Samsung Electronics Ltd.’s early-stage startup accelerator Samsung NEXT is looking to secure a foothold for its parent company in the emerging field of artificial intelligence, announcing a new fund dedicated to AI startups.

Samsung NEXT, launched in 2013 with a focus on emerging technologies, has already invested in a number of AI startups, including deep learning company Bonsai AI Inc. and nuTonomy, which builds software for self-driving cars.

Now, Samsung NEXT is launching the Samsung NEXT Q Fund with the specific goal of helping to “enable the full potential of AI’s future,” the company said.

Samsung NEXT Director Ajay Singh told SiliconANGLE that the company isn’t putting any limits on the Q Fund so it’s not clear how much will ultimately be invested. “We adhere to the philosophy that if the technology development is important for AI, it’s important for us to take a look at,” Singh said. “In that sense, we’re not capping our fund.”

Singh offered more clarity about the focus of the project, which he said is due to run for the next five years and beyond. The Q Fund will focus on identifying startups using AI to solve computer science problems, and also those that are solving AI problems themselves, he said. The idea is not only to fund these startups, but also to provide access to Samsung’s technical expertise and talent.

The company is also looking to invest in startups that have “nonobvious” and “forward-thinking” approaches to AI, rather than those that already have existing, applied AI technologies. “[The] focus is on early stage investments with an idea to invest small amounts and then double down as the company progresses,” Singh said.

Asked for some examples of these “nonobvious” approaches, Singh said that Samsung NEXT was interested in startups tackling some of the grand challenges facing AI right now. He said some of these challenges include learning in simulation, scene understanding, intuitive physics, program learning programs, robot control, human computer interaction and meta learning, just to name a few.

“We are interested in founders applying novel approaches to these challenges, which have historically been immune to conventional techniques,” Singh said.

One example of such a startup is Covariant.AI, which uses new techniques including imitation learning and deep reinforcement learning to train robots in new and complex skills. Formerly known as Embodied Intelligence, Covariant is currently building smart manufacturing robots that learn from human workers through virtual reality. Samsung NEXT said Covariant is an example of its focus on startups that replace traditional algorithm-based learning with newer and more efficient systems for training AI models.

“We’re at a critical juncture in the development of robotics — one where conventional techniques have reached their limits,” said Pieter Abbeel, founder of Covariant.ai. “Q Fund is one of the few venture funds to understand that many ‘Grand Challenge’ problems, including ours of robot control, could be solved with an AI-first approach.”

Image: geralt/pixabay