To interact with the real world, AI will acquire physical intelligence

recent A.I The models are surprisingly human-like in their ability to generate text, audio and video when prompted. However, so far these algorithms have largely remained in the digital world rather than the physical, three-dimensional world we live in. In fact, whenever we try to apply these models to the real world, even the most sophisticated struggle to perform adequately—think of how challenging it is to build safe and reliable self-driving cars, for example. Despite being artificially intelligent, these models not only have no understanding of physics but often hallucinate, which leads them to make inexplicable mistakes.

This year, however, is when AI will finally happen Jump from the digital world to the real world we live in. Extending AI beyond its digital boundaries demands reworking how machines think, connecting AI’s digital intelligence with robotics’ mechanical expertise. This is what I call “physical intelligence,” a new form of intelligent machine that can understand dynamic environments, adapt to unpredictability, and make decisions in real time. Unlike the model used by standard AI, physical intelligence is rooted in physics; To understand basic principles of the real world, such as cause-and-effect.

Such features allow physical intelligence models to interact and adapt to different environments. In my research group at MIT, we are developing physical intelligence models that we call fluid networks. In one experiment, for example, we trained two drones—one guided by a standard AI model and the other by a fluid network—to detect objects in a summer forest, using data captured by human pilots. While both drones performed equally well when they were trained to do exactly what they were supposed to do, when they were asked to detect objects in different conditions – in winter or in an urban environment – ​​only the Liquid Network drone successfully completed its task. This experiment showed us that, unlike traditional AI systems that stop evolving after their initial training phase, fluid networks continue to learn and adapt from experience just like humans.

Physical intelligence is able to interpret and physically execute complex commands derived from text or images, bridging the gap between digital instructions and real-world execution. For example, in my lab, we’ve built a physically intelligent system that, in less than a minute, can iteratively design and then 3D-print based on prompts like “robot that can walk” or “robot that can grip.” can object”.

Other labs are also making significant progress. For example, robotics startup Covariant, founded by UC-Berkeley researcher Pieter Abbeel, is developing chatbots—similar to ChatGTP—that can control robotic arms when requested. They have already secured more than $222 million to develop and deploy picking robots in warehouses worldwide. A group from Carnegie Mellon University also recently displayed A robot with just one camera and unspecified actuation can perform dynamic and complex parkour movements—including jumping over obstacles twice its height and across gaps twice its length—using a single neural network trained by reinforcement learning.

If 2023 was the year of text-to-image and 2024 was text-to-video, then 2025 will mark the age of physical intelligence, a new generation of devices—not just robots, but anything from power grids to smart homes—that’s what we’re telling them. Can interpret and perform tasks in the real world.