Integral AI Inc., a five-year-old company founded by former Google researchers Jad Tarifi and Nima Asgharbeygi, develops AI models geared for automated systems such as robots and self-driving cars. The company has worked with auto parts maker Denso Corp. since 2021 to help teach industrial robots new skills by observing demonstrations.

The 15-person startup is holding initial discussions with Toyota Motor Corp., Sony Group Corp., Honda Motor Co., Nissan Motor Co. and Mitsui Chemicals Inc. to pitch them on how artificial intelligence can advance manufacturing processes. The next step is for a human operator to give a robot a language prompt, like “make a coffee,” and have the robot teach itself how to do so, Tarifi told Bloomberg News.

Japan is home to many of the world’s biggest industrial robot makers, including Fanuc Corp. and Yaskawa Electric Corp., while SoftBank Group Corp. is buying the robotics unit of ABB Ltd. The country also hosts factory automation providers such as Mitsubishi Electric Corp. and Kawasaki Heavy Industries Ltd., with Japanese companies delivering an estimated 29 per cent of the global supply of industrial robots, according to the International Federation of Robotics.

Integral has a role to play because “Japan is strong in robotics, but they’re not strong in AI and computing,” Tarifi said.

The 42-year-old, who started Google’s first generative AI team in 2013, is one of a growing number of AI doctorate holders who see the workings of the brain’s neocortex as key to building AI architecture and algorithms that mimic the way a child learns.

Tarifi’s goal is to create AI models that can distil information with less data and process new information without accidentally deleting prior data — essential for learning. Such models would enable companies to push ahead in physical AI and handle sophisticated tasks such as designing new batteries, discovering materials and drugs or powering humanoid robots, Tarifi said.

Ultimately, the aim is to help companies teach robots to build new robots. “They might build a cooking robot, they might build a cleaning robot, or they might build a factory robot that builds an iPhone,” he said.

The ability to self-learn could eventually free machines from the need for updates. Existing large language models, such as OpenAI’s ChatGPT and Google’s Gemini, require more human-guided training, which can limit flexibility, efficiency and reliability, according to Tarifi.

Facing life-and-death situations as a child in war-torn Lebanon, Tarifi realised early on that “to actually have meaningful impact with AI, you need to affect the physical world, not just the digital world,” according to a 2024 interview with Nikola Danaylov — author of Conversations with the Future: 21 Visions for the 21st Century — on the Singularity.FM podcast.

Integral, which has raised about $5.5 million to date, now seeks about $10 million in a new funding round to scale up its model and ready it for public release. That’s pocket change compared with what big tech companies are spending on AI, but adequate for developing an algorithm, Tarifi said. The company will seek greater scale after the planned launch of Integral’s Genesis model later this year.

“The company’s claims are extremely bold,” Danaylov said in a recent interview. “But when you can’t afford to use or recreate the paradigm, you have no other option but to invent a new one.”

Source link