There are many potential problems with the way we train AI models. Data sets used to teach machines to think have been repeatedly shown to create biased algorithms, sometimes leading to discrimination. By some estimates, data scientists spend 80% of their valuable time acquiring, cleaning, and managing data sets. And in some cases, the data sets don’t even exist in the first place.
Physicist Nathan Kundtz has a plan to fix that, which could drive huge improvements in AI training.
Kundtz, who previously led the Bill Gates-backed satellite communications company Kymeta as CEO, has launched a company called Rendered.ai to create synthetic data sets to train artificial intelligence systems, as well as build an infrastructure that enables users to engineers customize and modify them.
The Bellevue, Washington-based startup is currently working on AI that learns from visual data sets, including images and video, as well as radar, X-ray, and thermal imaging data. This week it announced a $ 6 million seed round.
There are other companies that provide synthetic data sets for AI training, creating images for machines to learn when they cannot be collected in real life. But they take a different approach.
“Most of them think of this problem as a computer graphics problem. So the product they provide and the solution to that is a set of images,” Kundtz said.
“We think of it very differently. We think of this as an engineering problem. And the product we offer is a platform that leads you to the business results you need to drive your business forward.”
Kundtz provided an example. There are researchers who monitor economic activity through the use of satellite images. They use AI to scan images of construction cranes, trucks, and other physical manifestations of commerce and growth. But before that can happen, the model might need a training dataset with millions of images to capture examples of these objects, plus the skills of an expert annotator to recognize and identify them. And then there are irregularities to deal with, perhaps snow covering a crane, interference from clouds, or rust obscuring it, causing a scorer to lose the item.
To solve that problem, Rendered creates its own mini 3D universe and populates it with cranes and trucks for the AI to learn. They can add the same irregularities, but still find the cranes and trucks. It’s a solution that may sound just as labor-intensive, but the simulation relies on basic science to carry the load.
“Especially in the world of computer vision, a lot of the data is really based on physics that we understand very well, as light interacting with things is [an area] where we have tremendous experience and incredible simulation tools. “Kundtz said.
Rendered puts that dataset in a framework so that AI engineers can modify and customize it.
It’s an ambitious vision, but Kundtz brings business credibility to the company.
Kundtz began his career as a manager at Intellectual Ventures, the high-tech innovation hub and the brainchild of former Microsoft CTO Nathan Myhrvold. While there, Kundtz led a team working on cutting-edge antenna technology, which became the foundation for spin-off startup Kymeta. He was Kymeta’s CTO and then CEO before leaving in 2018.
Kymeta has raised more than $ 440 million, according to PitchBook, with about half of that accumulated under Kundtz’s leadership. The company continues to rack up milestones with its next-generation satellite broadband technology.
It was the experience at Kymeta that helped inspire Kundtz’s new venture. When interacting with satellite companies, he saw that his work was sometimes limited by the lack of data sets to train the AI.
In addition to satellite companies, Rendered’s platform can provide data sets for medical, transportation, robotics and security engineers. The startup is currently in beta mode, working with a handful of users to fine-tune its technology. The initial funding, which came from Tectonic Ventures, Congruent Ventures, Union Labs and Uncorrelated Ventures, will help the company expand its services to more clients.
Kundtz is excited about the work he has embarked on, addressing what he sees as a fundamental obstacle to AI.
“The problem is a bit of a ‘well duh’ problem,” Kundtz said. “And yet when you see something sofundamental and yet not addressed in a way that is scalable and that really gets to the heart of the problem, that’s the kind of thing that gets me up in the morning. “
