How fast do skills decay?

Quickly enough to matter, and unevenly. The best available synthesis puts skill loss at d = −0.01 immediately after training and d = −1.4 after more than 365 days without practice. The uneven part is the finding that should change how organisations think about automation: the tasks that decay fastest are the cognitive ones. The hands hold up. The judgement does not.

This question has been answered for decades in aviation, emergency medicine and the military, and almost none of it has reached the discussion about AI at work. What follows assembles it.

Minus 1.4 after a year

Arthur and colleagues meta-analysed 189 independent data points from 53 articles on skill decay and retention. Loss ran from d = −0.01 immediately after training to d = −1.4 after more than 365 days of non-use. Graded entry.

An effect size of 1.4 is very large. For scale, most workplace interventions that anyone bothers to publish sit under 0.5. This is the size of the gap between a person at the end of their training and the same person a year later, having not used what they learned.

The part that matters more is the moderator. Physical, natural and speed-based tasks decayed less. Cognitive, artificial and accuracy-based tasks decayed more. Skill decay is not one curve. It is at least two, and the steeper one belongs to the kind of work that is currently being automated.

The hands were fine. The thinking was not.

Casner and colleagues put sixteen airline pilots through routine and non-routine scenarios in a Boeing 747-400 simulator, varying the level of automation. Instrument scanning and manual control were mostly intact, even among pilots who reported little recent hand-flying practice.

What failed was the cognitive layer: tracking position without a map, deciding the next navigational step, and recognising instrument failures. Those showed frequent and significant problems. Graded entry.

Sixteen pilots in a simulator is a small study and the authors do not offer it as a general rule for knowledge work. Its value is that it arrives at Arthur's moderator by a completely different route. A meta-analysis of 189 data points across five decades of training research, and a cockpit study of sixteen people, both find that the manual half survives disuse better than the thinking half.

That is the finding this research keeps returning to, and here it has a number and a mechanism. When automation absorbs the cognitive work and leaves the manual work, it removes practice from precisely the faculty that loses it fastest.

Six months, and the trades that regulate to it

The Red Cross reviewed 47 studies of CPR skill retention across healthcare and lay populations, with retest intervals from six weeks to 24 months. Substantial degradation occurs within the first year, and retention declines between six and twelve months unless there is refresher training. Graded entry. The decay was measured on manikins rather than in resuscitations, so it establishes the curve rather than the consequence.

Aviation converted this into law. Under 14 CFR 121.441 a pilot in command must pass a proficiency check every 12 calendar months, and within every 6 calendar months either a proficiency check or an approved simulator course. Graded entry. Those intervals are a regulatory minimum rather than a figure calibrated to a measured decay curve, and they are still more than any other profession requires.

The FAA's own working group on flight path management identified vulnerabilities in manual handling after transition from automated control, and in how those skills are defined, developed and retained. It also recorded that pilots sometimes rely too heavily on automated systems and can be reluctant to intervene. Graded entry.

Crew resource management is the counter-example worth knowing, because it shows the limit of training as an answer. Helmreich's review found that line audits confirm CRM produces the intended behavioural change, and that measured attitudes decay over time even with recurrent training. Graded entry. Recurrent training slows the curve. It does not abolish it.

In months, when the tool is good enough

The intervals above come from disuse: the skill is not practised because the situation does not arise. AI introduces a different mechanism, where the situation arises constantly and the person no longer does the thinking.

Budzyń and colleagues examined 1,443 colonoscopies performed without AI assistance across four Polish centres, by nineteen endoscopists averaging 27.6 years of experience. Unassisted adenoma detection fell from 28.4 per cent to 22.4 per cent after routine exposure to an AI detection tool, within months. Graded entry.

The comparison is imperfect, and the imperfection matters. The classical literature measures what happens when a skill is not used. Budzyń measures what happens when it is used with help. Those are different exposures, and the second one has been studied for a fraction as long. What makes it notable is that the effect appeared in months rather than years, in practitioners with nearly three decades of practice, in a domain where the moderator says decay should be slower because the task has a large perceptual component.

What comes back

Decay is not deletion. Murre and Dros replicated Ebbinghaus across intervals from twenty minutes to thirty-one days and found relearning to criterion took less time than original learning at every interval tested, confirming the savings effect first reported in 1885. Graded entry.

Cepeda and colleagues, meta-analysing 839 assessments across 317 experiments, established that spacing and retention interval act jointly: the gap between practice sessions that produces the best retention widens as the target retention interval widens. Graded entry. If you want a capability to survive a year, the practice that maintains it should be spaced further apart than if you want it to survive a week.

Both results come from verbal material rather than professional judgement, and the Ebbinghaus replication is a single subject. The recovery side of this question is treated properly at can you regain a skill you have lost.

Two different things are called skill decay

Everything above measures decay through disuse: the skill degrades because it stops being practised. The popular discussion usually means something else, decay through obsolescence: the skill stays intact and the world stops needing it. The half-life framing belongs to the second.

Rahim Hirji set out the obsolescence case in The Half-Life of Skills (Box of Amazing, 8 June 2025), arguing that the risk is not forgetting what you knew but carrying skills whose value has gone, and that the people who cope are not more skilled but less attached.

Keeping the two apart matters because they call for opposite responses. Obsolescence is answered by letting go and learning something else. Disuse is answered by continuing to practise the thing you already have. Advice built for one will damage the other, and most published guidance does not say which it is addressing.

The half-life literature also carries figures that circulate far more confidently than they can be traced, including a widely repeated claim that the half-life of a technical skill has fallen to about two and a half years. This page does not use them, on the same basis as the most quoted AI statistics, checked.

What none of this establishes

No study here measures the decay of professional judgement over a career. Arthur's synthesis covers trained tasks with a defined criterion, which is not what a partner, a consultant or a physician does on a hard case. Arthur states this directly: the meta-analysis does not establish how fast any particular professional skill decays or how quickly it can be regained.

The intervals in aviation and CPR are also not transferable numbers. They were set by regulators and committees weighing decay evidence against cost and practicality, and neither is a measured optimum. Anyone quoting "six months" as the moment a skill goes should say which skill, measured how, and against what criterion.

And the AI mechanism is the least studied of all of them. One observational study in one procedure in one country carries most of the weight, alongside two experiments that produced the gap deliberately rather than observing it in the field.

What follows from the shape of the curve

Three things, and the first is the one organisations get wrong.

Protect the cognitive practice first. The instinct is to preserve the visible, manual, demonstrable part of a job because it is the part that looks like the work. The evidence points the other way twice over: the manual half survives disuse better, and the cognitive half is what automation takes.

Set an interval and make it fail-able. Aviation's advantage is not the frequency, it is that a proficiency check can be failed. An unaided exercise nobody can fail is a calendar event rather than a measurement. This is the argument in what professions can learn from aviation.

Expect the curve, and stop treating decay as a failure of the individual. Skill loss under non-use is the normal behaviour of trained capability, documented since 1885. What is new is an arrangement that removes the practice while the work continues, so nothing signals that the interval has started. That accumulation is capability debt.

Key sources

• Arthur, W., Bennett, W., Stanush, P. L. and McNelly, T. L. (1998). Factors that influence skill decay and retention: a quantitative review and analysis. Human Performance, 11(1). Graded entry.
• Casner, S. M., Geven, R. W., Recker, M. P. and Schooler, J. W. (2014). The retention of manual flying skills in the automated cockpit. Human Factors, 56(8). Graded entry.
• American Red Cross Scientific Advisory Council (2009). Scientific review: CPR skill retention. Graded entry.
• Murre, J. M. J. and Dros, J. (2015). Replication and analysis of Ebbinghaus' forgetting curve. PLoS ONE, 10(7). Graded entry.
• Cepeda, N. J. et al. (2006). Distributed practice in verbal recall tasks: a review and quantitative synthesis. Psychological Bulletin, 132(3). Graded entry.
• 14 CFR 121.441, Proficiency checks. Electronic Code of Federal Regulations. Graded entry.
• Federal Aviation Administration (2013). Operational use of flight path management systems: final report. Graded entry.
• Helmreich, R. L., Merritt, A. C. and Wilhelm, J. A. (1999). The evolution of crew resource management training in commercial aviation. International Journal of Aviation Psychology, 9(1). Graded entry.
• Budzyn, K. et al. (2025). Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy. The Lancet Gastroenterology and Hepatology, 10(10), 896-903. Graded entry.

Related SuperSkills research

On recovery, can you regain a skill you have lost. On the concept, deskilling and capability debt. On the mechanisms that maintain a skill, retrieval practice, desirable difficulty and productive struggle. On the professions that measure it, what professions can learn from aviation and which professions face the greatest deskilling risk. On why nobody notices, the illusion of competence. On measurement, assessing capability rather than output.

About this research

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. Skill decay is an established field in human factors and training research, not a SuperSkills coinage. Every finding on this page is attributed to the study that produced it and kept separate from the interpretation.

How this research works  ·  Reviewed quarterly  ·  Found an error? Tell me and it is corrected on the page.