Synthetaic, an AI firm that is part of the Titletown Tech portfolio, tracked the Chinese balloon shot down in U.S. airspace demonstrating the power of its technology.
That data was subsequently used in a New York Times article that detailed the path of the balloon from its launch to the end of its journey across the United States.
Just days after a Chinese balloon was shot down in U.S. airspace, Synthetaic tracked the balloon’s flight path across the continental U.S. in commercially-available Planet Labs satellite data.
AI has historically relied on extensive human labeling and the building of bespoke models. In time-sensitive scenarios, like this one, those processes can be untenable burdens. Synthetaic used its rapid automated image categorization software to cut the time it took to find the balloon to minutes.
The human-machine collaboration made it possible to instantly find the object of interest in vast quantities of video, image, or satellite data, without pre-labeled data or pre-trained models.
“Like any researcher, I was skeptical that we’d really found it so quickly,” said Synthetaic CEO, Corey Jaskolski. “We were able to validate it was the balloon through size, shape, social media reports near by, and wind maps showing where it was expected be. Further we were able to use parallax (since satellites capture the an airborne object from different angles as they move) to calculate its altitude. Once we knew that what we’d found was actually the balloon, we could use that image as our new starting point in RAIC to find more locations.”
From there, Synthetaic refined its search and confirmed five more images of the balloon during its journey across the continental U.S. Jaskolski said, “We’ve mapped the track backwards, starting in South Carolina just before it was shot down all the back through where it entered the U.S. from Canada”
The potential applications go far beyond this high-profile news story.
“This is new in AI,” said Jaskolski. “Instead of thousands of labelers, a single person can now nudge an unsupervised AI into what they need it to be. That allows you to build an AI and run it in seconds across millions of images. To go from a place of needing millions of labeled images to only needing, in this case, a hand drawing based on an educated guess — it fundamentally changes what’s possible.”
