By contrast, the latest statement was endorsed by Microsoft’s chief technology and science officers, as well as Demis Hassabis, CEO of Google’s AI research lab DeepMind, and two Google executives who lead its AI policy efforts. The statement doesn’t propose specific remedies but some, including Altman, have proposed an international regulator along the lines of the U.N. nuclear agency.
Some critics have complained that dire warnings about existential risks voiced by makers of AI have contributed to hyping up the capabilities of their products and distracting from calls for more immediate regulations to rein in their real-world problems.
Hendrycks said there’s no reason why society can’t manage the “urgent, ongoing harms” of products that generate new text or images, while also starting to address the “potential catastrophes around the corner.” He compared it to nuclear scientists in the 1930s warning people to be careful even though “we haven’t quite developed the bomb yet.” “Nobody is saying that GPT-4 or ChatGPT today is causing these sorts of concerns,” Hendrycks said. “We’re trying to address these risks before they happen rather than try and address catastrophes after the fact.” The letter also was signed by experts in nuclear science, pandemics and climate change. Among the signatories is the writer Bill McKibben, who sounded the alarm on global warming in his 1989 book “The End of Nature” and warned about AI and companion technologies two decades ago in another book. “Given our failure to heed the early warnings about climate change 35 years ago, it feels to me as if it would be smart to actually think this one through before it’s all a done deal,” he said by email Tuesday. An academic who helped push for the letter said he used to be mocked for his concerns about AI existential risk, even as rapid advancements in machine-learning research over the past decade have exceeded many people’s expectations. David Krueger, an assistant computer science professor at the University of Cambridge, said some of the hesitation in speaking out is that scientists don’t want to be seen as suggesting AI “consciousness or AI doing something magic,” but he said AI systems don’t need to be self-aware or setting their own goals to pose a threat to humanity. “I’m not wedded to some particular kind of risk. I think there’s a lot of different ways for things to go badly,” Krueger said. “But I think the one that is historically the most controversial is risk of extinction, specifically by AI systems that get out of control.”]]>
