Microsoft CEO Satya Nadella has called for AI systems to give an authorized person the ability to stop a model while it is working. In a Saturday morning post on X, he likened that control to an emergency brake.
Nadella proposed separating a model from the software that directs its work and placing safeguards outside the model. He also called for tamper-proof, human-readable records of consequential model actions and said systems should be designed to contain a model from the outset in case it is compromised.
His proposal comes amid accounts from leading AI companies of incidents in which they appeared to lose control of their models. Anthropic CEO Dario Amodei has also published a plan for a more cautious approach to AI development.
Please note! Comments are published by readers. The opinions expressed may not reflect the position of the editorial team. NYCT is not responsible for the content of readers' opinions and asks authors to observe moral standards, not to incite national or racial hatred, and to avoid rudeness. Think before you speak. You are responsible for your words! By publishing an opinion, you agree to the rules! If this request is ignored, NYCT reserves the right to deprive the user of the ability to leave comments.