Microsoft President Supports Independent AI Safety Assessments

Microsoft President Brad Smith has expressed support for the introduction of independent evaluators to assess the safety of artificial intelligence systems. Smith emphasized the need for multiple layers of human control over AI.

The topic of AI safety and global regulation was a key discussion point at the United Nations General Assembly in New York. In an interview with Bloomberg, Smith stated that progress on AI safety is difficult to achieve if only one or two organizations are responsible. He highlighted that while model developers are crucial, companies like Microsoft, which provide the software layer and control AI agents, are equally important.

Microsoft, in collaboration with OpenAI, has established a joint safety committee. This committee facilitates discussions on AI safety and helps determine when models are ready for release. Smith also indicated Microsoft's support for incorporating kill switches in advanced AI systems to maintain human oversight.

"Safety mechanisms really need to be deployed at multiple levels," Smith said. "If you can put a kill switch in three or four places, why only put it in one?"

Smith also acknowledged that Microsoft needs to adjust its climate targets. The company had previously committed to removing more carbon from the atmosphere than it emits by 2030. However, the use of natural gas power plants to supply energy for AI data centers has made this timeline more challenging. "We still believe climate is one of the most important issues of this century, but the question we're asking ourselves now is whether we can achieve those goals in the same way we envisioned six years ago," Smith stated. "I think we're discovering that we have to make some adjustments."