Microsoft detailed the project in a news article this week. The foam gliders are thrown into the air by a researcher and then handed over to the AI’s control. The AI, partially modelled on how birds stay aloft, uses a bevy of sensors and predictive technology to look for thermal air currents.
The columns of rising air are used by birds to gain height without making significant effort. Since gliding aircraft are lightweight and have no motor, the air columns are an obvious way to keep them aloft for longer. It’s difficult to find the columns and move between though, unless you’ve got AI on your side.
The AI assesses a range of factors including air temperature, wind direction and flight speed to predict the weather conditions ahead. Using meteorological analysis, it figures out where thermals are likely to occur. After deciding where to head for next, it controls the aircraft as it moves towards its destination. Microsoft believes this makes it the only AI in use in the real world that takes action based on its own predictions.
READ NEXT: Does Microsoft’s massive Office 365 spell the end for startups?
The prototype gliders Microsoft is testing do have motors fitted to them but they aren’t used unless the AI makes a serious mistake. While the glider is airborne, it’s the AI making all the decisions. The researchers consider it to be a useful experiment for the future of AI. It demonstrates that current neural networks can make decisions nearly instantaneously.
When AI-powered autonomous vehicles encounter a hazard, they’ll need to decide on the action to take without any hesitation. The gliders have a similarly small error allowance. If they drift out of the thermal, they could rapidly fall towards the ground with the AI powerless to regain control. Microsoft said the prototypes address how to teach AI the art of planning several steps ahead.
“For us, the sailplane is a testbed for technologies at the core of anything that will be considered intelligent in the next 10 years,” said Andrey Kolobov, the researcher leading the project’s engineering work. “AI in the real world will have very little room for error, like our sailplane.”
READ NEXT: LinkedIn forced to allow AI bots to scrape its data
“It’s really the question of, ‘How do you plan for the future, several steps ahead?'” added Ashish Kapoor, a principal researcher at Microsoft. “Computationally, that’s a very hard problem.”
Microsoft envisions the gliders could have a significant role in the future. Capable of remaining airborne for weeks or months, they’d be able to provide cellular coverage to isolated regions or disaster zones. The aircraft could also remotely monitor farms and provide live coverage of road traffic.
No ground infrastructure is required and the power could be sourced from solar panels. Still, there are some issues though, such as mechanical failure and the inability to keep the planes fixed in a certain region. A sudden change of wind speed or direction could send the gliders miles away from the area they’re needed in, forcing humans to recover and relaunch the planes.
