Fwd: Air Canada Flight 8646 and Artificial Intelligence
Chandana Liyanapatabendy
Air Canada Flight 8646 and Artificial Intelligence
The most immediate recommendation is the universal implementation of transponder or equivalent tracking devices on all ground vehicles operating within the movement area of an airport, without exception.
by Dr. Ruwantissa Abeyratne
March 29, 2026
9 mins read
NTSB investigators at the scene of the collision
The Facts
The tragedy of Air Canada Express Flight 8646 collision stands not merely as an isolated aberration in the annals of aviation mishaps, but as a profound epistemic rupture—an event that compels a re-examination of how intelligence, both human and artificial, is orchestrated within the delicate ecosystem of air navigation and ground movement. The collision between a landing aircraft and a fire truck, each ostensibly cleared to occupy the same runway space, reveals a systemic failure of anticipatory cognition rather than a mere lapse in procedural adherence. It is here that artificial intelligence, if properly conceived and deployed, could serve not merely as a tool, but as a form of anticipatory jurisprudence—an arbiter of probabilities that forestalls catastrophe before it manifests.
The known facts are stark. The aircraft, on final approach, was cleared to land while a fire truck responding to another emergency was cleared to cross the same runway. Communications were issued, yet comprehension failed. Surveillance systems did not generate timely alerts, partly because the vehicle lacked a transponder, rendering it invisible to certain detection systems. The controller’s desperate exhortation—“stop, stop, stop”—came too late, encapsulating the tragic latency between human reaction and unfolding reality.
Anticipatory Intelligence
This latency is precisely where artificial intelligence finds its raison d’être. Anticipatory intelligence, as distinct from reactive intelligence, does not wait for conflict to arise but predicts the confluence of trajectories—physical, procedural, and cognitive—that may culminate in danger. AI systems, trained on vast datasets of aircraft movements, human communications, and historical incidents, could continuously compute the probabilistic intersection of runway occupancy, issuing preemptive alerts not merely when a conflict is imminent, but when it becomes statistically plausible.
In LaGuardia, an AI-enabled ground surveillance system would not have depended solely on transponder inputs. By integrating computer vision, radar fusion, and predictive modeling, it could have identified the fire truck’s movement as anomalous relative to the aircraft’s approach path. Even in the absence of a transponder, pattern recognition algorithms could have flagged the vehicle’s trajectory as intersecting with an active runway clearance, triggering an automated “conflict veto”—a system override preventing simultaneous clearances. Such a system would embody what may be termed “direction setting intelligence,” wherein AI does not merely inform human operators but constrains the decision space within safe parameters.
Direction Setting
The concept of direction setting is particularly salient in the domain of air traffic control, where cognitive overload and divided attention are endemic. The LaGuardia incident reportedly involved limited staffing and concurrent emergencies, conditions under which human controllers are prone to attentional tunneling. AI, by contrast, does not fatigue, nor does it prioritize one stream of information at the expense of another. By continuously synthesizing inputs from aircraft, vehicles, weather systems, and communication channels, AI could provide controllers with a dynamically updated “safety envelope,” within which all clearances must reside. Any instruction that would breach this envelope would be flagged, delayed, or automatically rejected.
Learning by Puzzles
This brings us to the notion of “learning by puzzles,” a pedagogical paradigm wherein AI systems are trained not merely on historical data but on simulated scenarios that resemble complex, multi-variable puzzles. Aviation incidents, particularly runway incursions, are rarely the result of a single error; they are emergent phenomena arising from the interaction of multiple agents and conditions. By framing these interactions as puzzles—dynamic systems with incomplete information—AI can be trained to recognize patterns of convergence that humans might overlook.
Consider, for instance, the near-catastrophic event of Air Canada Flight 759 near-miss, where an aircraft lined up with a taxiway occupied by multiple aircraft instead of the intended runway. Or the 2023 near-collision at Austin-Bergstrom, where a cargo aircraft and a departing passenger jet came within a mere 150 feet of each other. These incidents, though varied in context, share a common structure: a breakdown in situational awareness compounded by imperfect communication and environmental ambiguity. Each represents a “puzzle” that, if encoded into AI training models, could enhance the system’s ability to anticipate similar configurations in real time.
The Airport
The integration of AI into airport ground management would necessitate a reconfiguration of existing systems such as ASDE-X (Airport Surface Detection Equipment). While current systems provide surveillance and alerting capabilities, they are fundamentally reactive, issuing warnings only when predefined thresholds are crossed. AI, by contrast, could operate on a continuum of risk, assigning probabilistic scores to evolving scenarios and issuing graded alerts that escalate as risk increases. In the LaGuardia case, the absence of a transponder on the fire truck rendered it invisible to ASDE-X, thereby nullifying its alerting function. An AI-enhanced system, however, could have compensated for this deficiency by integrating alternative data sources, such as video feeds and motion sensors, thereby maintaining situational awareness even in the face of equipment limitations.
Furthermore, AI could facilitate a form of “collaborative cognition” between air traffic controllers and ground personnel. Natural language processing systems could analyze radio communications in real time, detecting ambiguities, overlaps, or contradictions in instructions. For example, if a controller issues a landing clearance while simultaneously authorizing a vehicle to cross the same runway, the AI could flag the inconsistency, prompting immediate clarification. This would transform communication from a linear exchange into a monitored, validated process, reducing the likelihood of misinterpretation.
The role of AI in direction setting extends beyond immediate operational contexts to encompass strategic planning and policy formulation. By analyzing patterns of runway incursions across multiple airports and timeframes, AI could identify systemic vulnerabilities—such as staffing shortages, procedural ambiguities, or equipment deficiencies—and recommend targeted interventions. The LaGuardia incident, for instance, has already highlighted issues related to controller workload and the lack of transponder equipment on emergency vehicles. AI could aggregate such insights across jurisdictions, providing regulators with a data-driven basis for revising standards and protocols.
It must also be acknowledged that the deployment of AI in aviation raises profound legal and ethical questions, particularly in the context of liability. If an AI system fails to prevent an accident, where does responsibility lie? With the designers of the algorithm, the operators of the system, or the regulatory bodies that approved its use? These questions resonate deeply with the principles of international air law, particularly those enshrined in liability regimes such as the Montreal Convention. The introduction of AI as an active participant in decision-making processes may necessitate a redefinition of concepts such as “fault” and “negligence,” extending them to encompass algorithmic behavior.
Yet, despite these complexities, the imperative for integrating AI into aviation safety systems is undeniable. The increasing density of air traffic, coupled with the growing complexity of airport operations, renders purely human-centered systems increasingly untenable. AI offers not a replacement for human judgment, but an augmentation—a means of extending the cognitive horizon of controllers and ground personnel, enabling them to perceive and respond to risks that would otherwise remain latent.
In reflecting upon the LaGuardia tragedy, one is reminded that aviation safety has historically advanced through the crucible of catastrophe. Each accident, however tragic, has yielded lessons that have been codified into improved practices and technologies. The challenge, in the age of artificial intelligence, is to transcend this reactive paradigm and embrace a proactive ethos—one in which accidents are not the catalysts for change, but the anomalies that are precluded by design.
Artificial intelligence, with its capacity for anticipatory reasoning, directional guidance, and puzzle-based learning, offers a pathway toward such an ethos. It enables a reimagining of aviation safety not as a static set of rules, but as a dynamic, adaptive system—one that evolves in response to emerging risks and continuously refines its understanding of the complex interplay between human and machine.
In the final analysis, the lesson of the LaGuardia collision is not merely that errors were made, but that the system within which those errors occurred lacked the capacity to foresee and forestall their convergence. AI, if judiciously integrated, could provide that capacity, transforming the skies—and the runways beneath them—into domains not only of movement, but of mindful anticipation.
My Take
The lamentable collision at LaGuardia is not merely an operational failure but a philosophical indictment of the way we continue to conceive aviation safety—as a reactive architecture rather than a predictive enterprise. We have long relied on layered defenses, redundancy, and human vigilance, all of which have served aviation admirably. Yet, the complexity of modern airport ecosystems has outpaced the human capacity to synthesize, anticipate, and act within shrinking temporal margins. What occurred was not simply a breakdown in communication or surveillance; it was a failure to anticipate the convergence of risk. In this context, artificial intelligence should not be viewed as an optional enhancement, but as an indispensable instrument of foresight.
The first and most immediate recommendation is the universal implementation of transponder or equivalent tracking devices on all ground vehicles operating within the movement area of an airport, without exception. It is both astonishing and disquieting that, in an age where aircraft are tracked with exquisite precision, critical ground vehicles may still operate in relative invisibility to surveillance systems. This lacuna must be closed through regulatory mandate, enforced by national aviation authorities and harmonized through international bodies such as ICAO. The absence of such equipment is not a technical oversight; it is a systemic vulnerability.
Second, airport surface surveillance systems must evolve from their current reactive posture to anticipatory intelligence. Existing platforms, while technologically sophisticated, are constrained by rule-based alerting mechanisms that activate only when a conflict is imminent. What is required is a probabilistic framework—an AI-driven system that continuously assesses trajectories, velocities, and clearances, generating predictive alerts well before a hazardous intersection materializes. Such systems should be capable of issuing graduated warnings, escalating in urgency as the probability of conflict increases, thereby affording controllers the temporal latitude to intervene decisively.
Third, the concept of “clearance integrity” must be redefined through the integration of AI as a real-time validator of air traffic control instructions. Every clearance—whether to land, take off, or cross a runway—should be subjected to instantaneous algorithmic scrutiny. If a proposed instruction conflicts with existing or projected movements, the system should not merely alert the controller but, where necessary, inhibit the issuance of the clearance until the conflict is resolved. This introduces a paradigm shift from advisory systems to directive safeguards, wherein AI acts as a guardian of procedural coherence.
Fourth, communication—long regarded as the lifeblood of aviation safety—must be augmented through natural language processing technologies capable of detecting ambiguity, contradiction, and omission. Controllers and pilots operate within a highly standardized linguistic framework, yet deviations, overlaps, and cognitive slips do occur, particularly under conditions of stress or workload saturation. AI systems could monitor radio exchanges in real time, flagging inconsistencies such as simultaneous clearances for incompatible movements or incomplete readbacks. This would transform communication from a passive exchange into an actively supervised process.
Fifth, there must be a deliberate and sustained investment in simulation-based training that leverages the concept of “learning by puzzles.” Traditional training methodologies, while rigorous, often rely on predefined scenarios that may not capture the fluid and emergent nature of real-world incidents. AI-driven simulators could generate dynamic, multi-variable scenarios that challenge controllers and ground personnel to navigate complex, evolving situations. These “puzzles” would not have a single correct solution but would instead cultivate adaptive thinking, pattern recognition, and anticipatory judgment. Importantly, such systems could also learn from the responses of trainees, refining their scenarios to address observed weaknesses.
Sixth, airport operations must embrace a model of collaborative cognition, wherein human and artificial intelligence operate not in parallel but in symbiosis. Controllers should not perceive AI as an intrusive overseer but as an extension of their cognitive faculties. This requires careful interface design, ensuring that AI-generated insights are presented in a manner that is intuitive, actionable, and non-disruptive. Trust, in this context, is paramount; it must be earned through reliability, transparency, and demonstrable value.
Seventh, regulatory frameworks must be recalibrated to accommodate the integration of AI into safety-critical functions. This includes the development of certification standards for AI systems, protocols for their validation and verification, and mechanisms for accountability in the event of failure. The legal architecture of aviation, particularly in relation to liability, must evolve to address the role of algorithmic decision-making. This is not a trivial undertaking, but it is a necessary one if AI is to be entrusted with responsibilities that bear directly on human life.
Eighth, there must be a global effort to create and maintain comprehensive databases of runway incursions and surface incidents, enriched with detailed contextual information. These datasets should be made accessible for training and refinement of AI systems, subject to appropriate privacy and security safeguards. Each incident, whether catastrophic or narrowly avoided, constitutes a repository of lessons that can inform future prevention. The institutionalization of such knowledge is essential to the development of truly anticipatory systems.
Ninth, contingency protocols for emergency vehicle operations on active runways must be rigorously reviewed and standardized. While the urgency of emergency response is unquestionable, it must not compromise the integrity of runway operations. AI could play a pivotal role in coordinating such movements, ensuring that emergency vehicles are granted access only within a dynamically assessed safety envelope. This would reconcile the imperative of rapid response with the equally critical requirement of collision avoidance.
Finally, there must be a cultural shift within the aviation community—an acknowledgment that the future of safety lies not solely in human expertise, but in the intelligent augmentation of that expertise. This is not a diminution of the human role, but its elevation. By embracing AI as a partner in safety, we extend our capacity to foresee, to decide, and to act.
The LaGuardia incident should not be remembered solely as a tragedy, but as a turning point. This moment compels us to transcend the limitations of our current paradigms and to embrace a more anticipatory, intelligent approach to aviation safety. The recommendations outlined herein are neither speculative nor unattainable; they are practical measures grounded in existing technologies and informed by lessons from past experiences. What is required is not invention, but implementation; not aspiration, but action.
Dr. Ruwantissa Abeyratne DCL, PhD, LL.M, LLB, FRAeS, FCILT
Senior Associate, Aviation Law and Policy
Aviation Strategies International
This message (including any attachments) is sent without prejudice to any rights and privileges as shall legally accrue to ASI, and may contain privileged or confidential information that should not in any way be compromised, duplicated or misused in any manner whatsoever. Any disclosure, copy, distribution or any inappropriate action taken relying upon the content of this information is strictly prohibited. Any views or opinions expressed do not necessarily represent those of ASI. ASI accepts no liability for the content of this email, or for the consequences of any actions taken on the basis of the information provided. / Ce message (incluant ses pièces jointes) est envoyé sans préjudice aux droits et privilèges légaux d’ASI, et peut contenir des informations privilégiées et confidentielles ne devant pas être compromises, copiées ou utilisées de quelque manière que ce soit. Quelconque divulgation, copie, distribution ou action inappropriée liée à son contenu est strictement interdite. Les opinions y étant exprimées ne représentent pas nécessairement celles d’ASI. ASI n’accepte aucune responsabilité quant au contenu de ce message ou pour les conséquences des gestes posés sur la base des informations y étant fournies.