Why Pilots Forget: The Science of Memory, Decay, and What Actually Works

Sep 18, 2026

The Room You Never Forget

Dr. Stephen J. Ronan, a surgeon and instrument-rated pilot, has told this story before. Years ago, during a thoracic surgery to remove a lung cancer, something uncommon went wrong. The lead surgeon in the room — world-renowned, highly experienced — froze. Ronan had recently studied the exact problem. He spoke up, took over, and the patient survived.

"The difference between freezing and moving wasn't courage or talent," Ronan says of that room. "It was whether the right response was already loaded — retrievable under stress — or something you had to go hunting for while the clock ran out."

Pilots have their own version of that room: an engine failure after takeoff, an electrical fire, a rapid decompression, a vacuum failure in IMC. Rare, unforgiving, and survivable only if the knowledge is already there. Which leaves only one question that actually matters: how much of what you studied is still there?

The uncomfortable answer, according to decades of memory research, is less than you think — and you have no reliable way to feel the difference.

The Forgetting Curve Is Not a Metaphor

In 1885, German psychologist Hermann Ebbinghaus ran the first quantitative study of memory decay, testing himself on lists of nonsense syllables. Retention collapsed fast: roughly 58% gone within 20 minutes, a third of what remained gone by the next day, and only about 21% of the original material still retrievable after a month without review.

That's easy to dismiss as a 19th-century curiosity — nonsense syllables aren't a checkride oral. But in 2015, Jaap Murre and Joeri Dros replicated Ebbinghaus's experiment with modern methods, publishing in PLOS ONE. They found the same curve, 130 years later. Meaningful, structured knowledge decays more slowly than nonsense syllables — but it still decays. Nothing you learn is exempt from the curve; some material just buys a gentler slope.

Pilots Aren't Exempt — and They Can't Tell

If the forgetting curve is universal, does aviation training and real-world stakes change the equation for pilots? Three studies say no — and the second and third findings are more unsettling than the first.

Casner, Heraldez, and Jones (2006) cold-tested 60 certificated pilots on FAA knowledge items — material they had already passed, sometimes years earlier. The average score: 74.8%. The passing line is 70. These weren't students; they were signed-off pilots, and cold, they landed barely five points above the line.

Childs, Spears, and Prophet (1983) tracked 42 private pilots over 8, 16, and 24 months and found something more troubling than decay itself: all 27 measured aviation skills deteriorated — worst for pilots who did no additional training — but the pilots' confidence in their own abilities stayed flat. Competence fell. Self-assessment didn't move. The pilots could not accurately judge their own degradation. "I feel current" is not data; it's a feeling that persists independent of whether it's true.

The Dunning-Kruger effect. Daniel Ellis (2018), a CW5 Black Hawk standardization examiner studied this for his doctorate at Trident University International. Testing 91 aviators, he found they rated their own competence at an average of 76.7% while their measured, actual competence averaged 55.5% — roughly a 21-point overestimation gap. The gap was worst in the lowest-scoring quartiles, and flying hours didn't reliably close it. Ellis called for a much larger, database-backed study across thousands of pilots to see how the gap moves over a career — a study that simply wasn't possible with a sample of 91.

A Safety Gap Nobody Talks About

Compare aviation's approach to knowledge maintenance with any other safety-critical profession. Medicine requires continuing medical education and board recertification. Law requires continuing legal education for an entire career. Aviation requires a knowledge test once for a certificate that lasts a lifetime. The Flight Review required under 61.56 tests mostly practical skill, not comprehensive knowledge, and the FAA's WINGS program credits completing an activity, not retaining what it taught. Aviation is the only safety-critical profession with no ongoing, official verification of knowledge.

The irony: the FAA's own Aviation Instructor's Handbook (FAA-H-8083-9B, Chapter 3) recommends spaced practice over massed practice — the exact fix this article gets to below. No FAA-sanctioned tool implements its own agency's recommendation.

The cost shows up in the numbers. Pilot error is a factor in 80–85% of fatal general aviation accidents (NTSB / AOPA Nall Report). 2024 had the lowest fatal GA accident rate since 2009 — real progress, largely from safer aircraft and equipment — but general aviation still incurred $1.7 billion in insurance losses that year, a 22-year high (Milliman 2024 GA Insurance Analysis). The airplanes got safer. The knowledge didn't get more durable. What's left in the fatal-accident category is stubbornly cognitive.

The Science of Remembering

If the problem is well understood, so is the fix — and it has been sitting in the research literature for decades, mostly ignored. Three mechanisms, applied together, produce an estimated 200–300% better retention for the same amount of study time.

The testing effect. Retrieving a fact from memory strengthens that memory far more than re-reading it does. Roediger and Karpicke (2006) found that repeated self-testing beat repeated re-studying by 50–80% on delayed tests. The best way to study is to be tested — testing isn't just how you measure learning, it's how you cause it.

The spacing effect. When you review matters more than how much. Cepeda and colleagues' 2006 meta-analysis of 184 studies and 839 assessments found that distributed practice — the same total study time, spread out rather than crammed — consistently beats massed practice. There's a mathematically identifiable sweet spot: review a fact right before you'd forget it, and the retrieval strengthens the memory; review too early and you're just re-reading, too late and it's already gone.

Desirable difficulties. If studying feels easy, Robert Bjork (1994) argued, it probably isn't working. Cramming feels productive precisely because it's easy — and easy means the memory isn't being made durable. The techniques that build lasting memory tend to feel harder while you're doing them.

The practical catch with spacing has always been this: no person can track the individually optimal review moment for hundreds of facts across a career. A person can't, but software can. FSRS-5, an open algorithm fit and validated on roughly 700 million real spaced-repetition reviews from 10,000 learners (Ye, Su, and Cao, 2022, KDD), models three things for every fact, for every learner: stability (days until recall probability drops to 90%), difficulty (how hard this item is for this learner), and retrievability (the probability of recall right now). It schedules each review at the last useful moment — just before you'd forget — and on public benchmarks it predicts recall more accurately than the older SM-2 algorithm, with simulations suggesting roughly 20–30% fewer reviews for the same retention.

Two Lies About Memory

Two beliefs about memory are common, comfortable, and wrong. The first: "I'm just bad at memorizing." Memory isn't a fixed talent some people have and others don't — it's a technique, and most people were never taught the one that actually works.

The second, more dangerous belief: "I studied hard, so I know it." Hard work with the wrong technique builds the feeling of knowing while the actual memory quietly evaporates over the following weeks. Effort is an input. A cockpit — or an operating room — only cares about output.

The wrong techniques are also the popular ones. In 2013, Dunlosky, Rawson, Marsh, Nathan, and Willingham reviewed ten common study techniques against the research evidence in Psychological Science in the Public Interest. Only two earned a "high utility" rating: practice testing and distributed practice — the same two mechanisms described above. Elaborative interrogation, self-explanation, and interleaved practice landed in a "moderate" middle tier. The three techniques most people actually use — highlighting, re-reading, and summarization — landed at the bottom, rated low utility.

Re-reading fails for a specific, well-documented reason: the fluency illusion. The second time you read a page, it feels easier — familiar. Your brain interprets "familiar" as "I know this." But familiarity is recognition, and an emergency doesn't hand you a multiple-choice test; it demands recall, cold, from nothing. Childs's 1983 finding — confidence staying flat while competence fell — is this same illusion showing up in the cockpit. "I feel current" is the fluency illusion wearing a flight jacket.

Cramming has its own honest limitation: it works, for about one day. Casner, Jones, Puentes, and Irani (2004) showed that FAA knowledge questions can be passed through rote memorization — but reword the same question slightly and scores collapse, because the rote memory never became real understanding. You cram, you pass, and the knowledge is largely gone before the certificate is even printed.

What AI Changes

In 1984, educational researcher Benjamin Bloom documented what he called the "2 sigma problem": one-on-one tutoring is so effective that the average tutored student outperforms 98% of students taught in a conventional classroom — a two-standard-deviation improvement. Bloom's problem was that nobody could scale it. A personal tutor for every student, or every pilot, was never economically realistic.

That's the constraint AI removes — and the evidence isn't a demo-day anecdote. Kestin and colleagues (2025), in a Harvard-run crossover randomized controlled trial published in Nature Scientific Reports, found an AI tutor outperformed in-class active learning by an effect size of 0.73 to 1.3 standard deviations — and the AI-tutored group needed less time to do it, 49 minutes versus 60. Maye's 2026 meta-analysis of 21,415 learners in medical education found a large effect size (d = 1.01) for spaced retrieval practice. This isn't theoretical for medicine already: surveys show 86.2% of U.S. medical students use spaced-repetition tools, 66.5% of them daily.

A separate 2024 meta-analysis by Wang and colleagues found that adaptive learning systems match or exceed the results of expert human instructors — meaning the benefit doesn't depend on finding a great teacher. The science behind all of this has been solid for decades. What changed is the ability to deliver it at the scale and personalization Bloom said the classroom could never reach.

When It Matters Most: Under Stress

All of this matters most exactly when it's hardest to access. Landman and colleagues (2017), writing in Human Factors, describe the startle effect: under sudden, severe stress, attention narrows and working memory collapses. What survives isn't everything you technically know — it's only what's been consolidated to automaticity. A procedure memorized once, two years ago, for a checkride, is not consolidated. Under startle, it will not be there. "Sort of remember" performs, in the moment, exactly like "don't know."

The military has known this for decades, which is why it never drills boldface procedures — the handful of steps flown from memory before the checklist even comes out — just once. Boldface items are drilled verbatim, handwritten from memory, on a recurring schedule, for an entire career. Stripped to its mechanism, that discipline is simply retrieval practice plus spacing, aimed deliberately at the small number of items where a wrong order kills people. The general aviation translation isn't to memorize the entire emergency-procedures section — most of it is correctly a checklist item. It's to give boldface-level drilling only to the time-critical, order-matters items: engine failure after takeoff, in-flight fire, runaway trim, emergency descent. Military pilots have a squadron training officer enforcing that schedule. A GA pilot has to be their own.

The same mechanism extends past pure recall, into judgment. Hendrickson, Goldsmith, and Johnson (2006) found that maneuvers pilots had mentally rehearsed — chair-flown — before an evaluation decayed less than those that hadn't been rehearsed. Chair-flying a scenario pulls an entire decision chain out of memory rather than passively re-reading it: the same three techniques — testing, spacing, desirable difficulty — aimed at decisions instead of isolated facts.

The Takeaway: A 10-Minute Protocol

None of this requires an hour a day. It requires roughly ten minutes, most days, correctly timed.

Ten minutes of properly spaced retrieval practice can hold knowledge that would otherwise decay to around 21% within a month — the same collapse Ebbinghaus measured in 1885 and Murre and Dros confirmed in 2015. A well-tuned spacing algorithm schedules something like 15 to 25 reviews a day — not busywork, but items timed to the last useful moment before they'd fade. After roughly six months of consistent practice, the curve flattens: more material has consolidated, and reviews become less frequent because less is actively decaying. The compounding does the rest.

The science here isn't new or speculative. Ebbinghaus's curve is 140 years old, the testing effect and spacing effect are decades-old findings, and the FAA's own instructor handbook already recommends spaced practice. What's changed is the ability to actually run that schedule — for a career's worth of material, across thousands of individually timed facts and procedures, at the moment each one is about to be forgotten. The pilots most likely to freeze in the room that never leaves them aren't the ones who never studied. They're the ones whose knowledge quietly decayed below the threshold of what they could feel, while their confidence stayed exactly where it was.

If you something to help with this problem, check out FlyCurrent.ai/CUCKOO. FlyCurrent uses the FSRS-5 algorithm to space learning of aviation material. Now you can get a special deal of $10/month for 3 months to try out FlyCurrent. After 3 months the rate goes to $39.99/month (first 1000 people to join), with the rate guaranteed for 3 years. After the first 1000, the rate is $59.99/month. There is no obligation. I think you will find it a valuable platform for learning and retaining aviation information.

Mike “Cuckoo” Kloch from information compiled by Dr. Stephen Ronan

Sources

• Ebbinghaus, H. (1885). Über das Gedächtnis.

• Murre, J. M. J., & Dros, J. (2015). Replication and Analysis of Ebbinghaus' Forgetting Curve. PLOS ONE.

• Casner, S. M., Heraldez, D., & Jones, K. M. (2006). International Journal of Applied Aviation Studies, 6(1), 71–97.

• Childs, J. M., Spears, W. D., & Prophet, W. W. (1983). Private Pilot Flight Skill Retention (DOT/FAA/CT-83/34).

• Ellis, D. B. (2018). Doctoral dissertation, Trident University International. ProQuest 13427420.

• Roediger, H. L., & Karpicke, J. D. (2006). Test-Enhanced Learning. Psychological Science, 17(3).

• Karpicke, J. D., & Blunt, J. R. (2011). Science, 331, 772–775.

• Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Psychological Bulletin, 132(3), 354–380.

• Bjork, R. A. (1994). Memory and Metamemory Considerations in the Training of Human Beings. Metacognition: Knowing About Knowing.

• Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Psychological Science in the Public Interest, 14(1), 4–58.

• Ye, J., Su, J., & Cao, Y. (2022). FSRS-5. KDD '22.

• Casner, S. M., Jones, K. M., Puentes, A. R., & Irani, T. (2004). Retention of Airman Knowledge (NASA/TM-2004-212814).

• Bloom, B. S. (1984). The 2 Sigma Problem. Educational Researcher, 13(6).

• Kestin, G., et al. (2025). Nature Scientific Reports.

• Maye, A. (2026). Meta-analysis of spaced retrieval in medical education.

• Landman, A., Groen, E. L., van Paassen, M. M., Bronkhorst, A. W., & Mulder, M. (2017). Human Factors, 59(8), 1161–1172.

• Hendrickson, J. M., Goldsmith, T. E., & Johnson, P. J. (2006). Mental rehearsal and retention of aviation skills.

• Wang, Z., et al. (2024). Adaptive learning vs. non-adaptive instruction: a meta-analysis.

• NTSB / AOPA Nall Report.

• Milliman (2024). GA Insurance Analysis.