How this startup tracked that Chinese spy balloon using AI
Synthetaic CEO on finding Beijing's 'needle in a haystack' of sat imagery
Interview Most of us binge Netflix when we're feeling sick. Corey Jaskolski, on the other hand, looks for Chinese spy balloons in satellite imagery.
Jaskolski is the CEO and founder of Synthetaic, an AI software startup that traced the journey the balloon took until it was shot down in February off the coast of South Carolina.
It did this using his company's machine-learning tool called RAIC – which stands for Rapid Automated Image Categorization – with commercial satellite imagery from Planet Labs, and a hand-drawn image of a Chinese spy balloon that "would have embarrassed a 4-year-old child."
And it only took about two minutes for the AI model to spot the balloon in those millions of square miles of unlabeled photos, we're told.
Jaskolski told The Register all about it, and you can listen to the conversation in the video below. Now you may think object detection in satellite photos is old hat, or nothing our intelligence services can't do, though it's interesting to hear how this technology is well and truly in our own hands, and how well it actually works.
"We ended up RAICing the equivalent of two times the entire surface area of Earth because we had to go back in time and we didn't know where the balloon would be," he told us. Ultimately, the AI model found the balloon 13 times in the data, which could also indicate whether the balloon was under active control by Beijing at each point.
Synthetaic shared this data with the US government before publishing what it found, and since then "we have had a lot of interest from the government" in its AI technology.
The technology has several other defense and intelligence use cases, Jaskolski added.
This includes analyzing historical satellite images to "find, every day, where every ship is on earth, where every airplane is, where every new balloon that gets launched," he explained.
Spoiler alert: he's already started in on this.
"These massive datasets — whether they're video, whether they're geospatial, whether they're still imagery, whether they're drone feeds — when when you get to a certain amount of data, it becomes wildly intractable for humans to review it and look for it," Jaskolski said. "And RAIC can help find that needle in the haystack." ®