Seven leading artificial intelligence companies have agreed to a handful of industry best practices, a first step toward more meaningful regulation, the White House announced Thursday.
The companies (Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI) agreed to principles that include security, transparency with the public, and internal testing of their products before introducing them to the public.
In a call with reporters Wednesday night ahead of the announcement, a White House official, who asked not to be named as part of the terms of the call, said President Joe Biden will eventually sign an executive order that will more strictly regulate AI, though officials are still working out the details.
“The White House is actively developing executive action to govern the use of AI for the president’s consideration,” the official said. “This is a high priority for the president.”
Top executives from the companies involved will also meet with the White House on Friday, including Microsoft Chairman Brad Smith, Google and Alphabet President of Global Affairs Kent Walker, Amazon Web Services CEO Adam Selipsky, Meta President of Global Affairs Nick Clegg, OpenAI President Greg Brockman, Anthropic CEO Dario Amodei and Inflection AI CEO Mustafa Suleyman.
Each of the seven companies has agreed to hire independent experts to test their systems for vulnerabilities and share the information they discover with each other, governments and researchers, the official said. They also agreed to develop so-called “watermarking” mechanisms to help users identify when the content they watch or listen to is generated by AI.
Commitments are voluntary and non-binding.
Gary Marcus, a leading critic of artificial intelligence, said the commitments were important, but noted that they do not require AI companies to say what data they are using to train their models.
“I think it’s a great first step and it doesn’t go far enough,” Marcus said. “First of all, because it is voluntary. Second, one of the most important things we need here is what data is used to train the models, and that’s not part of this.”
“The red team is great. Information sharing between companies is fantastic. An agreement on the watermark is excellent. These are all good steps. But until we have real transparency around the data, we’re not done,” he added.
Generative AI systems like OpenAI’s ChatGPT became a sensation late last year, sparking a swarm of new users and leaving the US government scrambling to find an appropriate role. In May, OpenAI CEO Sam Altman courted members of Congress, asking them to regulate the industry. In June, Biden flew to San Francisco to meet with leaders from some of the Silicon Valley companies leading the way in AI in the US.
The White House official said the deal’s emphasis is on creating transparency that will allow experts to better scrutinize AI systems.
“What the companies are committing to is independent analysis by domain experts and the establishment of a broader, multifaceted regime to ensure that that analysis is credible and reliable,” the official said. “And they are also committing, more generally, to engaging with academia and civil society and US governments and others in establishing best practices to safeguard these systems and then adhering to those practices.”
