As part of a much broader push to modernize America’s aging air traffic systems, the FAA has announced several new AI tools over the last week that will address ongoing problems with flight safety and congestion. This comes just two months after Secretary of Transportation Sean Duffy modified his budget request to Congress to emphasize AI. While there are several different models made by several different companies here, the main throughline is that they will be synthesizing data that the FAA keeps in distinct databases. That ought to allow for analysis that just can’t be done quickly or efficiently right now, leading to fewer airport collisions and late flights.
One AI model is called Foundry, made by defense contractor Palantir. By looking at data that is currently “scattered across the federal government and other sources,” it will look for identifiable patterns in airport safety issues, per Politico. In other words, it will suss out specific actions or locations that cause more problems than normal, which the FAA can then turn into new guidance, investments, or upgrades. The AI’s conclusions will be reviewed by actual living humans, so we shouldn’t be implementing any hallucinations if all goes well. The first year of development is estimated to cost $4 million.
Just to make it clear, Foundry will be looking for repeatable behaviors or hotspots that increase statistical risk; it is absolutely not Minority Report, predicting individual instances. The March collision between Air Canada Express Flight 8646 and a fire truck, for example, is not something Foundry could have foreseen. At the end of the day, people in the air traffic control towers and the vehicles have to be making the right calls in the moment. No AI can help you with that.
Smarter flying with SMART
Separately, Boston defense company Air Space Intelligence (ASI) won a bid to create AI models to ease traffic congestion by predicting bottlenecks and delays. The FAA announced that “the Flow Management Data and Services (FMDS) will be the new technological backbone of the FAA’s Air Traffic Control System Command Center.” ASI will also create another system within FMDS called Strategic Management of Airspace, Routes and Trajectories (SMART).
Basically, right now, planes take off without enough regard to how many other planes also want to land at the destination airport at the same time. That’s how holding patterns and landing delays happen. On top of that, once again, lots of available data, such as weather, airport construction, and runway closures, are all held in different systems, so it’s difficult to coordinate all this information in real time. ASI’s AI models are promising to finally do this at a cost of $875 million, per Bloomberg.
What could possibly go wrong?
“It’s not often you hear industry say to the FAA, ‘You’re going too fast,’ but we’re getting that from our operator partners as we’re working with them to be able to deploy this in September,” said FAA Senior Certification Advisor Steve Fulton, per Aviation Week. Yes, the rollout for ASI’s SMART is scheduled for just a few months from now. Handing something as important as air traffic over to AI, even with human oversight, is a pretty big move. The private sector in general may be hungry for AI, but even it thinks the government needs to let up on the throttle a bit.
It will at least be a staged rollout, in that it will only be used on traffic above 24,000 feet initially. But that’s still a lot of planes! The question would be whether these AI systems need more testing — potentially a lot more — before being introduced. Of course, the government has long been accused of adapting too slowly to innovative technology. One might argue that’s how the nation’s air traffic system deteriorated to its current crisis state in the first place. Is the answer to adopt cutting-edge tech right away? We’re about to find out.
