A simple drawing of what a weather balloon might look like from above led to Delafield-based artificial intelligence startup Synthetaic being featured in The New York Times. The company’s Rapid Automatic Image Categorization platform was able to track the Chinese spy balloon that was detected over the U.S. in January. BizTimes Milwaukee reporter Ashley Smart caught up with Corey Jaskolski, founder and president of Synthetaic, to learn how the company’s platform was able to successfully locate the balloon 13 times and what that feat means for the future of artificial intelligence.
How did you first learn about the Chinese spy balloon?
“I think when I first learned about it was when it was first sighted in Alaska, and no one was really sure what it was, and then it came to the U.S. and people became pretty certain what it was. I just heard about it in the news cycle and having launched weather balloons of my own, it immediately caught my attention. When I first heard of it, I actually didn’t even think about trying to find it with our RAIC tool. Not until a week or so later when I was home sick on the couch with nothing to do.”
Why were you interested in trying to track the balloon?
“We had just developed a new feature (geospatial object detection) for our product. Our product had always been able to find pretty much everything in images and video, but we just came up with this feature that allowed us to find individual objects in satellite imagery. That was somewhere in the back of my mind, but then the other thing was I’ve always been interested in using AI to try to find things that seem relatively impossible. While I was at home sick I said, ‘Hey, I think this tool is ready to deploy at a massive scale like this’ and gave it a try.”
What allowed for the RAIC platform to successfully track the balloon?
“I thought there was a pretty big chance it wouldn’t work. The first drawing (of the balloon), I made it thinking of the way I’ve seen airplanes in satellite imagery. When you see an airplane in satellite imagery, the satellite takes multiple images as it’s going over. It takes a red image, a green image and a blue image and then merges them together to get a color image. Anything that’s not on the ground – that red, green and blue – gets spread out as three separate images. I thought this balloon would be going a lot slower than an airplane, so it wouldn’t be as spread out in the imagery. I ended up drawing a sort of snowman with a red, blue and green circle on top of it. The red, green and blue were separated by what I thought the wind speed would be. I thought it would be an accurate representation of what it looked like in the satellite data, but I was really quite wrong. Because the balloon was flying so high, the pattern I drew ended up being spread out a lot more. It wasn’t a snowman. It was three separate dots separated by a very large distance. But RAIC still got it and still knew what we were looking for from that one drawing. My drawing looked enough like the balloon that RAIC was able to generalize.”
How did you end up working with The New York Times?
“Wired magazine published an article on us about a week or so before we were contacted by The New York Times. We had already found the balloon throughout the entirety of the United States. When we first started working with them all we had was (the balloon) throughout the United States, we didn’t have it anywhere else yet. Working together with them, we made these additional detections all the way back to its launch site Hainan, China. The New York Times was super involved. There were three reporters working on it. They actually ran wind models themselves and did calculations to back up our calculations. This group is their visual investigations group. They, in fact, brought information to us in some cases that we did not have.”
What does RAIC’s ability to find the balloon illustrate about the power of AI technology?
“What it really illustrates is the ability of AI, and specifically the way we do AI with RAIC, to really find needles in a haystack. If you imagine as a general user, you’ve looked through Apple Maps or Google Earth and zoomed in and looked at your house or other properties. Imagine trying to find a single object that’s hidden in Google Earth. It’s sort of the equivalent of what we can do with AI now. We’re hopeful that we’ll be able to use it for everything from defense and security but also for nonprofit and NGO work.”
How has the New York Times story helped the company’s reputation and growth?
“The impact has been tremendous. We’ve been telling folks that our tool can do this, and we do lots of demos and people are always blown away. I think this is still so new of a concept in AI that we still run into a lot of skepticism and disbelief, which is important in the world of AI, where a lot of people have some misguided views of what AI can really do. Having that quick response and being able to find the true needle in the haystack – as if we brought a magnet to the party – I think really showed people what we’ve been saying is true.”
What are your thoughts on how quickly AI has developed? Are things moving too fast?
“In general, the pace of AI has been so rapid lately. We’ve always been on an exponential curve with technology and there are just some points in time where you really feel that acceleration. This is one of those times, with ChatGPT and the other generative AIs. One thing I think people don’t realize is AI has been around in the background for a very long time. These (current) applications of it capture the public’s imagination. AI researchers have been using data to check for credit card fraud, to try to estimate housing crises from real estate listings, all that stuff has been around.
“Do I think it’s moving too fast? No. I think one misperception is that AI can run amok and do its own thing. We’ve all seen Terminator and The Matrix. The reality is AI right now is nowhere near that. We are at a point where AI only works in a human-machine collaboration. You still have to prompt ChatGPT with a sensible question for it to come back with a sensible answer. What AI can really do right now is make us better at our jobs.”
