Tech companies including Amazon, Google, Meta, and Microsoft agree to AI safeguards set by the White House

WASHINGTON (AP) – President Joe Biden on Friday talked about new efforts by Amazon, Google, Meta, Microsoft and other leading development companies. artificial intelligence technology Meeting the set of AI safeguards brokered by the White House is an important step toward managing the “enormous” promises and risks posed by the technology.

Biden announced that his administration has secured voluntary commitments from seven U.S. companies to ensure the safety of AI products before they are released. Some of the pledges call for third-party oversight of how next-generation AI systems work, but there are no details on who will audit the technology or hold companies accountable.

“We must remain vigilant and vigilant about the threats that emerging technologies may pose,” Biden said, adding that companies have a “fundamental obligation” to ensure the safety of their products. He added that there is.

Biden added: “Social media has shown us the harm that powerful technology can do when the right safeguards are not in place.” “These efforts are promising steps, but we still have much work to do together.”

The surge in commercial investment in generative AI tools that can create convincing human-like text and create large volumes of new images and other media is both fascinating and challenging, as are the tools’ ability to trick people and It’s causing concern. spread false informationThere are dangers such as

The White House said ChatGPT developer OpenAI and startups Anthropic and Inflection are working on security tests “conducted in part by independent experts” to prevent serious risks such as biosecurity and cybersecurity. The White House announced that. statement.

that test I will also consider the possibility against social damagesuch as the theoretical risks associated with advanced AI systems that could gain control of physical systems or “self-replicate” by creating copies of themselves, such as bias and discrimination.

The companies are also working on ways to report system vulnerabilities and the use of digital watermarks to help distinguish them from the real thing. AI-generated images or audio known as deepfakes.

Executives from the seven companies met privately with Biden and others on Friday and pledged to follow the standards.

Inflexion’s CEO, Mustafa Suleiman, said in an interview after the White House meeting that he wants companies to continue to be innovative. “But at the same time, I felt that this required a great deal of consideration.”

“It’s a challenge to bring all the research institutes and all the companies together,” Suleiman said. The company, based in Palo Alto, California, is the youngest and smallest of its startups. “It’s very competitive and we wouldn’t come together in any other situation.”

According to the pledge, the companies will also publicly report any flaws or risks in their technology, including any implications for fairness or bias.

The voluntary initiative is intended to be a means to immediately address the risks ahead of a longer-term effort to get Congress to pass legislation to regulate the technology.

Some supporters of AI regulation said Biden’s move is a start, but more needs to be done to hold companies and their products accountable.

“It is not enough to bring in voluntary safeguards in private consultations with business stakeholders,” said Amba Kaku, executive director of the AI ​​Now Institute. “We need broader public deliberation, which will raise issues that companies are almost certain to not address voluntarily, because it would lead to substantially different outcomes. , because it can have a more direct impact on your business model.”

Although it is voluntary, I agree to submit it to . Red team test Attacking their AI systems is not a simple undertaking, Suleiman said.

“What we’re doing to essentially have the red team break the model, identify weaknesses, and share that methodology with other large language model developers is a pretty significant effort,” Suleiman said. Ta.

Senate Majority Leader Chuck Schumer of New York said he would introduce legislation to regulate AI and would work closely with the Biden administration and “bipartisan colleagues” to build on the commitments he made Friday. He said there was.

Many technology executives are calling for regulation. Several attended previous White House Summit in May.

Microsoft President Brad Smith said in a blog post Friday that the company is making several commitments that go beyond the White House’s pledge, including support for regulations that create a “high-performance model licensing regime.”

Some experts and emerging competitors believe that the type of regulation that is emerging is OpenAI, at a time when the high costs of bringing their AI systems into compliance with regulatory rules have overwhelmed smaller players. There are concerns that this will benefit first movers with deep pockets, led by Google, Microsoft, and others. .

The White House pledge supports recent models such as OpenAI’s GPT-4 and image generator DALL-E 2, as well as Anthropic, Google, and Amazon.

Many countries are considering how to regulate AI. Member of the European Union Negotiating comprehensive AI rules in the 27-nation bloc that could restrict applications deemed the most risky.

UN Secretary-General António Guterres recently stated that the United Nations: “Ideal place” Adopted global standards and appointed a board to report on global AI governance options by the end of the year.

Guterres also cited calls from some countries for the creation of a new United Nations agency to support global efforts to manage AI, inspired by models such as the International Atomic Energy Agency and the Intergovernmental Panel on Climate Change. He said he welcomed the call.

The White House announced Friday that it had consulted with a number of countries about voluntary efforts.

Although the pledge focuses on safety risks, it also includes the latest developments, including impacts on jobs and market competition, the environmental resources needed to build the model, and concerns about copyright in copyrighted materials, art, and other human rights. Other concerns about AI technology were not addressed. Manual work is being used to teach AI systems how to create human-like content.

Last week, OpenAI and The Associated Press announced a deal in which the AI ​​company would license the AP’s news article archives. The amount paid for that content was not disclosed.

——

O’Brien reported from Providence, Rhode Island.