Can you regain a skill you have lost?

Usually yes, and faster than you learned it the first time. That is the one genuinely reassuring finding in this whole territory, and it changes what capability loss actually costs. It does not make the loss free, because the recovery has to be paid for in the one currency organisations are least willing to spend, which is time on work that produces nothing.

Almost every page on this site describes capability draining away. This one asks the question that follows, and the answer is more hopeful than the rest of the argument implies.

How fast skill actually goes#

Arthur and colleagues pooled 189 data points from 53 studies to ask what happens to a trained skill over an interval of not using it. The headline is a range rather than a number: skill loss ran from a d of about zero immediately after training to a d of -1.4 after more than a year of non-use. That is a large effect by any conventional reading.

The moderators matter more than the headline. Physical, natural and speed-based tasks decayed less. Cognitive, artificial and accuracy-based tasks decayed more.

Which is the wrong way round for anyone reading this. The skills that survive disuse best are the ones AI is least interested in taking. The skills that go first are judgement, diagnosis, and knowing which of several defensible answers is right, which is the entire content of professional work.

The same split shows up when you look at one profession closely. Casner and colleagues found airline pilots' instrument scanning and manual control mostly intact after periods of little practice, while the cognitive tasks, tracking position, deciding the next step, spotting an instrument failure, showed frequent and significant problems. The hands remembered. The judgement did not.

A number worth holding: months, not years#

The clearest interval evidence comes from somewhere unglamorous. A systematic review of 47 studies on CPR skill retention found substantial degradation within the first year, with retention declining from around six to twelve months unless there was refresher training. CPR is heavily trained, procedurally simple, high stakes and widely practised. It still goes in months.

Two caveats. The decay was measured on manikins, and the review found no studies linking manikin performance to actual patient outcomes. And CPR is not a fair proxy for a complex cognitive skill, though if anything the direction of that error is unhelpful: the Arthur moderators say cognitive skills should decay faster, not slower.

Do not merge the aviation and clinical numbers into a single figure. They measure different things over different intervals and are not comparable. What they agree on is the order of magnitude. Months.

The savings effect, which is the good news#

Ebbinghaus noticed in the 1880s that relearning something apparently forgotten takes less effort than learning it did. He called it savings. The measure is not what you can recall but how much faster you get back.

Murre and Dros replicated the whole thing in 2015 across intervals from twenty minutes to thirty-one days, with ten replications at each. Relearning to criterion took less time than the original learning at every interval tested.

So the thing you cannot do any more is not gone. It has become cheaper to rebuild than it was to build. That is a materially different proposition from starting again, and the reason this page exists.

The limits are real and should be stated. This is one subject learning nonsense syllables. Nobody has demonstrated savings for professional judgement over the timescale of a career, and it would be a considerable stretch to assume the effect transfers unchanged from lists of syllables to the ability to read a room or price a risk. What the finding supports is a direction, not a discount rate.

What the recovery evidence actually says to do#

The strongest evidence on how to rebuild is about when you practise rather than how much. Cepeda and colleagues synthesised 839 assessments across 317 experiments and found spacing and retention interval acting jointly: the gap between practice sessions that produces the best retention gets longer as the period you want to retain over gets longer.

Put plainly, if you want a capability to survive a year, practising it four times across that year beats practising it four times in a fortnight, for the same total effort. The synthesis covers verbal recall rather than professional skill, so this is applied by analogy and should be read that way.

The practical consequence is that recovery is a scheduling problem more than a volume problem, which makes it cheaper than most organisations assume and easier to keep deferring than almost anything else on a plan.

Three things this cannot tell you#

All three are limits of the evidence rather than of the argument.

Nobody has measured skill recovery in AI-displaced knowledge work specifically. Every study here predates the question. The mechanism is well established and the setting is not.

Savings is demonstrated for verbal material and simple procedures. Whether the compressed, pattern-based judgement that takes a decade to build behaves the same way is unknown, and there is a plausible argument that it does not, because what decays may be the pattern library rather than the skill of using it.

And there is a floor nobody has located. A skill never built cannot be regained, which is the whole point of the missing rungs. Savings applies to recovery, not to acquisition that never happened.

What follows for anyone worried about their own capability#

• Test before you assume. The uncomfortable version is doing a piece of real work unaided and noticing where it is harder than it used to be. Most people have never checked and are guessing in one direction or the other.
• Expect the judgement to have gone before the technique. If you feel rusty, the fluent-looking part is probably fine and the deciding part is where the damage is.
• Space the rebuild. Four sessions across a year beats four in a fortnight. Recovery is a calendar problem.
• Count in months. Whatever your own decay interval is, the evidence across several domains says it is shorter than people expect.
• Do not wait for the crisis to find out. The moment you need a capability is the worst moment to discover you no longer have it, which is the argument the whole aviation literature is built on.

Related SuperSkills research#

On what is being lost, capability debt and deskilling. On the repetitions that build judgement in the first place, the missed reps and the missing rungs. On the mechanism that makes effortful practice work, desirable difficulty. On the profession that has thought about this longest, what professions can learn from aviation. On checking yourself, am I becoming dependent on AI.

Key research and primary sources

• Arthur, W., Bennett, W., Stanush, P. L. and McNelly, T. L. (1998). Factors that influence skill decay and retention. Human Performance, 11(1).
• Murre, J. M. J. and Dros, J. (2015). Replication and Analysis of Ebbinghaus' Forgetting Curve. PLoS ONE, 10(7).
• Casner, S. M. et al. (2014). The Retention of Manual Flying Skills in the Automated Cockpit. Human Factors, 56(8).
• American Red Cross Scientific Advisory Council (2009). Scientific Review: CPR Skill Retention.
• Cepeda, N. J. et al. (2006). Distributed Practice in Verbal Recall Tasks. Psychological Bulletin, 132(3).

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. Every study here predates generative AI, and the transfer to AI-displaced knowledge work is inference from adjacent evidence rather than direct measurement, which the page states rather than obscures. The savings effect is demonstrated for verbal material and has not been shown for professional judgement.

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