The company’s DeepMind division compiled guardrails that can be fed into the large language models (LLMs) that instruct machines, giving them “a set of safety-focused prompts to abide by when selecting tasks for the robots.” (The Logic)

Talking point: The Robot Constitution, as the company called it, was inspired by science-fiction icon Isaac Asimov’s Three Laws of Robotics. The I, Robot author incorporated the laws into several of his books, composing them in a way that, at least notionally, would ensure robots safeguard humans and themselves, while remaining under human control. Among Google’s safety rules, robots are restricted from attempting tasks that involve “humans, animals, sharp objects or electrical appliances.” DeepMind’s guidelines come as concerns grow over the rapid advancement of the technology, and as leaders in the field call for tighter regulations to ensure AI is controllable and safe.