The startup spun out of UCL research into how the brain works is building a new kind of AI model – much better, it claims – to help hardware move more intelligently through the physical world – and has now raised $8 million in funding to get it off the ground.
Stanhope AI isn’t the first company to build an AI system to power autonomous machines. But the approach is different. Based on the neuroscience concept of “active reasoning,” Stanhope’s AI learns in the same way that the human brain is thought to learn, and then continually modifies what it does in real time. Stanhope believes its approach produces a system that is not only easier (and cheaper) to start using in active environments, but also more responsive to spontaneous situations.
If Stanhope’s technology works as expected, it could be a major step forward for autonomous systems. Introducing more autonomy is a key priority in military, industrial, and other industries that are introducing new generations of equipment and seeing AI as the future of their industries.
But as events involving two of the most high-profile AI drone startups, Stark and Helsing, demonstrated, real-world deployments have been far from perfect in physical scenarios. Until now, the standard and expensive approach has been to use huge datasets to “train” AI. Stanhope believes that its unique paradigm of how AI can “learn” can significantly improve outcomes.
“This is completely different from deep learning,” said Professor Rosalyn Moran, co-founder of Stanhope AI, in an interview with Resilience Media. “This is actually what we made. [to be like] Memories in the human brain. You get all the great efficiency and logic, but you also get something basic: independence. This is a true generative model. ”
In fact, Moran believes that current versions of “generative AI” aren’t actually generative at all. “They stole the word,” she says offhandedly about the hyperscalers who use the word a lot these days.
The GenAI startup has been around for more than just one term. Companies like OpenAI, Anthropic, and self-driving car companies have poured hundreds of billions of dollars into the training challenge of building models, but they’re not done yet.
Moran said Stanhope’s technology is already being tested by an unnamed robotics and drone company. You can see part of the test here. It shows how drones equipped with Stanhope’s ‘agents’ are introduced into new environments with obstacles and figure out how to navigate around them to reach their destination. The startup will use the seed money to continue expanding its research and business funnel.
Frontline Ventures is leading the seed. New backers include Paladin Capital Group and Auxxo Female Catalyst Fund. Previous investors UCL Technology Fund and MMC Ventures also participate.
A meaningful time at UCL
Stanhope was the brainchild of Moran and another professor in UCL’s neuroscience department, Carl Friston. Moran originally came to study and work in this field from the world of electrical engineering.
Friston’s research into understanding how the brain works was crystallized in the “free energy” theory. In very basic terms, this theory posits that in order to survive, self-organizing organisms continually learn from their external environment by comparing it with what they already know. Given that the brain functions through a series of electrical impulses, Moran joined the team to expand on that angle of research.
It was a busy time at UCL (just kidding). In the same hallway was Demis Hassabis, who was conducting original neuroscience research that ultimately led to founding DeepMind. Naturally, perhaps learning from their environment, the duo saw an opportunity to spin an idea for an alternative to physical-world AI that better reflected how they think the brain works.
One of the unique aspects of how Stanhope’s technology is implemented is that the system’s processing occurs at the “edge” of the device itself, rather than in the cloud (as is the case with autonomous system training today).
This significantly reduces costs and energy usage in building and operating the system. It could potentially provide solutions for how military drones can function even in electronic warfare environments, where communication with the operator is cut off, for example.
This makes the technology applicable to a wider range of use cases and has brought investors to the attention of the issue.
“We have been looking at a significant number of new hardware vendors in the defense technology space,” said Nazo Moosa, partner at Paladin. “What we were really excited about about Stanhope is that this goes beyond drones. Drones are an important test bed, but we looked at it from a different angle. If you look at the physical world or the digital world, the data is still being used up. It’s hard to be sure, but in the physical world, data is the biggest crisis point.” She noted that it takes Waymo to market to start collecting training data in the physical world, and while warehouses are an easier environment, combat zones are not.
“In a combat zone, amazing things happen,” she said. “We don’t have the ability or flexibility to train for millions of hours, and that’s what Rosalynn and her team are fixing.”
