AI Is Moving From Chatting to Walking

A computer scientist who spent eight years digging into the nuts and bolts of large language models—those programs that power chatbots like ChatGPT—started to feel stuck. He said the field had moved beyond basic research and was now mostly about building products. So he left his PhD program at Brown University. Instead of polishing existing chatbots, he launched a startup called Overworld. The company’s goal is clear in its name: create AI that can see, understand, and move through real environments, not just read text. The shift reflects a broader trend. Many developers are tired of tweaking algorithms that only work on screens and want to see machines learn how to interact with the world—pick up objects, navigate rooms, or help in factories.Overworld’s approach is to give AI a sense of space. By training models on data from cameras and sensors, the system learns to map its surroundings and plan actions. It’s a step toward robots that can follow humans, perform tasks safely, or explore unknown places. Critics point out the challenge: building AI that works outside controlled labs is hard. Sensors can be noisy, and real‑world situations are unpredictable. Yet the potential payoff—more useful robots in homes, hospitals, and industry—is huge. The move also highlights a philosophical change. Instead of treating AI as a purely digital helper, researchers now see it as a partner that shares our physical world. The question becomes: how can we make these systems reliable, ethical, and safe as they step onto our streets? The future of AI may not be in answering questions online, but in walking alongside us. It will need careful design, robust testing, and a clear vision of what it means for machines to truly understand the places they inhabit.