In 2015 the psychologists Joshua Hartshorne and Laura Germine analysed the scores of 48,537 people tested on the web and reported that vocabulary does not peak until around the age of 65. Now that generative AI has entered everyday work, what actually grows as we age? Vocabulary and knowledge grow, as does the ability to reason about a conflict from several sides, and so does day-to-day emotional stability. Processing speed falls, and the large gains from generative AI so far have come on the side of less experienced workers' output.

01Vocabulary and reasoning about conflict grow with age; speed declines

Three lines of research answer the question with numbers. Hartshorne and Germine found that vocabulary peaks around 65 in web-based testing. The social psychologist Igor Grossmann and colleagues found that people aged 60 to 90 were better than younger adults at reasoning about a conflict from several perspectives. Laura Carstensen and colleagues, tracking the same people for ten years, found that emotional life becomes steadier with age.

The losses are just as clear. The speed of matching digits to symbols peaks in the late teens and declines after that. Ageing, in other words, does not pull every ability down at once. Some abilities rise while others fall across the same span of ages.

One link to generative AI belongs at the start. In a large study of customer-support work, the people whose output rose most with an AI assistant were the less experienced agents. The abilities that grow with age and the gap that AI narrowed are two different things.

If we treat age as a single story of decline, we lose sight of that difference. Separating which abilities grow from which gaps AI has closed gives us a better way to think about our own role at work, and about the roles of older and younger colleagues. That separation is what I want to argue for here.

02Each ability peaks at a different age; none brings them all together

The figure of 65 for vocabulary means more when it is set beside the other abilities. Using the published norms of a standard intelligence and memory battery, Hartshorne and Germine compared the age at which performance was highest on each task. Digit-symbol substitution, a test of how fast a person can match symbols to numbers, peaked in the late teens. Memory for strings of digits peaked in the early twenties. Vocabulary peaked at around 50.

Figure 1 Peak age by ability
Symbol-substitutionspeedlate teensDigit memoryearly 20sVocabulary (testnorms)around 50Vocabulary (webdata)around 65Symbol-substitution speedlate teensDigit memoryearly 20sVocabulary (test norms)around 50Vocabulary (web data)around 65
Peak ages differ by more than 40 years depending on the ability. Vocabulary peaks around 50 or around 65 depending on how it is measured.

The vocabulary peak moves with the method. In the test norms it came at around 50; in the web data it came at around 65. The authors state plainly that the web data put the peak later. I cannot say which of the two ages is the true one. What both measurements share is that vocabulary keeps growing until at least around 50.

The authors go further. On many tasks, they write, there may be no single age at which a person is at their best on all of them. At the age when speed is highest, vocabulary is still climbing. By the age when vocabulary is highest, speed has long been in decline.

On this evidence, there is no one age that can be called the prime of life. Whether a 20-year-old or a 60-year-old does better depends entirely on which ability is being compared.

03Gains in reasoning about conflict appeared among Americans but not among Japanese participants

Vocabulary is not the only ability that researchers have found rising with age. In 2010 Grossmann and colleagues gave residents of Michigan short accounts of conflicts to read. Some described conflicts between groups; others described disputes between individuals. Participants were asked how they thought the situations would unfold, and their answers were scored.

People aged 60 to 90 scored higher than younger adults. The older participants made more use of reasoning that considers other people's perspectives, looks for room to compromise, and admits the limits of what one knows. Grossmann and colleagues presented this as a strength that grows into old age.

The result did not hold everywhere. In 2012 Grossmann and colleagues compared Japanese and American participants using the same approach, and the findings split by country and by the type of conflict.

1

American participants

Older age went with higher scores for wise reasoning.

2

Japanese participants

No relationship appeared between age and wise-reasoning scores.

3

Conflicts between groups

Young and middle-aged Japanese scored higher than Americans of the same age.

Among the Japanese participants, older people did not outscore younger ones. On conflicts between groups, Japanese participants already scored high when young. Either way, this research gives no basis for carrying the American finding over to Japanese workplaces, or to the people who review promotional materials there.

04Speed and vocabulary or reasoning are driven by separate factors

If the gain in reasoning about conflict differs between Japan and the United States, a more basic puzzle remains: why should vocabulary rise while speed falls? Hartshorne and Germine concluded that because peak ages are scattered so widely, age-related change cannot be explained by a single cause. In their account, several dissociable factors act separately on different areas of thinking.

Figure 2 Direction of each ability with age
Growing olderProcessing speedfalls from late teensVocabulary andknowledgerises to about 65Reasoning aboutconflicthigh at 60-90 in the USEmotional stabilityrises over 10 years of reportsGrowing olderProcessing speedfalls from late teensVocabulary and knowledgerises to about 65Reasoning about conflicthigh at 60-90 in the USEmotional stabilityrises over 10 years of reports
Speed is the branch that declines; the other three rise. The gain in reasoning about conflict was measured among American participants.

Grossmann and colleagues reached a compatible conclusion. The ability to solve unfamiliar problems on the spot, which psychologists call fluid intelligence, declines with age. Even so, reasoning about social conflict improved across the same age range. The ability that declined and the ability that improved moved in opposite directions across the same age range.

Neither study identifies which factor drives which ability, and I will not fill that gap with guesses.

What can be said is that drawing age as one falling line erases the abilities that rise. At work, the changes people notice first are often the slower reply and the longer time it takes to learn a new tool. Those are changes in speed and in solving unfamiliar problems. Growth in vocabulary, or in the capacity to see a dispute from several sides, is much harder to see from outside.

05Emotional stability was measured with five daily reports repeated over ten years

If separate factors move separate abilities, then the third ability on the rising side, emotional stability, needs its own measurement. Carstensen and colleagues did not rely on test scores or on the scoring of written answers. They recorded feelings as people lived them.

About 180 people aged 18 to 94 reported their emotional state five times a day for one week, at moments chosen at random. The same people repeated the exercise five years later and again ten years later.

As people aged, positive emotional states became more common and their emotions fluctuated less. People who reported more positive emotion were more likely to be alive thirteen years later. That is an association found in the same people, not proof of a cause. The study does not show that positive feeling lengthened anyone's life.

AspectHartshorne et al. 2015Grossmann et al. 2010Carstensen et al. 2011
Who was measured48,537 web participants and test norms247 community residents in MichiganAbout 180 people aged 18 to 94
MethodTask performanceScored responses to conflict storiesFive emotion reports a day, over ten years
What rose with ageVocabularyReasoning about perspectives, compromise and the limits of knowledgePositive states and emotional stability

The three studies measured different people in different ways. They still agree on one point: some abilities grow with age. The second study's gain, though, was measured among American participants and did not appear among Japanese ones. That agreement across unrelated methods is the main reason I trust the answer this column gives.

06Generative AI helped the less experienced most, and few older Americans use it yet

Three separate studies found vocabulary, reasoning about conflict and emotional stability all rising with age. Placed next to research on generative AI and data on who uses it, those findings raise three points.

First, the gain in reasoning about conflict depends on where and on whom it was measured. The reasoning that improved among Americans aged 60 to 90 did not improve with age among the Japanese participants. Anyone who says at work that older colleagues handle disputes better should add whose data that claim rests on. Grossmann's results describe a tendency among American participants; they say nothing certain about the person in front of you.

1

34% for less experienced agents

An AI assistant raised resolutions per hour, with minimal effect on experienced, highly skilled agents.

2

10% among those 65 and over

In 2025, 58% of US adults under 30 had used ChatGPT, against 10% of those 65 and older.

Second, the gains from generative AI were largest for less experienced workers. The economist Erik Brynjolfsson and colleagues studied 5,179 customer-support agents who were given an AI assistant and compared how many issues they resolved per hour. Novice and low-skilled agents improved by 34%. Experienced, highly skilled agents saw almost no change. What AI narrowed was a gap in output, which is a different thing from the gap in vocabulary or reasoning that age widens. One caution matters here: the study divided workers by tenure and skill, not by age. A less experienced worker is not necessarily a young one.

Third, the people using these tools skew young. In a Pew Research Center survey of 5,123 US adults in early 2025, 58% of those under 30 said they had used ChatGPT, compared with 10% of those 65 and older. At least for ChatGPT, older Americans still have little contact with AI output.

07No study yet found measures whether older workers' vocabulary and reasoning help in reviewing AI drafts

Once we know that AI's help falls mostly on the less experienced side, the obvious next question is where the abilities that grow with age might count in AI-assisted work. Reviewing a draft written by generative AI is one candidate, including drafts of promotional materials. That work means catching a poorly chosen word, or noticing that a sentence ignores how a reader with a different stake will take it. Vocabulary and reasoning about conflict look relevant, but that is my guess and nothing more.

The research so far measured age and ability on one side, and generative AI and output on the other, in separate studies. Within what I was able to check, I found no study that tested whether older workers make a difference when reviewing AI output. I do not claim none exists. I can only say that I looked and did not find one.

The research does tell us how such a study would have to be built.

Figure 3 Steps to test what has not been measured
The same AIdraftOlder andyounger…Count theerrors foundSeparateage from…Check for adifferenceThe same AI draftOlder and younger staff review itCount the errors foundSeparate age fromyears of experienceCheck for a difference
Age and years of experience tend to move together. Unless the two are counted separately, no effect of age can be claimed.

The key is to separate age from years of experience. Older workers usually also have longer careers. If the two are not counted separately, any difference that shows up could come from either. Because Brynjolfsson and colleagues divided workers by experience and not by age, their results say nothing about age on its own.

One possibility also follows from the facts above. The more AI closes the gap in output between workers, the more the remaining differences may sit in reviewing drafts and making judgements about them. If that happens, and if vocabulary and reasoning about conflict do help in review, the abilities that grow with age would have more places to matter. That is my inference from the evidence so far; no one has yet put a number on it.

  1. Vocabulary peaks around 65 in web-based testing, while processing speed peaks in the late teens. Ageing does not bring every ability down at once.
  2. Reasoning about conflict from several perspectives was stronger among Americans aged 60 to 90, but no age gain appeared among Japanese participants. The gain in this ability depends on which country's participants were measured.
  3. Generative AI raised resolutions per hour by 34% among less experienced agents, and only 10% of US adults aged 65 and over had used ChatGPT in 2025. I found no study measuring how the abilities that grow with age apply to AI work.
Closing

Ageing adds vocabulary and knowledge, the ability to reason about a conflict from several perspectives (a gain seen among American participants), and emotional stability. What declines is speed. Generative AI narrowed a different gap, in the output of less experienced workers. Within what I was able to check, I found no study measuring whether the gains of age help in reviewing AI output. A study that wants to find out should begin by counting age and years of experience separately.

Sources & references
  1. Hartshorne, J. K. & Germine, L. T. (Psychological Science). When does cognitive functioning peak? The asynchronous rise and fall of different cognitive abilities across the lifespan. 2015-03-13. https://pmc.ncbi.nlm.nih.gov/articles/PMC4441622/ (different peak ages by ability; vocabulary peak around 65 in web data and around 50 in test norms; dissociable factors act on different domains)
  2. Grossmann, I., Na, J., Varnum, M. E. W., Park, D. C., Kitayama, S. & Nisbett, R. E. (PNAS). Reasoning about social conflicts improves into old age. 2010-04-20. https://pmc.ncbi.nlm.nih.gov/articles/PMC2867718 (people aged 60 to 90 outperformed younger adults in reasoning about conflict from several perspectives)
  3. Association for Psychological Science (via EurekAlert!). Does wisdom really come with age? It depends on the culture. 2012-08-30. https://www.eurekalert.org/news-releases/812600 (age gain in wise reasoning among Americans but not among Japanese participants)
  4. Carstensen, L. L. et al. (Psychology and Aging). Emotional experience improves with age: Evidence based on over 10 years of experience sampling. 2011. https://pmc.ncbi.nlm.nih.gov/articles/PMC3332527 (five reports a day over ten years; greater emotional stability and positive well-being with age; association with survival 13 years later)
  5. Brynjolfsson, E., Li, D. & Raymond, L. R. (NBER Working Paper 31161). Generative AI at Work. 2023-04. https://www.nber.org/papers/w31161 (34% rise in resolutions per hour for novice and low-skilled agents among 5,179 workers)
  6. Pew Research Center. 34% of U.S. adults have used ChatGPT, about double the share in 2023. 2025-06-25. https://www.pewresearch.org/short-reads/2025/06/25/34-of-us-adults-have-used-chatgpt-about-double-the-share-in-2023/ (ChatGPT use at 58% under 30 and 10% at 65 and older)