Aviation Professionals Were Asked How They Feel About AI. The Answer Was “4.4 out of 7.”
When regulators talk about artificial intelligence in aviation, the conversation usually stays technical: model robustness, data quality, certification pathways, learning assurance. All necessary. But there’s a second question no amount of engineering rigour answers — will the pilots, controllers, and technicians who work alongside these systems actually accept them?
The European Union Aviation Safety Agency went and asked. Its report Ethics for AI in Aviation collects answers from 231 aviation professionals, and the result is more interesting than a yes or no. It’s a shrug.
How the survey worked
EASA’s AI Programme started looking beyond the purely technical dimension around 2019, in step with the European Commission’s High-Level Expert Group on AI. The framing it settled on treats AI as a tool rather than a moral agent in its own right — technology that helps people do what they couldn’t otherwise do, and that reshapes the work and the workers in the process.
The ethical anchor is the EU Charter of Fundamental Rights, particularly the third generation of rights covering data protection, bioethics, and transparent administration. Nine concepts came into the survey design from there, among them fairness and non-discrimination, labour protection, privacy, professional development, the right to make decisions, transparency, and accountability.
Rather than asking about these abstractions directly, EASA embedded them in eight hypothetical scenarios spanning the flight deck, maintenance, air traffic control, and aerodromes. Respondents were placed inside each story as the person affected, then rated it on a seven-point scale across three dimensions: comfort (feeling relaxed and free from tension), trust (believing the thing is safe and reliable), and acceptance (being willing to agree to it).
The survey ran three weeks and closed in January 2024. To qualify you had to be an aviation professional with some link to AI. The sample skews senior: 77.5% senior professionals, most aged 40 to 59, four fifths male, about 80% in technical areas of industry and 17% at national aviation authorities. Three quarters work directly with AI systems. They rate their own understanding of AI in aviation as good, their teams’ as merely sufficient.
The shrug
Averaged across all eight cases and all three dimensions, the score was 4.4 out of 7. On a scale where 4 is dead centre, that is a lean toward the positive so slight it barely registers.
The ordering is telling. Comfort came highest at 4.58, acceptance next at 4.34, trust lowest at 4.28. People are more at ease with these systems than they are convinced by them. There’s no polarisation, no enthusiasts squaring off against refuseniks — just a large body of experienced professionals who haven’t made up their minds, and say so. EASA reads it as a gap that explainability, demonstration, and hands-on experience still have to close.
Where the discomfort concentrates
The spread between scenarios is where the real signal is.
The worst-scoring case by some distance was deskilling. Respondents were put in 2035 as a controller supervising a fully automated separation system that handles conflicts on its own and needs human intervention roughly once a year — until a cyberattack takes it down during heavy daytime traffic. Comfort scored 3.62, the lowest single figure in the study. Controllers are convinced that if they don’t practise the job they will lose the ability to do it, and that when the technology fails they won’t be able to recover by hand.
Second lowest was physiological monitoring at 4.18, with trust at 3.86. The scenario: single-pilot operations, with a ground-based AI assistant reading cardiorespiratory signals, brain and electrodermal activity, body temperature, and eye movement to gauge your cognitive workload. Concerns clustered around personal invasion, professional threat, psychological effects, and the data itself — several respondents said this category of data should never be shared with an AI system at all.
Both low scorers have something in common: each puts an AI system in direct contact with human work performance, either replacing it or assessing it.
At the other end, airport terminal allocation scored highest at 4.88, with comfort at 5.17 — the highest figure anywhere in the study. Here an AI system optimises gate assignments for connecting-passenger walking times, with an acknowledged risk of skewed behaviour from training on one airport’s dominant domestic carrier. Bias and market-fairness concerns came up, but didn’t dent comfort much. The distinguishing feature is obvious: it isn’t a safety case.
The rest landed in the middle — maintenance sign-off assisted by a drone-scanning traffic-light system (4.36), speech recognition monitoring controller–pilot exchanges (4.43), automatic go-around initiation (4.46), AI-driven crew rostering built on personal and family data (4.56), and human–AI teaming in a control centre (4.71).
The one thing everyone agreed on
Uncertainty about ethics did not translate into uncertainty about governance. Support for regulating these situations was overwhelming and consistent. Teaming with AI and physiological monitoring both cleared 90%. Crew rostering, speech recognition, and deskilling sat between 80% and 90%. Even the least-demanded case, terminal allocation, drew 80%. On who should hold the pen, EASA was named first in at least 52% of responses across every case, with national authorities in a supporting role.
Responsibility and accountability produced the sharpest numbers in the report. Asked who bears responsibility when things go wrong, respondents put the deploying organisation first at 34% and the human end user last at 13%. For accountability the gap widens: deployer 39%, end user 6%. The people at the sharp end are saying plainly that they should not absorb the liability for a system somebody else chose to install.
Why the non-acceptance matters most
Two thirds of respondents — 171 of 231 — rejected at least one scenario outright. EASA asked why, and coded 2,395 individual comments. Roughly 30% concerned the AI systems themselves: performance, poorly defined operational domains, missing mitigations, plain mistrust. Another 28% concerned harm to the humans using them — privacy, professional threat, psychological impact, perceived unfairness. Data handling accounted for 11%, direct threats to safety 6%.
Buried in the competencies question is the finding I’d flag to anyone deploying these systems. Alongside the expected asks for AI literacy, data skills, and IT knowledge, respondents named emotional intelligence as a genuinely new requirement — and were specific: emotion regulation, coping with boredom, assertiveness, building trust in a machine, resilience, handling AI disruptions, and the mental strength to work in an environment where relationships have been partly dehumanised.
Nobody trains for that yet.
What happens next
EASA is folding these findings into Issue 03 of its AI Concept Paper and into rulemaking task RMT.0742 on AI trustworthiness, using a two-step approach: a preliminary ethics-based risk assessment, followed by mitigation where the assessment demands it. A second survey aimed at the general public is planned.
The strategic point is simple enough. A system can be technically robust, formally certified, and still fail — because the controller doesn’t trust it, the pilot resents being measured by it, or the technician won’t sign the release. EASA’s report is an argument that ethics belongs in the design phase, not the communications plan.