"You cannot simply plug current large language models (LLMs) and algorithms into humanoid robots and expect them to work properly. Only when we examine how artificial intelligence (AI) operates, as thoroughly as taking a magnetic resonance imaging (MRI) scan, can AI robotics be precisely implemented."
Daniel Lee, a world-renowned scholar in AI robotics and professor at Cornell University in the United States, made these remarks in an interview with Seoul Economic Daily at the Shilla Hotel in Jung-gu, Seoul, on the 27th, stressing that "physical AI is the final frontier of AI." Professor Lee researches robotics, neuroscience, and AI at Cornell University, and previously led AI research as head of Samsung Electronics' Global AI Center. Ahead of his lecture on "The Physical Revolution Led by Robotics" on the 28th, the second day of Seoul Forum 2026, Lee shared his views on how the "brain" of AI robotics should be built.
Lee emphasized that physical AI — designed to accurately perceive the real world and act like humans — will face problems of a different dimension than those previously addressed. He posed the question, "Why don't we yet have a robot butler in our homes?" His answer: "Cooking, washing dishes, or folding laundry are everyday tasks for humans, but they are difficult for AI, which lacks reasoning capabilities." He explained that while Google's AI AlphaGo won its battle of wits against former Go player Lee Sedol 10 years ago, what is now needed is the further development of AI that can act like humans.
Mistakes that AI makes in the digital space may end as mere hallucination problems, but the issue grows more serious when it comes to robots operating in the real world. If a robot misidentifies a dangerous object or misjudges a person's movement, it can go beyond a simple error and lead to safety issues.
He stressed that for physical AI to reach a certain level, "advances in basic science are needed, including better algorithms and a deeper understanding of AI." In this process, he said, what matters as much as AI performance is having a system that can verify how trustworthy AI is. The era of physical AI, in other words, could become a competition over whose algorithms are reliable enough to operate robots.
Lee defined current AI as a "black box." He pointed out, "Just as people in Freud's era inferred human psychology through clues such as dreams and behavior, AI today is still at a stage where we can only infer its internal processes from its outputs," adding, "Going forward, we need to understand AI to the level where we can explain by what principles it operates and under what circumstances it may produce errors."
Lee assessed that Korea, given its strengths in semiconductors, manufacturing, and industrial automation, can hold an important position in the era of physical AI as well. He particularly forecast that the areas where physical AI will be applied first are likely to be in manufacturing — factories, logistics centers, and industrial automation sites.
He also expressed expectations for Korea's role. "Korea is in a unique position to bridge the United States and China," Lee said. "I hope Korea can become both a connection point and a leader where different technologies and ideas are freely shared."
