SMART’s airport test puts human override before AI control

What SMART is supposed to do

Speaking to News Nation, Yiri said specialists still lack a basic definition of SMART’s role. The answer will determine how its flight-safety risks should be assessed.

The system could remain a forecasting tool. Or it could take on a more prescriptive role, proposing how air-traffic conflicts should be handled. At an airport, that would mean operating against conditions that change continuously rather than against a fixed dataset.

Yiri is skeptical of SMART’s usefulness until its intended use is clear. His caution is practical: the closer the system gets to directing live traffic, the less room there is for uncertainty about what it is allowed to decide.

The failure mode is not obscure

Asked about the main danger, Yiri gave one answer: “hallucinations.” AI systems can produce inaccurate or unverified information, making the provenance of their data a safety issue before the system ever reaches an airport.

That last condition is the clearest boundary in the proposal. Yiri said the authority to intervene should belong to the controller on duty, not to a remote process or an abstract system rule.

Start in a simulator

The planned first launch is not, in Yiri’s view, a reason to put SMART into live operations. He said the system should first be tested in a simulator using scenarios as close as possible to real airport conditions.

Until those trials happen, he does not think there is a basis for relying on SMART’s information in real time. Specialists still have too many unanswered questions, and Yiri’s point is that they should not be answering them while an actual airport situation is unfolding.

I think that makes the simulator less a cautious extra than the real test of the project. If SMART cannot demonstrate reliable behavior under realistic scenarios without controlling live traffic, the “soft launch” says little about whether it belongs in the operational system at all.

The timing raises the stakes

The plan emerged after a major aviation communications failure on Monday. Near Philadelphia, both the primary telecommunications line and its backup failed at the same time, leaving hundreds of aircraft on the East Coast delayed for several hours.

That outage sharpened an existing concern about the United States’ weakened air-traffic-control system and an overburdened workforce. Adding an experimental AI system to that environment could create another dependency before the underlying operating conditions are stable.

The announcement is quiet about the question that matters most: what exactly happens when SMART’s data is wrong, incomplete or suddenly outdated? Human override is necessary, but the system’s safety case will depend on whether controllers can recognize the failure and act before it becomes part of a live decision.