Synthetaic’s AI is unlocking the previously impossible

The Business News Staff

Staff

April 5, 2023

GREEN BAY – You’d be hard-pressed to find someone who hasn’t heard about the presence of a Chinese balloon over the United States.

A lesser-known tidbit, however, is the part where a Wisconsin-based company’s game-changing AI technology played in pinpointing the exact location the balloon originated from.

“For (Synthetaic), this marks the first broad obvious case of what we’ve always said we could do with RAIC (pronounced rake) – find anything in massive amounts of satellite data,” Corey Jaskolski, Synthetaic founder and president, said. “We’ve shown that to be true.”

A bit of background
Synthetaic – an end-to-end artificial intelligence (AI) company – was founded in 2019 by Jaskolski, who said he was inspired to reimagine the limits of AI and data.

Jaskolski earned dual degrees in both physics and mathematics from the University of Wisconsin-Stevens Point in 2000, before enrolling in grad school for electrical engineering and computer science at Massachusetts Institute of Technology (MIT).

Jaskolski said he always figured he’d end up at Lockheed or Boeing doing engineering work.

“But I had a huge passion for photography, and while I was in grad school at MIT, I worked for an underwater robotics company called Bluefin Robotics, which built small robots that would explore the bottom of the ocean by themselves,” he said. “You let them go and they come back 24 hours later with imagery and sonar data – all kinds of information about the ocean.”

Jaskolski said his work with the batteries that powered these robots led to a once-in-a-lifetime opportunity.

“While I was working there, Filmmaker James Cameron called up Bluefin and said, ‘Hey, the batteries you guys are producing, I need those for little tiny robots I’m using to film inside the Titanic,’” he said. “So, I went out to show Cameron’s team how to use the batteries.”

While walking them through the process, Jaskloski said one thing led to another and he found himself tagging along on the exhibition.

“While I was out there, they said, ‘These seem pretty complicated, why don’t you just come on the trip with us?’” he said.

What was meant to be a “just-in-case” role on the ship, Jaskolski said, turned into a much closer experience.

“I was meant to just be out on the ship, I wasn’t meant to actually go see the Titanic,” he said. “But one of the robots got stuck and the battery was damaged. So, they asked me to descend 12,500 feet to the Titanic in a three-person submarine to help extract the robot from inside the Titanic.”

Jaskolski said at that moment, he realized he had managed to combine two things he loved – photography and engineering.

“The robots were used to film inside the Titanic,” he said. “I managed to combine engineering by building the power systems for the robots. It was over for me at that point – I knew I needed to keep doing explorations and engineering and find ways to merge the two.”

That eventually led to a fellowship with National Geographic (Nat Geo).

As a fellow, Jaskolski designed, built and deployed dozens of technologies. 

Jaskolski said his work with Nat Geo took him to all seven continents on assignments.

“We scanned Mount Everest by helicopter so we could use AI to look at the vegetation growth up the side of it,” he said. “We built AI to detect gunshots from poachers in areas of southern Africa.”

In reflection of his efforts, Jaskolski was named the 2020 Rolex National Geographic Explorer Of The Year.

Appreciate failures
Jaskolski said although he had many successes, he also had failures.

“The reality is, I failed at like 80% of my AI projects,” he said. “You only hear about the successes. But it was the failures that kept me up at night. Not because you have to fail to succeed – but the why. Why are so many of these AI projects that start out with great ideas outright failing? Is it the algorithms? Are we dumb?”

Corey Jaskolski is the founder and president of Synthetaic – an end-to-end artificial intelligence (AI) company based in Delafield, Wisconsin. Heather Graves Photo

Jaskolski said he realized that much of the issues stemmed from the data required to train AI.

“Normally, if you want to train an AI model to try to find any object – for instance, if you’re trying to find a logo – you would need to take many, many, thousands of pictures of the logo – on the walls, signs, etc. – and around each picture, we’d draw a box,” he said.

Jaskolski said Google, Facebook or Microsoft, for example, have billions of labeled images.

“Unless you’re a huge company like that, getting that many images is nearly impossible,” he said. “That’s the Achilles heel of AI – you need that many labeled images to take advantage of the modern AIs that are out there. And it has got worse and worse as AI algorithms have gotten better and use more and more data.”

Which Jaskolski said sparked the idea for Synthetaic.

“We started the company to literally ask the question – what if we didn’t need to start with any labeled data?” he said. “And that’s crazy – that is like sacrilegious in AI, you don’t ask that question.”

But, Jaskolski said Synthetaic did.

“We had a couple of misguided attempts at first that didn’t work well,” he said. “Then we came up with this tool called RAIC (Rapid Automatic Image Categorization) that worked spectacularly well without labeled data.”

Jaskolski said what Synthetaic’s AI does differently is instead of having to have the human-labeled pieces, it just pulls in all the raw data.

“It doesn’t have to go through a human labeler first, you literally pull all the raw data and it starts understanding the way data relates to each other,” he said.

Jaskolski said it’s similar to the way people learn different languages.

“For example, if you wanted to learn Japanese, and you took four years of Japanese in high school and four in college – this is similar to normal AI,” he said. “You’re learning by people telling you words, being tested on the words, seeing how well you can understand it, having specific little exercises you do over and over. And eight years in, maybe you’re pretty good at Japanese.”

Another way to learn Japanese, Jaskolski said, is to go to Japan and totally immerse yourself in this language – this is similar to how Synthetaic’s AI technology works.

“That’s what our tool does – unsupervised AI, meaning a human isn’t telling it things a million times over – it learns that context by itself,” he said.

Jaskolski said the way the technology works is “we give it one example” – whether that is a hand drawing, a picture or a logo – and RAIC looks through the data set and says, “Oh, I think I know what this means – do you mean this sort of stuff?”

“Then a human would nudge the AI – a human-machine interaction – and within a couple of minutes of doing that, we get something at nearly the same quality as the painstakingly labeled human-trained data,” he said.

On a mission to find a balloon
Jaskolski said Synthetaic’s part in pinpointing where the balloon originated from was more “something fun to do,” than a planned-out project – one that stemmed out of a near team-wide COVID-19 quarantine.

He said the team had just unlocked a new feature of RAIC – which allowed them to search for objects in geospatial data. 
“We call that geospatial object detection mode,” he said.

Jaskolski said the team was brainstorming on “feats of strength” they could do with the new feature.

“We had, in retrospect, all these bad ideas,” he laughed. “Let’s take the satellite data of the United States and find the Oscar Meyer Wienermobile – cool, but not necessarily important.”

The brainstorming eventually landed on finding the origins of the Chinese balloon.

Corey Jaskolski

“I said, ‘I know the balloon was in South Carolina because it was shot down off the coast Feb. 4, 2023,’” he said. “‘What if I pull all the data in that date range over all of South Carolina, would I be able to find the balloon?’” he said. “I imputed a badly hand-drawn picture of the balloon and in like two minutes, I actually found it.”

Jaskolski said this “fun, I’m on COVID quarantine” project was the first time Synthetaic deployed the new feature at that scale.

“Once we found it once, we could run the wind models backward – so then I knew not only where it was from satellite data, but could actually calculate the altitude of the balloon,” he said. “Once we got to Canada, unfortunately, we weren’t able to find the balloon anymore.”

At that point, Jaskolski said the team changed course.

“We scaled RAIC so it could work over an entire state,” he said. “So, we asked ourselves, ‘What about running it over a huge swath of the world at once – could we do that?’”

From there, Jaskolski said the team downloaded all the satellite data from all of China, South Korea, North Korea, Japan and the Asian Islands over those few weeks.
“We had no idea when the balloon was launched,” he said. “All we knew is that on Jan. 31, we had it sitting in Canada. We didn’t know how long it was airborne – was it five days, was it 15 days? We ended up having this huge cross-section of time and space. We downloaded well more than 100 terabytes of satellite data (from Planet Labs).”

Jaskolski said searching across that much data with a regular AI would have taken many weeks, even with a big array of servers.

“In RAIC, it’s now looking for things in what we call RAIC space, which is the pixels – a representation of the imagery,” he said. “I was surprised when we found it.”

Once the team located the balloon in Asia, Jaskolski said they reverted back to wind models, and “we found it all the way back to the launch source.”

Jaskolski said RAIC’s new feature that can search on a global scale in just a few minutes was the “force multiplier” allowing the Synthetaic team to track the balloon.

Synthetaic’s process was the first to track the balloon itself, not just its expected path based on weather projections.

“Not knowing what the government may or may not be doing on tracking things like these balloons, I think what we offer is an interesting alternative – Because we use commercial satellite data,” he said. “We’re a commercial company. This wasn’t done on top secret networks – this was done with commercial data. And the implication is – what else can we find in the data? But also, what else can people find about us?”

Making a name
Synthetaic is based in Delafield – a city of a little more than 7,000 people in Waukesha County.

Jaskolski said approximately 38 of the company’s 45 employees live and work in Wisconsin.

“A lot of times, people think of AI companies as being in Silicon Valley, or if not there, on the East Coast,” he said.

The ability to grow AI companies, or any technical company, in the Midwest, Jaskolski said, is important for two reasons. 

“One, I subscribe to the concept that ‘talent is distributed evenly across the country, but opportunities are not,’” he said. “Being able to have an AI company here in the Midwest, we’ve been able to attract amazing talent – and these are people who didn’t want to move to the coasts to be in AI. They want to stay here where they can raise their families in the Midwest.”

Jaskolski said he also thinks it’s important for the government and national security to have technology companies distributed around the nation.

Creating and building an AI company in the Badger State, Jaskolski said he’s often been asked if it was hard.

“I’m happy to report we have had no problem getting amazing talent – PhDs in AI and occupational neuroscience – just amazing talent we have been able to recruit locally,” he said.

Jaskolski said the University of Wisconsin System is a big part of that.

From the investment side of things, Jaskolski said having TitletownTech in Wisconsin has been both an “amazing partner” and a “great business resource.”

Jill Enos, TitletownTech (TTT) managing director, said the firm met Jaskolski when he was first getting Synthetaic off the ground.

“He wanted to build the company in Wisconsin, and he wasn’t even looking for venture capital here because he didn’t think any existed that would meet his needs,” she said. “So, it was very opportune we were able to find him and he find us, and we’ve been involved with Synthetaic since 2020.”

Since then, Enos said TTT has served as an adviser to Jaskolski.

“We’ve been closely working with him as both an investor and a strategic partner,” she said. “He’s grown the team to a tremendous group of talented people.”

Enos said TTT is proud Jaskolski has been able to build in Wisconsin and attract talent from and to here.

On the technology front, Enos said what Jaskolski is able to solve with Synthetaic’s AI is unlike what any others in the market have been able to do. 

“Being involved with him and evaluating and exploring with him some of the applications for what this technology can do has been an honor,” she said. “It has also been incredibly exciting to help him be able to take this to the next level. And I think there’s so much more that is going to open up with this technology.”

What’s next?
Jaskolski said the sky’s the limit for what’s next for Synthetaic and its RAIC AI technology.

“Now that we’ve scaled this thing and we can run it at that level – it’s kind of like a needle in a haystack, and we have a magnet,” he said.

Jaskolski said that opens up all kinds of possibilities.

“One of the things we’re working on is with a climate tracer organization that is looking for greenhouse gas emission sources,” he said. 

Jaskolski said Synthetaic is helping the organization look for concentrated animal feed operations for methane emission sources, which was “completely un-trackable for them in terms of what it would take to do that budgetarily with regular AI.”

Jaskolski said the Chinese balloon was something everyone – national security, media, public interest – wanted answers to.
For Synthetaic, he said that’s just the beginning.

“It was something we were able to do, right after the balloon was popped,” he said. “The next time something like this comes up, we hope to be able to answer this before it even gets to the United States. Whether it’s a balloon, a climate-related phenomena or anything else RAIC can get answers to quickly and get it out there – that’s what I get excited about.”