Roboticist Ken Goldberg Cautions Against 'Humanoid Hype,' Citing 100,000-Year Data Gap

UC Berkeley roboticist Ken Goldberg recently tempered expectations regarding the immediate widespread adoption of general-purpose humanoid robots, stating on social media, > "Humanoid robots are coming, but don't hold your breath waiting for them." His statement, accompanied by a link to a UC Berkeley News article published on August 27, 2025, underscores a significant disparity between the rapid advancements in AI chatbots and the slower progress in physical robotics.

Professor Goldberg, a leading expert in the field, attributes this extended timeline primarily to what he terms the "100,000-year data gap." This concept highlights the vast difference in the amount of training data available for large language models (LLMs), which are trained on centuries' worth of internet text, versus the limited real-world interaction data available for robots. He emphasizes that tasks requiring physical dexterity, such as picking up a wine glass or changing a light bulb, remain profoundly challenging for robots, a phenomenon known as Moravec's paradox.

The Berkeley News article further elaborates on this data challenge, explaining that while LLMs benefit from immense digital datasets, collecting comparable real-world data for robots through methods like teleoperation is painstakingly slow. Each eight hours of human-controlled robot operation yields only eight hours of data, making the accumulation of sufficient experience for complex physical tasks a monumental undertaking. Simulations, while useful for some robotic movements, have proven insufficient for teaching intricate dexterity.

Goldberg's perspective contrasts with the "humanoid hype" propagated by some tech leaders, including Elon Musk and NVIDIA CEO Jensen Huang, who foresee humanoid robots performing complex tasks like surgery or household chores within a few years. He notes that the robotics research community is currently undergoing a "paradigm shift," with an ongoing debate between traditional model-based engineering and the newer data-driven approaches.

Despite the challenges, Professor Goldberg believes robots will serve as "intelligence amplifiers," augmenting human capabilities rather than replacing them. He points to task-specific robots, such as those used in surgical assistance or logistics, as current examples of effective automation. His work with Ambi Robotics, focused on package sorting, exemplifies how specialized robots can tackle laborious tasks, but he maintains that jobs requiring nuanced manual skills, like those of plumbers or electricians, are far from full automation.