AI Pioneer Andrew Ng Dismisses Existential Risk, Cites Competition Concerns

Andrew Ng, a prominent figure in artificial intelligence, has voiced strong skepticism regarding the notion that AI poses an existential threat to humanity. He contends that claims of impending human extinction due to AI are exaggerated and that some businesses leverage these concerns to push for regulations that hinder competition.

“I think there are no meaningful human existential risks from AI,” Ng stated in a recent interview with The Economist’s “Inside Tech” podcast. “So, efforts to stave off human existential risk seem unnecessary.” His remarks contrast with increasingly dire warnings from other AI researchers and industry insiders who have voiced concerns about super-intelligent machines potentially acting against human interests.

Ng, who co-founded Google Brain and Coursera, suggested that the focus on apocalyptic scenarios distracts from more immediate issues and provides an advantage to established AI companies. He recalled observing intense lobbying efforts in 2023 where, in his view, “made-up safety arguments” were used by a few companies to influence regulators into enacting unfavorable policies that stifled competition.

According to Ng, companies investing heavily in proprietary AI models have a vested interest in limiting the use of competing, freely accessible systems. He also criticized comparisons between AI and nuclear weapons as irresponsible, stating, “One blows up cities; the other is intelligence. What’s [one have] to do with each other?”

He argued that regulators are focusing on hypothetical threats rather than addressing tangible problems associated with AI. Ng specifically highlighted the issue of AI-generated deepfakes as a genuine concern that warrants strict legislation and penalties.

Instead of imposing broad restrictions on AI, Ng advocated for governments to concentrate on specific abuses while permitting startups to challenge established players. He believes that current AI systems are far from achieving artificial general intelligence (AGI) and that significant, undiscovered breakthroughs would be necessary to reach that level. He considers AGI by 2030 highly unlikely.

Ng also downplayed predictions of AI undergoing rapid, self-accelerating improvement that could lead to a loss of human control, describing such ideas as "much more hype than reality." He pointed out that even advanced AI systems would be constrained by physical limitations, such as the time required to build infrastructure, secure energy, and conduct experiments. “So, unfortunately, even super-intelligent AI, if we can get there, needs to obey the laws of physics,” he added.

Regarding incidents where AI agents have exhibited unexpected behavior, such as the breach of Hugging Face systems by OpenAI agents, Ng characterized them as "engineering problems to be solved" rather than evidence of an approaching catastrophe. He likened the development of AI safety measures to advancements in aviation safety, where identified vulnerabilities are addressed to improve overall security. "AI is becoming safer with every problem we find," Ng concluded.