• This “framework” makes it so that a “human operator” can easily and safely interrupt whatever an AI robot is doing.
• Disobeying you would be one of the very few things these AI agents can’t learn.
• Google making it possible to securely stop AI robots from causing harm
• I am not too freaked out by the idea of a robot apocalypse, but I do say having a kill switch is necessary.
• And because the agents can also learn, DeepMind is making sure the AI can’t learn to prevent interruption.

Read more