 percent against 81.3 percent for 30 to 54 year-olds. The difference in participation is driven overwhelmingly by how many people in each group wanted training.
What is the single biggest obstacle for people over 40 who want to reskill?
Schedule. It is the most cited main reason at every age band examined, at 24.0 percent for 45 to 54 year-olds and 20.1 percent for 55 to 64 year-olds. Health or age reasons account for 6.7 percent and 12.2 percent respectively.
Why are the older age groups improving fastest?
The data shows the pattern but does not explain it, and part of the movement in any repeated cross-section is cohort replacement rather than individual learning. What can be said is that the relative four-year gain in both digital skills and training participation rises monotonically with age across every band from 16 to 74.
Should I worry that these are European figures?
The skills and training series are European Union aggregates and should be read as such. The cognitive research is not geographically bounded in the same way, since it draws on standardised test norms and a large international web sample. Anyone outside the European Union should treat the gradients as indicative of a pattern rather than as their own national numbers.
Does the cross-sectional caveat undermine the whole argument?
It limits one specific claim, that individuals of a given age are improving, and that claim is not made here. It does not affect the conversion-rate finding, which compares intent to participation within the same survey and the same year, and that is where the practical conclusion comes from.
Sources
Hartshorne, J. K. and Germine, L. T. When does cognitive functioning peak? The asynchronous rise and fall of different cognitive abilities across the lifespan. Psychological Science, 2015.
Salthouse, T. A. When does age-related cognitive decline begin? Neurobiology of Aging, 2009.
Schaie, K. W. Commentary on Salthouse, published in the same issue of Neurobiology of Aging, 2009, on the cross-sectional fallacy and cohort effects.
European statistical office, individuals’ level of digital skills, European Union aggregate, survey years 2021, 2023 and 2025.
European statistical office, participation rate in education and training in the last four weeks, European Union aggregate, annual series 2020 to 2025.
European statistical office, Adult Education Survey 2022, persons who want to participate in education and training, and main reason for non-participation among those who wanted to, European Union aggregate.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
