The question posed—whether aviation's posture toward artificial intelligence has shifted since 2022—reflects a genuine inflection point in the industry rather than a settled matter. In the years since large language models and machine learning tools became mainstream, aviation has moved from a posture of near-total avoidance to selective, heavily governed adoption in specific domains. The shift is not uniform: flight operations, dispatch, and anything touching the certificated cockpit environment remain conservative by design, while back-office functions—maintenance planning, crew scheduling optimization, weather interpretation, and predictive analytics—have seen meaningfully more AI integration. The distinction matters because it explains why a former dispatcher asking this question would get a mixed answer depending on which corner of the industry they're looking at.
For working pilots and flight operations personnel, the practical changes are incremental rather than revolutionary. Airlines and business aviation operators have expanded use of machine learning in predictive maintenance (flagging component failures before they occur), fuel optimization algorithms, and turbulence/weather forecasting tools that ingest far more data than legacy models. Dispatch and flight planning software increasingly uses AI-driven route optimization that accounts for winds, restricted airspace, and fuel burn in ways that go beyond older deterministic systems. Some carriers have also begun experimenting with AI-assisted training analytics—reviewing simulator data or line operations safety audit (LOSA) data to spot trends in pilot performance or procedural deviations. None of this replaces pilot-in-command authority or dispatcher judgment; rather, it's positioned as decision-support, similar to how EFBs and TCAS became normalized tools rather than autonomous decision-makers.
The caution that existed in 2022 hasn't disappeared so much as it's been channeled into regulatory and procedural frameworks. The FAA, EASA, and ICAO have all issued guidance or working papers emphasizing that AI/ML systems used in safety-critical functions require explainability, traceability, and rigorous certification pathways—concepts that don't map cleanly onto how most modern AI models (especially deep learning and generative AI) actually work as "black boxes." This has produced a bifurcated industry response: enthusiastic adoption in non-safety-critical administrative and analytical functions, and extreme caution—bordering on prohibition—in anything that touches certified avionics, autopilot logic, or dispatch release authority. Boeing, Airbus, and avionics manufacturers like Garmin and Honeywell are all researching AI-enhanced systems (voice-command interfaces, anomaly detection, single-pilot operations research), but none of this has reached operational deployment in transport-category aircraft cockpits.
The broader trend worth watching is the growing pressure toward reduced-crew and single-pilot operations, where AI is being framed as an enabling technology rather than a novelty. Airbus's DragonFly program and various NASA/industry single-pilot ops studies explicitly lean on AI/automation to justify removing a crewmember, which has generated significant pushback from pilot unions (ALPA in particular has been vocal) precisely because the safety case rests on AI performing functions traditionally reserved for a second set of human eyes. This is the area where the "stay away from AI" sentiment from 2022 persists most strongly among line pilots, even as management and OEMs push adoption forward. For a former dispatcher-turned-IT professional, the honest answer is that AI has quietly become embedded in the industry's operational infrastructure—scheduling, maintenance, weather, training analytics—while remaining largely absent, and actively resisted, in the cockpit itself. That gap between back-office enthusiasm and front-line caution is likely to define the next several years of aviation AI policy debates.