Mode Confusion, the Glass Cockpit Automation Trap the NTSB Has Documented for Thirty Years, and Why the Annunciator Every Pilot Trusts Is the One Most Likely to Get Missed

Mode confusion - when a pilot's mental model of the autopilot's state diverges from reality - has contributed to fatal accidents at every level of aviation for three decades.

Aviation Technology Analyst

Mode confusion is one of aviation’s most persistent and least understood safety problems. It occurs when a pilot’s mental model of the automation’s current state diverges from what the automation is actually doing - and the aircraft behaves in ways the pilot does not anticipate. The NTSB has flagged it as a contributing factor in accidents since the 1990s, and it is not an airline-only problem: it appears in general aviation reports involving glass cockpit aircraft on routine instrument approaches.

What Is Mode Confusion and Why Does It Happen?

Modern autopilots, autothrottle systems, and flight management computers operate as state machines. Each system is always in a specific mode - heading mode, altitude hold, vertical speed, approach mode - and behaves according to the rules of that state. The problem is transition.

These systems move between modes automatically, based on conditions the pilot may not have anticipated. An autopilot captures an altitude and transitions from vertical speed mode to altitude hold. An autothrottle shifts from speed mode to thrust mode during an approach. A flight management system sequences to the next waypoint and arms a new constraint. Each transition is announced - typically by an annunciator changing on the primary flight display, sometimes accompanied by a brief tone.

If the pilot is looking at the right place at the right time, they catch it. If not, they don’t. And in a modern glass cockpit, transitions are frequently happening during the phases of flight when workload is already climbing.

How Long Has the NTSB Been Documenting This?

The NTSB began specifically identifying mode confusion as a contributing factor in accidents in the 1990s, shortly after the first generation of fully glass cockpit airliners entered service. The Airbus A320, certified in 1988, was the first commercial aircraft with a full fly-by-wire system and a truly complex autoflight mode architecture. By the mid-1990s, the NTSB had already flagged mode confusion in connection with several incidents and accidents involving glass cockpit aircraft.

The formal safety community term is automation surprise - a phrase that comes largely from the research of Earl Wiener, a professor at the University of Miami who spent his career studying human factors in aviation. Wiener coined the term “clumsy automation” to describe systems designed to reduce pilot workload that sometimes increase it at exactly the wrong moments: during mode transitions, during the unexpected, at the edges of normal operation.

Wiener’s research identified a consistent pattern. Pilots flying highly automated aircraft performed well during normal, predictable phases of flight. They were worse at the edges - the transitions, the surprises, the moments when automation did something they had not fully anticipated. That pattern has shown up in the accident record again and again.

What Does Mode Confusion Look Like at Its Worst? Air France Flight 447

Air France Flight 447 departed Rio de Janeiro on June 1, 2009, and disappeared over the South Atlantic. The investigation that followed became one of the most cited automation studies in aviation safety literature.

The aircraft flew into convective activity over the Intertropical Convergence Zone. The pitot tubes iced over, rendering airspeed data unreliable. The flight computers automatically transitioned from Normal Law to Alternate Law - a mode change that significantly reduced the envelope protections the crew had been relying on. The autopilot disconnected because it could not trust the airspeed data it required to function.

The crew had approximately four minutes from autopilot disconnect to impact with the ocean.

What the final investigation by the Bureau d’Enquêtes et d’Analyses (BEA) revealed was that the crew never established a shared understanding of what mode the aircraft was in. One pilot was making nose-up inputs. The other was making nose-down inputs at different points. They were working at cross-purposes because nobody had diagnosed the aircraft’s actual automation state.

The BEA was careful not to characterize the crew as careless. The captain had over 11,000 flight hours. What they lacked was adequate training for the specific scenario where the automation’s rules change - where the aircraft responds differently to control inputs in Alternate Law than it does in Normal Law - without a clear and unmistakable warning that this transition has occurred. The safety nets were partially down. The crew did not know it.

The BEA final report runs to 228 pages. Its core recommendation throughout is not about unusual attitude training or systems knowledge in isolation. It is about the specific, learnable skill of maintaining awareness of automation state under workload and pressure.

Why Did Turkish Airlines 1951 Crash on a Clear Approach?

Turkish Airlines Flight 1951 crashed on approach to Amsterdam Schiphol Airport in February 2009. The aircraft was a Boeing 737-900.

One of the radio altimeters failed and fed incorrect altitude data to the autothrottle system. The autothrottle was in retard mode for the approach. When the bad altimeter data told the system the aircraft was closer to the ground than it actually was, the autothrottle reduced thrust to idle at 1,900 feet above the ground - during an instrument approach in progress. The crew did not notice for over a minute. By the time the stick shaker activated, the aircraft was at 1,100 feet with insufficient energy to recover. Nine people died.

The Dutch Safety Board investigation found that the transition to idle thrust happened without an adequate alert. The automation did exactly what it was programmed to do given the data it had. The information the crew needed was present - a small piece of text in the corner of the primary flight display showing the autothrottle mode annunciator. But it was not communicated in a way that cut through the crew’s existing workload and scan pattern.

That is the core human factors failure: the information was displayed. It was not absorbed.

Does Mode Confusion Happen in GA Glass Cockpit Aircraft?

The glass cockpit revolution reached general aviation in the early 2000s. The Garmin G1000 debuted in a certificated aircraft - the Diamond DA40 - in 2003, and spread quickly across the fleet. By 2010, it was difficult to purchase a new certified trainer without a glass panel.

The GFC 700 autopilot, which integrates with the G1000 and later the G3000 suite, is one of the most capable autopilots ever installed in a light aircraft. It has heading mode, navigation mode, approach mode, vertical speed mode, flight level change mode, altitude hold, altitude select, and VNAV. Each behaves differently. Each responds to pilot inputs differently. Transitions happen automatically when capture criteria are met - sometimes triggered by setup the pilot performed earlier and then forgot about.

The pattern that appears repeatedly in NTSB general aviation reports involving glass cockpit aircraft follows a consistent shape. A pilot is flying an instrument approach in marginal conditions with the autopilot engaged. The aircraft gets high on the glideslope. The pilot reduces power to increase the descent rate. But the autopilot is in altitude hold mode, not approach mode. The mode transition the pilot believed had occurred - or intended to trigger - did not happen, or was never confirmed.

The autopilot holds altitude. Airspeed bleeds off. The pilot, fixated on the glideslope deviation, does not catch the energy state until the stall warning activates. The instrument approach environment has generated enough workload to keep the scan away from the mode annunciator at the exact moment it matters.

That is mode confusion. In a Cessna equipped with a G1000. On an approach to a Class D airport on a Tuesday afternoon.

How Do You Monitor Automation State Actively?

The core issue is not complexity - it is expectation management. The automation does exactly what it was programmed to do. It follows its rules precisely. The failure point is the gap between what the pilot believes the automation is doing and what it is actually doing.

Flying a glass cockpit aircraft with a capable autopilot means managing a system that is itself flying the aircraft - not simply flying the aircraft. That requires a different kind of situational awareness: not just what is the aircraft doing, but what mode is the automation in, what will it do next, and what will it do when conditions change.

There is a specific technique that experienced instrument instructors consistently pass along:

When you change any autopilot or autothrottle state, say aloud what mode you just selected. Then read the annunciator to confirm it matches what you said.

This verbal confirmation loop forces verification rather than assumption. It takes two seconds. It is the same closed-loop discipline used in professional flight operations.

The underlying framework is three steps, not one:

  1. Program it. Set the automation to what you intend.
  2. Verify it. Confirm the annunciator matches your intent immediately after.
  3. Monitor it. Continuously confirm it is still doing what you expect as conditions and phases change.

Instrument training has historically emphasized the first step. The human factors research says all three are required.

What Has the FAA Done to Address Automation Over-Reliance?

The FAA addressed the problem directly in 2013 with Safety Alert for Operators 13-0002, which explicitly told airlines that over-reliance on automation was eroding both manual flying skills and automation management skills simultaneously. The FAA was specifically concerned about pilots who could use automation competently in normal conditions but could not manage an automation failure or mode surprise effectively.

Advisory Circular 121-11, issued in 2016, established guidelines for upset prevention and recovery training partly in response to the same concern. The FAA’s Aviation Rulemaking Advisory Committee on flight simulation and training has been pushing for more scenario-based automation training in instrument curricula - specifically scenarios designed to present mode surprises and unexpected automation behavior in controlled environments where pilots can practice recovering from them.

The manufacturers are responding as well. The G3000 architecture found in modern turboprops and business jets simplified the mode annunciation display compared to some earlier systems, with a philosophy of making the current mode and armed modes visually unambiguous at a glance. Collins Aerospace and Honeywell have made similar adjustments to their latest avionics suites.

Will New Avionics Solve the Mode Confusion Problem?

The underlying design principle - the dark cockpit philosophy, where silence means everything is normal and warnings appear only when attention is needed - remains sound. Mode confusion represents a specific failure of that ideal: the annunciators are present, the information is displayed, but it is not communicated in a way the pilot reliably absorbs under workload during the phases of flight when it matters most.

The next generation of avionics systems are experimenting with active state communication - proactively signaling mode transitions with subtle audio cues or momentary display highlights rather than waiting for a pilot to look at the right annunciator at the right moment. Whether this reduces confusion or creates its own form of alert saturation is a research question that has not been fully answered. The direction, however, is correct.

The industry has not solved mode confusion. What it has done is gotten better at training for it, albeit slowly. The skill of maintaining awareness of automation state under workload is not innate, and it is not developed simply by accumulating more hours in glass cockpit aircraft. It has to be deliberately trained, in scenarios designed to surface the edges where automation behavior becomes unexpected.

Key Takeaways

  • Mode confusion occurs when a pilot’s mental model of the autopilot’s state diverges from what the system is actually doing; the NTSB has documented this in accidents since the 1990s.
  • Air France 447 (June 2009) and Turkish Airlines 1951 (February 2009) both demonstrate that experienced crews in capable aircraft can be caught by unanticipated mode transitions - with fatal results.
  • General aviation glass cockpit aircraft present the same risk on routine instrument approaches; NTSB reports document the scenario repeatedly in aircraft like the Cessna with the G1000.
  • The core discipline is program → verify → monitor: set the mode, confirm the annunciator immediately, and continuously track automation state throughout the flight.
  • Saying the mode aloud and reading the annunciator to confirm is the simplest, most effective single habit for closing the automation awareness loop.

Radio Hangar. Aviation talk, built by pilots. Listen live | More articles