Can We Build Safe Robots?

Amit Goel, head of robotics ecosystem at Nvidia, discussed the challenges of physical AI safety and how companies can prepare for the future. He explained the ‘Halos for Robotics’ concept, a new safety paradigm, and emphasized the need for safety layers to manage risks associated with physical AI. Goel also highlighted that agentic AI necessitates a ‘zero-trust’ philosophy and robust external infrastructure for monitoring interactions between AI agents and physical systems.

• Discussion on the challenges of physical AI safety.
• Introduction of ‘Halos for Robotics’ and a new safety paradigm for AI development.
• Emphasis on implementing safety layers to address new risks in the physical world.
• Agentic AI requires a ‘zero-trust’ philosophy and external monitoring infrastructure.
• Optimism about physical AI handling dangerous, strenuous, unsanitary, or inconvenient tasks.
• Mention of interviews with Amit Goel and Ali Golshan during GTC Berlin.
  https://foxvector.com/articles/418e3deb-e0ac-4bd7-9fd7-0c9bc15cf58a