The world is not ready for advances in artificial intelligence, according to a group of high-level experts, including two “godfathers” of AI, who warn that governments have not made enough progress in regulating the technology.
A shift by technology companies toward autonomous systems could “massively amplify” the impact of AI and governments need safety regimes that trigger regulatory action if products reach certain levels of capability, the group said.
The recommendations are made by 25 experts, including Geoffrey Hinton and Yoshua Bengio, two of the three “godfathers of AI” who have won the ACM Turing Award, the computer science equivalent of the Nobel Prize, for their work.
The intervention comes as politicians, experts and technology executives prepare to meet for a two-day summit in Seoul on Tuesday.
He academic articleCalled “Managing AI Extreme Risks Amid Rapid Progress,” it recommends government security frameworks introduce stricter requirements if the technology advances rapidly.
It also calls for increased funding for newly created bodies such as the UK and US AI safety institutes; force technology companies to carry out more rigorous risk controls; and restrict the use of autonomous AI systems in key social functions.
“Society’s response, despite promising first steps, is incommensurable with the possibility of rapid and transformative progress that many experts expect,” according to the article published in the journal Science on Monday. “Research into AI safety is overdue. “Current governance initiatives lack mechanisms and institutions to prevent misuse and recklessness, and barely address autonomous systems.”
A global summit on AI safety held last year at Bletchley Park in the UK negotiated a voluntary testing agreement with tech companies including Google, Microsoft and Mark Zuckerberg’s Meta, while the EU has enacted a law on AI and in the US, a White House executive order has established new AI safety requirements.
The document says that advanced artificial intelligence systems (technology that performs tasks typically associated with intelligent beings) could help cure diseases and raise living standards, but also carry the threat of eroding social stability and enabling automated warfare. . He warns, however, that the technology industry’s move toward the development of autonomous systems poses an even greater threat.
“Companies are shifting their focus toward developing generalist AI systems that can act and pursue goals autonomously. “Increases in capabilities and autonomy could soon greatly amplify the impact of AI, with risks including large-scale societal harm, malicious uses, and an irreversible loss of human control over autonomous AI systems,” the experts said, and They added that the unbridled advance of AI could lead to the “marginalization or extinction of humanity.”
The next stage in the development of commercial AI is “agent” AI, a term for systems that can act autonomously and, in theory, carry out and complete tasks such as booking vacations.
after newsletter promotion
Last week, two tech companies gave a glimpse into that future with OpenAI’s GPT-4o, which can conduct real-time voice conversations, and Google’s Project Astra, which was able to use a smartphone camera to identify locations, read and explain computer codes. and create alliterative sentences.
Other co-authors of the proposals include best-selling author Sapiens, Yuval Noah Harari, the late Daniel Kahneman, a Nobel laureate in economics, Sheila McIlraith, a professor of AI at the University of Toronto, and Dawn Song, a professor at the University of Toronto. University of California, Berkeley. The article published Monday is a peer-reviewed update of initial proposals produced before the Bletchley meeting.
A UK government spokesperson said: “We disagree with this assessment. “This week’s Seoul AI Summit will play an important role in advancing the legacy of the Bletchley Park summit and will see a number of companies update world leaders on how they are delivering on the commitments made at Bletchley to ensure the safety of their models.” .