The Drift Part Two: What Can We Actually Do About It

The instinct when confronted with a problem like Normalization of Deviance is to reach for a solution. A new procedure. A better checklist. A training module. More oversight. The uncomfortable truth is that none of those things fix it. They manage it, imperfectly, if the conditions are right. Understanding what can realistically be done requires first being honest about what cannot.

Normalization of Deviance cannot be engineered out of human beings. This is not pessimism. It is the clear-eyed starting point that Vaughan's work demands. The process she documented at NASA was not the result of inadequate training, insufficient checklists, or weak safety culture. NASA in 1986 had all of those things. It had some of the most technically sophisticated and safety-conscious engineers in the world. It had procedures, review boards, documentation requirements, and a culture that took safety seriously. Nonetheless, because an O-ring deviation had been normalized over years of successful launches, the result was the death of a crew of seven shortly after launch.

If the solution were simply "follow the procedures," Challenger would not have happened. Normalization of Deviance is the process by which "follow the procedures" gets quietly redefined to mean "follow the procedures as we currently understand them to apply" — which is a different thing, and a moving target.

What can be done is not elimination. It is detection, friction, and culture — applied consistently, without ever assuming the problem has been solved.

What Training Can Actually Do

The first thing training can do is name the mechanism explicitly.

Most pilots learn about complacency. Far fewer learn about its cousin: Normalization of Deviance as a distinct and specific process. A process  that one operates not through carelessness but through the gradual reclassification of deviant behavior as acceptable. The difference matters enormously. Complacency implies inattention. Normalization of Deviance can occur in highly attentive, highly experienced pilots who have simply recalibrated their standard over time without noticing. Telling a pilot to "stay vigilant" does not address a process that feels, from the inside, like nothing is wrong.

Teaching student pilots that the drift is normal, that it happens to experienced pilots more than inexperienced ones, and that uneventful outcomes are not evidence of adequate safety margin. It does not require a new curriculum. It requires honesty about what the research shows.

The second thing training can do is address the transmission problem. Students absorb the informal curriculum alongside the formal one. A CFI who flows from memory instead of reading checklists, who treats fuel sump checks as optional in good weather, who talks about regulations as bureaucratic obstacles rather than safety margins, is teaching Normalization of Deviance before the student has the experience to recognize it. The content of training matters. The behavior of the instructor during training matters more. You cannot teach a standard you do not personally hold.

This is not a small problem. It is how deviance propagates across generations of pilots without anyone intending to propagate it. The instructor who skips steps was taught by an instructor who skipped steps. The informal curriculum has its own lineage.

The third thing is where training intersects directly with the next generation: automation management. Student pilots entering training today are learning on glass cockpits with autopilots, moving maps, terrain warning systems, and traffic alerts. The automation is genuinely helpful. However, it also produces pilots who have never experienced the workload and attention demands of flying without it. Their baseline for what normal feels like is already buffered by layers of technology they have never had to fly without. (See part 1!)

That baseline is where their drift starts.

Training needs to deliberately impose conditions where automation is absent or degraded,  not to impose the “back in my day” mentality on students, but to ensure the pilot's baseline competence does not require the system to be working perfectly. A pilot who has only ever flown with a functioning autopilot does not know what they do not know. And what they do not know will become relevant at the worst possible moment.

Former two-time Secretary of Defense and Navy pilot Donald Rumsfeld famously said, “Reports that say that something hasn't happened are always interesting to me, because as we know, there are known knowns; there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns—the ones we don't know we don't know....it is the latter category that tends to be the difficult ones.”

What Design Can Do

Let's return to Bedford for a moment.

The QAR that revealed 98% non-compliance across 175 takeoffs existed on that aircraft before the accident. The data was there. The pattern of deviation was recorded and stored. Nobody was looking at it systematically until after seven people were dead. That is not an aircraft design failure. It is a monitoring infrastructure failure. And it is the gap that design can most realistically address.

Digital flight data monitoring programs like Flight Operations Quality Assurance (FOQA) in airline operations and other similar systems are increasingly available to corporate and charter operators and present an opportunity to resolve this.

Yet, they do not prevent individual deviations. They detect patterns of deviation that indicate drift has occurred, creating the opportunity to intervene before the margin runs out entirely. After the Bedford accident, the National Business Aviation Association (NBAA) analyzed 143,756 flights and found that crews performed no flight control check whatsoever on more than 2% of departures, and only a partial check on an additional 15%. That data existed. The question is whether anyone was positioned to act on it before the accident rather than after.

The gap in general aviation is significant and largely unaddressed. The GA pilot who has been skipping preflight checks has no equivalent monitoring system. Their drift is invisible to everyone, including themselves. The feedback that would alert them that a standard has been abandoned is absent. 

Better design can raise the cost of specific deviations. A checklist application that requires task completion rather than acknowledgment. A system that flags when a preflight flow runs faster than a physically complete inspection possibly could. These tools exist in varying forms. Their adoption is incomplete, their use is uneven, and their effectiveness depends entirely on whether the pilot using them is engaging genuinely or tapping through.

Which leads to the honest limit of design: the gap between checking the box and doing the thing is exactly where normalization of deviance lives. No interface can close that gap. It can only make the gap more visible to those willing to look at it.

What Culture Does That Nothing Else Can

The most durable defense against Normalization of Deviance is a culture where deviation is surfaced rather than concealed.  (This is something that medicine, Clinical Peer Review, only really started to adopt when the discussions were considered protected from discovery in liability cases.)

Where a pilot who notices a colleague skipping steps says something.

Where a junior crew member who observes a captain cutting corners feels empowered to challenge it. Where near-misses and small deviations are reported because the organization treats them as information rather than evidence of failure. Where the conversation after a long day includes "I felt pressure to rush the preflight" rather than just "The flight went fine."

This is harder to build than a procedure and easier to destroy than a regulation.

The conditions that enable normalization (production pressure, time scarcity, the desire to just get the job done, the reluctance to be the person who slows everything down) exist at every level of aviation and do not disappear when the safety briefing ends. They existed at NASA. They existed at American Airlines maintenance in 1979. They existed in the cockpit of that Gulfstream IV for years before the accident. They exist in every flight school, every charter operation, every airline, every weekend flyer who has somewhere to be and a weather window that won't last.

A culture that acknowledges this honestly is more useful than one that pretends professional pilots are immune to the pressures that affect every other human being. The pilot who admits they felt pressure to depart before the weather window closed, and chose not to go, is doing something more valuable than the pilot who says they never felt the pressure at all. Acknowledging the pull is not a weakness. It is the prerequisite to resisting it.

The Next Generation Question

The deviations that will emerge from the next generation are not fully predictable. But the mechanism that will produce them is identical to the one that produced Bedford, American 191, and Challenger. Technology changes. The human process does not.

What the next generation needs is not just competency in the new systems. They need a stable internal standard that does not drift with the automation. A personal baseline that holds even as the aircraft becomes more capable. The ability to ask, on every flight, not "Is the system handling it?" but "Do I understand what the system is doing, and could I continue without it?"

Moreover, they need instructors who model this. Not in the safety briefings they give, but in the preflight checks they actually complete, the checklists they actually read, the margins they actually hold when there is no one watching and it would be easier not to. Because the informal curriculum is always running. It was running when your instructor trained you. It is running when you train someone else. The standard you actually hold, not the one you teach, is what gets passed on.

The Honest Conclusion

There is no clean solution to Normalization of Deviance. Vaughan did not offer one. The accident record does not suggest one exists.

What the research does suggest is that organizations and individuals who understand the mechanism are harder to affect by it. Awareness does not eliminate the drift, but it slows it. Explicit standards do not prevent normalization, but they create reference points from which drift can be measured. Cultures of transparency do not stop deviations, but they surface them early enough that the margin has not yet run out.

None of this is satisfying in the way a technical fix is satisfying. The problem is not technical. It is human, running through systems that humans built, maintained, and fly.

The pilot who has successfully skipped the preflight fuel check a hundred times has not demonstrated the check is unnecessary. They have demonstrated only that contaminated fuel was not present on those hundred occasions.

You made it to the airport this morning. The stop sign, the speed, the lane change — none of it caught up with you. It rarely does. The margin erodes in the silence between the deviation and the consequence.