Tesla engineer highlights AI safety challenges in physical robotics
Remote-controlled robot demonstrations underscore the complexity of ensuring safe AI behavior in the physical world.
Tesla engineer Ashok Elluswamy offered perspective on the state of robotics development, noting that current demonstrations of robot movement rely on remote control rather than autonomous decision-making. "These robot moves are essentially remote controlled. The robot rather blindly performs those while maintaining balance and such," Elluswamy wrote on X.
Elluswamy framed the challenge as fundamentally different from AI safety concerns in large language models. "AI safety for physical robots makes the current crop of AI safety for large models look like child's play. Extremely important to get it right," he added, emphasizing that deploying autonomous systems in the physical world carries distinct risks that require careful engineering.
The remarks come as Tesla continues development of Optimus, its general-purpose humanoid robot. While the company has shown video demonstrations of the robot performing tasks, Elluswamy's comments clarify that these clips showcase mechanical capability and balance control rather than independent decision-making by onboard AI systems.
"For those who don’t know, these robot moves are essentially remote controlled. The robot rather blindly performs those while maintaining balance and such. AI safety for physical robots makes the current crop of AI safety for large models look like child’s play. Extremely important to get it right. Quoting @cixliv: “This video is dedicated to the AI safety kids”"— X · @aelluswamy