The next phase of AI development is moving beyond text and into the physical world. Researchers warn that US policymakers don’t yet grasp the scale of what’s coming, while China is already pulling ahead in robotics.<\/p>\n
While large language models predict the next word in a sentence, so-called world models<\/a> predict what might happen in a physical environment. They analyze multimodal data from video, images, text, audio, and other sensors to understand three-dimensional spaces. Researchers believe the technology could be just as transformative as large language models. Some, like AI pioneer Yann LeCun, see world models as a core building block for advanced AI<\/a>.<\/p>\n The applications range from robotics for warehouses and homes<\/a> to molecular environment simulation<\/a> for drug development and autonomous driving<\/a>. The umbrella term for all of this is “Physical AI.”<\/p>\n But policymakers are falling behind, researchers warn in a Politico feature. Russell Wald, director of the Stanford Institute for Human-Centered Artificial Intelligence, recalls warning Congress about large language models before ChatGPT launched in 2022 – and being ignored. He sees the same pattern repeating with world models, noting that many lawmakers don’t even know what a world model is.<\/p>\n Missing supply chains and China’s robotics lead<\/p>\n World models amplify the already massive demand for computing power. Blaine Fisher of Tulane University told Politico that keeping up with the data demands of language models alone was already a struggle and world models need physical hardware like robots on top of that. The US tech industry is pushing for a national robotics strategy<\/a> that would also strengthen supply chains against China.<\/p>\n Wald warns of a bottleneck similar to what happened with 5G: if there’s a breakthrough in world model research, the US will have given these systems a brain but won’t have the supply chains for the hardware they need. Meanwhile, a bipedal robot built by Chinese smartphone maker Honor recently broke the human half-marathon record<\/a>.<\/p>\n The societal risks also go far beyond the familiar problems with large language models, according to Wald. World models require a different kind of governance, especially around privacy, labor markets, and national security. The ability to analyze the real world could significantly boost surveillance and autonomous weapons systems. Fisher also warns that people could retreat into virtual worlds with lifelike physics and AI avatars, predicting that some people simply won’t want to leave their homes anymore.<\/p>\n \t\t\t\tAI News Without the Hype \u2013 Curated by Humans<\/p>\n \n\t\t\t\t\tSubscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive “AI Radar” frontier report six times a year, full archive access, and access to our comment section.\t\t\t\t<\/p>\n