Microsoft president backs deployment of AI safety evaluators

Brad Smith said Microsoft participated in AI safety discussions with its close partner OpenAI, through a safety board staffed by both companies. Photo by F. Carter Smith/Bloomberg via Getty Images

Article content

Microsoft Corp. president Brad Smith said the company backs the use of independent evaluators for artificial intelligence.

Article content

Smith’s comments coincided with the United Nations General Assembly in New York, where safety — and options for regulating the industry globally — have been top of mind for world leaders and corporate chiefs.

Article content

Article content

Article content

Article content

“The model companies are obviously critical, but so are companies like ours that produce the layer of software that controls AI agents, that monitors what they do,” he said.

Article content

Article content

Smith said Microsoft participated in AI safety discussions with its close partner OpenAI, through a safety board staffed by both companies that helps determine when models are safe to release. He also cited Microsoft’s support for an off-switch to keep advanced systems under human controls.

Article content

“You really want safety to be built in multiple layers,” he said. “At times, you may want the government to be able to have the on-off switch, but why would you put that switch in only one place when you can put it in three or four?”

Article content

Smith, who earlier announced that Microsoft would spend an additional US$2 billion in the Middle East, said it was the right time to invest in the region, despite the ongoing United States-Iran war.

Article content

The company has already invested US$1.5 billion in G42, an Emirati tech firm, and agreed to use a planned AI data centre in the country. Smith confirmed in the interview that Microsoft still planned to use 200 megawatts of computing capacity at the facility.

Article content

Article content

He also acknowledged that Microsoft would have to make adjustments to its climate goals, which committed the company to pull more carbon from the atmosphere than it emits by 2030.

Article content

Article content

That timeline has appeared to slip out of reach, in part thanks to the company’s embrace of natural gas-fired plants being brought online quickly to serve AI data centres.

Article content

“We do continue to believe that climate remains one of the most important issues of this century, but we’re now asking, well, can we do it in the ways that we thought six years ago?” Smith said. “And I think what we’re finding is we’re going to have to make some adjustments.”

Article content

—With assistance from Mark Bergen.

Article content

Article content

Article content