FAA takes its AI traffic adviser nationwide, raising questions about accountability

The system, known as SMART, does not fly aircraft or replace air traffic controllers. It studies large volumes of information and recommends ways to sequence arrivals, adjust routes and manage traffic. The final decisions remain with human officials.

That distinction is central to the FAA’s case for the program. It is also at the heart of the debate over whether the agency has provided enough evidence to justify expanding an experimental tool into a safety-critical national system.

Ars Technica reported that the FAA is moving SMART from an initial deployment in the Washington, DC, region to all 22 air-route traffic-control centers in the country. The expansion comes as airlines and passengers face persistent delays, high fuel costs and staffing pressure inside the air traffic system.

An adviser, not an autopilot

The FAA describes SMART as a forecasting and recommendation tool. Its September announcement says the system draws on roughly 200 data streams, including weather, flight paths, traffic conditions and controller staffing information.

It then looks for patterns that may not be obvious when traffic managers examine each problem separately. SMART can suggest a different order for incoming aircraft, recommend route changes or identify ways to distribute demand more evenly across the system.

The FAA’s SMART fact sheet says the program is intended to improve traffic-flow management while preserving the existing structure of human oversight. The system cannot take control of an aircraft, and it cannot substitute for the professional judgment of controllers and traffic managers.

Under the arrangement described by the agency, a specialist team in Washington reviews SMART’s recommendations before sending them to facilities. Controllers at those facilities can accept or reject the suggestions. In practice, that makes SMART less like an autonomous pilot and more like a highly specialized adviser whose recommendations enter an established chain of decisions.

The FAA says that model has already produced early results. After its September 21 rollout in the National Capital Region, SMART helped reroute 180 flights, with an average saving of 10 minutes and nearly 80 gallons of fuel per flight, according to the agency’s announcement.

The system has since expanded to centers serving Atlanta, Jacksonville and Miami. The FAA has also cited a case in which SMART recommended a shorter route for a cargo flight headed to Dallas-Fort Worth. The agency has not identified the flight or released detailed evidence showing how the system has affected passenger traffic.

The missing test is public confidence

Those early examples may show that the tool can identify useful options. They do not, by themselves, establish how often its recommendations are correct, how the system performs during unusual weather or emergencies, or whether its benefits hold across different regions and traffic patterns.

That is where critics have focused their objections.

Rep. Don Beyer and five other lawmakers said the FAA should pause the Washington-area deployment. They cited the absence of independently verified testing, limited consultation with controllers, inadequate training and insufficient public transparency. They also raised unresolved questions about accountability if a recommendation contributes to a bad decision.

Their concern is not simply that an algorithm might make a mistake. Human professionals also make mistakes. The harder question is what happens when a controller relies on a system that the FAA has encouraged facilities to use, then something goes wrong. If the recommendation was advisory, responsibility may remain with the human decision-maker. Yet if the system becomes embedded in daily operations, the line between advice and institutional expectation can become less clear.

The issue also involves the workforce that must use the tool. Controllers will need to understand what SMART is optimizing, what information it may overlook and when its recommendations should be rejected. Without transparent performance measures and meaningful training, human oversight can become a formal safeguard rather than an effective one.

Efficiency with a human responsibility

The aviation industry is not uniformly opposed to using AI. The Airline Dispatchers Federation says it supports artificial intelligence that augments rather than replaces professional judgment. It also stresses that certificated aircraft dispatchers must retain authority over operational decisions.

That position points to a possible path for SMART. An advisory system could reduce routine congestion while allowing experienced professionals to concentrate on exceptions, conflicting signals and rapidly changing conditions. If the recommendations are accurate and understandable, the technology could help the system absorb more traffic without simply asking exhausted workers to make decisions faster.

But that promise depends on more than fuel savings and individual success stories. A national rollout requires independent evaluation, clear measures of delay reduction and safety, a published training plan and an explanation of who is accountable when humans accept flawed advice.

SMART is therefore not a story about machines taking over American airspace. It is a story about how algorithmic judgment enters a workplace where human responsibility cannot be delegated so easily. The FAA has supplied a case for trying the system. The next test is whether it will supply enough evidence, transparency and worker involvement to show that the technology deserves to shape decisions nationwide.