Physically embodied artificial intelligence faces significant hurdles because, unlike large language models, there is no vast "physical internet" to scrape for training data. Companies are currently resorting to controversial methods, such as recording human movements in domestic settings, to bridge this data gap. Moravec’s paradox highlights why this remains difficult: machines easily master complex cognitive tasks like chess but struggle with simple physical actions like folding laundry or handing over a bottle, which require nuanced force control and environmental interaction. While humanoid robots serve as a valuable "moonshot" for research, the near-term future of robotics lies in specialized, collaborative machines designed for specific domestic assistance, such as elderly care. Experts like Grace Shao and Professor Subramanian Ramamoorthy emphasize that true physical intelligence requires seamless integration into human environments, a goal that remains a long-term engineering challenge.
Sign in to continue reading, translating and more.
Open full episode in Podwise
