Key facts
- Microsoft CEO Satya Nadella called for an 'emergency brake' for AI models.
- Nadella proposed separating AI models from their orchestrating systems.
- He advocated for externalizing controls and safeguards for AI.
- Nadella suggested documenting all meaningful model actions with tamper-proof evidence.
- He called for systems that allow authorized individuals to pause or shut down models mid-task.
Microsoft CEO Satya Nadella has outlined a new approach to AI safety, emphasizing the need for robust controls and safeguards. In a post on X, Nadella stated that AI models should not be treated as "nested black boxes" whose recommendations are simply accepted or rejected.
Nadella proposed separating the AI model itself from the systems that manage its operations, advocating for externalized controls and safeguards. He also called for "tamper-proof human readable evidence" to document every significant model action. Furthermore, he suggested implementing systems that grant authorized individuals the ability to "pause or shut down a model mid-task," likening this capability to an "emergency brake."