What professions can learn from aviation

Aviation met this problem forty years ago. Automation took over the flying, pilots became monitors, and the industry discovered that a monitor who rarely intervenes is worse at intervening than the operator who used to do the job. Knowledge work is arriving at the same place with no institutional memory of how anyone handled it, and the answers aviation reached are unglamorous, expensive and mostly ignored elsewhere.

The useful thing about aviation is not that it solved the problem. It is that it was forced to write its solutions into law, which means they are specific enough to copy.

Bainbridge named the trap in 1983#

Lisanne Bainbridge's paper Ironies of Automation is four pages long, predates the personal computer, and describes what is happening in professional services in 2026 with uncomfortable precision. Her ironies, in her order:

Designers who regard the operator as unreliable automate what they can and leave the operator whatever they could not think how to automate, usually without redesigning the support around it. Manual control skills deteriorate when they are not used, so the person called on to take over is less capable than they would have been before automation existed. Monitoring for rare failures is, in her word, humanly impossible to sustain much beyond half an hour. And the sting in the tail: it is the most successful automated systems, with the rarest need for intervention, that need the greatest investment in operator training.

That last line is the one to take into an AI deployment meeting. The better the system, the more you have to spend on the humans around it, which is the opposite of the business case almost everyone writes.

The paper is conceptual. It reports no accident statistics and no effect sizes, and it should be read as a framework rather than as evidence. What has accumulated since is the evidence.

What happened when the industry checked#

In 2013 the Federal Aviation Administration published the report of its Flight Deck Automation Working Group, a synthesis of accident data, incident reports and operator surveys running to 28 findings. Two matter here. The group found vulnerabilities in manual handling after transition from automated control, and specifically in the definition, development and retention of those skills. It also found that pilots sometimes rely too much on automated systems and may be reluctant to intervene.

A regulator, examining an entire industry, named skill retention as a finding. No equivalent exercise has been run for any knowledge profession.

The most-cited case sits underneath it. The French investigation into Air France 447, which fell into the Atlantic in 2009 after its airspeed sensors iced, cited among its contributing factors the lack of practical training in high-altitude manual handling and in the procedure for speed anomalies. Three qualified pilots, a serviceable aircraft, and a manual flying situation none of them had practised recently.

One accident is not a base rate, and it should not be used as one. What it is, is the reason the rest of this list exists.

The finding that should worry knowledge work most#

In 2014 Casner and colleagues put 16 airline pilots into a Boeing 747-400 simulator and varied how much automation they had. The result splits neatly in two.

Instrument scanning and manual control were mostly intact, even where the pilots themselves reported little recent practice. The hands remembered. What did not hold up were the cognitive tasks: tracking position without a map, working out the next navigational step, recognising an instrument failure. Those showed frequent and significant problems, and the degree of trouble tracked how much task-unrelated thinking the pilots had been doing while the automation flew.

Read that against a profession where the entire job is cognitive. Aviation's reassuring half, the motor skills, is the half knowledge work does not have. What decayed in the cockpit is the only thing a lawyer, an analyst or a clinician does.

The sample is 16 pilots in a simulator, so this is a signal rather than a settled quantity. It is the sharpest signal available.

Four things aviation wrote into law#

1 · Recurrent tested practice, as a condition of the licence. Under 14 CFR 121.441 a pilot in command must pass a proficiency check every 12 calendar months, and within every 6 months either another check or an approved simulator course. Not training attendance. A check, which can be failed, after which you do not fly.

No knowledge profession does this. Continuing professional development almost everywhere counts hours attended. It is worth being clear that the aviation intervals are a regulatory minimum rather than a figure calibrated to a measured decay curve, but the principle underneath is the part to copy: the test is periodic, it is real, and failing it has consequences.

2 · Protected attention, as a rule rather than a virtue. The sterile cockpit rule, 14 CFR 121.542, prohibits any crew activity during a critical phase of flight other than what is needed to operate the aircraft. Critical phases are defined precisely:

"Critical phases of flight includes all ground operations involving taxi, takeoff and landing, and all other flight operations conducted below 10,000 feet, except cruise flight."

The interesting move is that the rule does not ask anyone to concentrate. It removes the things that would prevent concentration, during a period defined in advance, and puts the obligation on the operator as well as the individual.

3 · A named training response to a named failure. Crew Resource Management started at a 1979 NASA workshop, prompted by a finding that a captain had failed to accept input from junior crew. The industry did not issue a values statement about speaking up. It built a training programme and audited behaviour on the line.

Helmreich and colleagues, reviewing twenty years of it, are careful about what can be claimed. Accident rates are too rare to serve as a validation criterion, so the evidence is behavioural: line audits show the intended changes appear. They also record that measured attitudes decay over time even with recurrent training, and that a subset of pilots is never reached. CRM should not be described as proven to reduce accidents. Its own architects say that cannot be measured.

4 · Designing for the monitoring problem rather than exhorting people out of it. Molloy and Parasuraman showed in the laboratory that detection of an automation failure degrades with time on task, and that the effect is strongest where the automation has been consistently reliable. Reliability is part of the cause, not the cure.

What does not transfer#

Four honest differences, because the analogy gets stretched.

Aviation has a rare, catastrophic, immediately visible failure mode. Knowledge work has a common, gradual, largely invisible one. Nobody convenes an investigation because a strategy was mediocre. That difference is the whole reason capability loss goes unmeasured in offices and is measured obsessively in cockpits.

Aviation has a simulator. Most professions cannot rehearse the real thing at low cost, which is what makes recurrent testing affordable there and expensive elsewhere.

Aviation has a single regulator per jurisdiction with the power to ground you. Professional services have a patchwork.

And aviation's task is bounded. Flying an approach has a correct answer. Advising a client does not, which makes the proficiency check harder to design and easier to fudge.

What to take anyway#

• Test capability periodically, unaided, on real work. Aviation's insight is not the interval. It is that the check is separate from the doing, and that it can be failed.
• Define the critical phase. Which decisions in your process are the equivalent of below 10,000 feet? Protect those by rule, not by asking people to focus.
• Expect monitoring to fail, and design around it. A reviewer who has approved forty accurate outputs in a row is not being careless on the forty-first. That is what attention does.
• Spend more on the humans as the system gets better, not less. Bainbridge's fourth irony, and the hardest sentence to get past a finance director.
• Do not claim your training works until you have measured behaviour. Aviation measures on the line and still declines to claim an accident-rate effect. Most corporate programmes claim more on far less.

Related SuperSkills research#

On the underlying mechanism, automation complacency and automation bias. On what erodes, capability debt and deskilling. On whether it comes back, can you regain a skill you have lost. On the oversight duty, meaningful human oversight, and on why review is the weakest position, human in the loop is not a safeguard. On testing people rather than output, how to assess capability rather than output.

Key research and primary sources

• Bainbridge, L. (1983). Ironies of Automation. Automatica, 19(6).
• PARC/CAST Flight Deck Automation Working Group (2013). Operational Use of Flight Path Management Systems. Federal Aviation Administration.
• Bureau d'Enquetes et d'Analyses (2012). Final report on the accident to flight AF 447.
• Casner, S. M. et al. (2014). The Retention of Manual Flying Skills in the Automated Cockpit. Human Factors, 56(8).
• 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).
• Molloy, R. and Parasuraman, R. (1996). Monitoring an Automated System for a Single Failure. Human Factors, 38(2).
• 14 CFR 121.441, Proficiency checks, and 14 CFR 121.542, the sterile cockpit rule.

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. The regulations cited are United States federal aviation regulations, quoted from the current Electronic Code of Federal Regulations. European requirements are broadly comparable but are not quoted here, because the primary text could not be confirmed directly. The transfer from aviation to knowledge work is an argument, and the four differences that limit it are set out above rather than left out. Not legal or operational advice.

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