US, Britain announce partnership on AI safety, testing

This post was originally published on this site

WASHINGTON (Reuters) -The United States and Britain on Monday announced a new partnership on the science of artificial intelligence safety, amid growing concerns about upcoming next-generation versions.

Commerce Secretary Gina Raimondo and British Technology Secretary Michelle Donelan signed a memorandum of understanding in Washington to jointly develop advanced AI model testing, following commitments announced at an AI Safety Summit in Bletchley Park in November.

“We all know AI is the defining technology of our generation,” Raimondo said. “This partnership will accelerate both of our institutes work across the full spectrum to address the risks of our national security concerns and the concerns of our broader society.”

Britain and the United States are among countries establishing government-led AI safety institutes.

Britain said in October its institute would examine and test new types of AI, while the United States said in November it was launching its own safety institute to evaluate risks from so-called frontier AI models and is now working with 200 companies and entites.

Under the formal partnership, Britain and the United States plan to perform at least one joint testing exercise on a publicly accessible model and are considering exploring personnel exchanges between the institutes. Both are working to develop similar partnerships with other countries to promote AI safety.

“This is the first agreement of its kind anywhere in the world,” Donelan said. “AI is already an extraordinary force for good in our society, and has vast potential to tackle some of the world’s biggest challenges, but only if we are able to grip those risks.”

Generative AI – which can create text, photos and videos in response to open-ended prompts – has spurred excitement as well as fears it could make some jobs obsolete, upend elections and potentially overpower humans and catastrophic effects.

In a joint interview with Reuters Monday, Raimondo and Donelan urgent joint action was needed to address AI risks.

“Time is of the essence because the next set of models are about to be released, which will be much, much more capable,” Donelan said. “We have a focus one the areas that we are dividing and conquering and really specializing.”

Raimondo said she would raise AI issues at a meeting of the U.S.-EU Trade and Technology Council in Belgium Thursday.

The Biden administration plans to soon announce additions to its AI team, Raimondo said. “We are pulling in the full resources of the U.S. government.”

Both countries plan to share key information on capabilities and risks associated with AI models and systems and technical research on AI safety and security.

In October, Biden signed an executive order that aims to reduce the risks of AI. In January, the Commerce Department said it was proposing to require U.S. cloud companies to determine whether foreign entities are accessing U.S. data centers to train AI models.

Britain said in February it would spend more than 100 million pounds ($125.5 million) to launch nine new research hubs and AI train regulators about the technology.

Raimondo said she was especially concerned about the threat of AI applied to bioterrorism or a nuclear war simulation.

“Those are the things where the consequences could be catastrophic and so we really have to have zero tolerance for some of these models being used for that capability,” she said.