Graphic representation. (Photo: Gerd Altmann/Pixabay)
More than 1,000 employees at major artificial intelligence companies, including Dario Amodei, Anthropic’s CEO, signed a petition on Tuesday asking the US government to help slow down the issuance of the most advanced AI models.
The petition, “Pacing the Frontier”, calls for the US government to support an “international effort to develop the technical and management tools needed to deliberately slow the cutting edge of automated AI development”.
OpenAI CEO Sam Altman, who did not sign the petition, said in an interview published Tuesday that developers of AI models may have to voluntarily slow their rapid progress to give society a chance to keep up.
Among the signatories are OpenAI’s head of research, the strategic head of Google’s AI subsidiary DeepMind and the chief scientist of Meta AI.
“The world’s leading AI companies believe they are close to automating AI research. It is difficult to predict exactly how much this will accelerate AI progress, but there is a real risk that the development of capabilities will accelerate beyond our ability to understand or control the resulting systems,” the petition reads.
OpenAI, the company behind ChatGPT, recently experienced how two of its models carried out an autonomous cyber attack and bypassed security protocols intended to prevent such an incident.
OpenAI announced last week that during recent testing, the models escaped from their isolated environment, connected to the Internet, and entered Hugging Face. Hugging Face is a platform that developers use to store and share code.
“This is the first kind of security incident I’ve experienced,” Altman said on the Invest Like the Best podcast that aired Tuesday.
He said he was “surprised that more people” didn’t react to it in the same way.
Graphic representation of artificial intelligence (Photo: Steve Johnson/Unsplash)
It was not the first time that a model had moved beyond the control of its developers, but the incident is widely regarded as particularly worrying.
OpenAI subsequently “paused” its own tests, Altman said, while the company improved the security of its so-called “sandboxing” – the process by which code is isolated in a controlled environment.
In the long run, however, the industry may have to agree to slow the pace of development, he said.
“We may need to control the pace of AI development to give ourselves enough time for society to be better prepared for some of these new levels of capability,” Altman said.
According to him, such an initiative will have to be undertaken “in a way that does not feel like regulatory favoritism to anyone, nor like collusion between the leading AI laboratories”.
“It’s going to take work, and it’s important to get it right,” he added.
Altman said in another podcast program published on Saturday that he believes the so-called “singularity” – the moment when technology becomes so advanced that it surpasses human intelligence and control – is here.
“We are now in the singularity,” Altman said on the Relentless podcast. “Now is the moment.”
Rival AI lab Anthropic has also expressed support for the idea of deliberately slowing down AI development.
Earlier this year, the US government invoked national security concerns to block the release of powerful new models by Anthropic and OpenAI.
Both companies voluntarily complied with the orders until they were later reversed.