Microsoft CEO Nadella calls for adding an emergency brake to AI — CNBC

Microsoft CEO Satya Nadella said that advanced artificial intelligence systems should have containment mechanisms, independent oversight and an emergency shutdown mechanism that would allow authorized people to pause or turn off a model while it is performing a task. He wrote this on the social network X, CNBC reports.

Control over models

Nadella believes that non-deterministic models should be surrounded by reliable deterministic system architecture, human oversight and proven operating procedures. He also called for developing industry standards where existing rules are insufficient.

According to him, an approach similar to the handling of internal risks should be applied to advanced closed models and open-weight models. This should become one of the ways to build a system of control over AI.

More current news is available on the UA.News Telegram channel Telegram.

Principles of observability

The Microsoft chief proposed designing AI systems according to the principles of observability. Among these principles, he named model diversity, a human-understandable trace of the model's actions, continuous system testing, independent oversight and audit mechanisms, containment, and disclosure of incident information.

Nadella noted that the most reliable superintelligence system will not be the one whose model is trusted the most, but the one that allows the least reliance on trust in the model. His statement came amid calls from other technology company executives and researchers to strengthen safeguards in AI and, in some cases, slow the development of advanced technologies.