Artificial intelligence pioneer Andrew Ng is dismissing fears that AI could wipe out humanity — and accusing some businesses of using “made-up safety arguments” to push regulations aimed at crushing competition.
“I think there are no meaningful human existential risks from AI,” Ng told The Economist’s “Inside Tech” podcast in an interview published Tuesday.
“So, efforts to stave off human existential risk seem unnecessary.”
The blunt remarks came against a backdrop of increasingly dire warnings from AI researchers and industry insiders — some of whom have embraced so-called “doomerism” and predict that super-intelligent machines could go rogue, seize control of critical infrastructure and ultimately bring about human extinction.
Ng, a former Stanford prof who helped launch the pioneering Google Brain project and co-founded online education giant Coursera, argued that the apocalyptic rhetoric has distracted policymakers from genuine problems while giving established AI companies an opportunity to hamstring smaller rivals.
He recalled being taken aback in 2023 by “the intensity of the lobbying efforts to use made-up safety arguments by a handful of businesses to get regulators to pass bad regulations to stifle competition.”
Ng suggested that companies spending billions of dollars developing proprietary AI models have an incentive to restrict competing systems that can be downloaded and used for free.
He also blasted attempts to compare artificial intelligence with nuclear weapons, calling the analogy irresponsible.
“One blows up cities; the other is intelligence. What’s [one have] to do with each other?” Ng said.
He warned that “regulators are distracted fighting phantoms rather than looking at the actual problems of AI.”
Ng pointed specifically to AI-generated, non-consensual intimate deepfakes as an abuse that warrants tough legislation and penalties.
Rather than restrict a technology with overwhelmingly beneficial applications, Ng argued that governments should focus on specific abuses while allowing startups to challenge established players.
He argued that today’s systems remain a long way from achieving artificial general intelligence, or AGI — the ability to perform any intellectual task a human can — despite enormous advances in recent years.
Getting there would require breakthroughs that have yet to be discovered, Ng said, adding that he considers reaching that level by 2030 highly unlikely.
He also poured cold water on predictions that AI could rapidly improve itself in an accelerating cycle that eventually leaves humanity powerless to control it.
Such notions of compounding progress and “recursive” self-improvement are “much more hype than reality,” he said.
Even an extraordinarily powerful AI system would face practical obstacles, including the time needed to construct data centers, secure electricity and conduct experiments.
“So, unfortunately, even super-intelligent AI, if we can get there, needs to obey the laws of physics,” Ng said.
He also disputed the notion that recent incidents involving rogue AI agents offer evidence of an approaching catastrophe.
Ng cited a widely publicized episode in which a swarm of experimental OpenAI agents breached the systems of rival AI platform Hugging Face — an incident that prompted a Senate investigation.
Rather than viewing the breach as evidence of an existential threat, he characterized it as “an engineering problem to be solved.”
Ng compared the development of AI safety protections to the history of commercial aviation, which became dramatically safer as engineers identified vulnerabilities and corrected them.
The techie argued that the same process is already underway as developers strengthen security barriers and monitoring systems to prevent autonomous agents from escaping their intended operating environments.
“AI is becoming safer with every problem we find,” he said.
His skepticism stands in sharp contrast to warnings from prominent researchers who have argued that the industry is racing toward a potentially irreversible disaster.
Former Anthropic researcher Jacob Coxon warned after quitting the company that “AI can kill us all by the end of the decade.”
“These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources,” Coxon said.
Anthropic researcher Evan Hubinger likewise wrote on X in September that he believes there is a greater than 10% chance AI will kill every human being within the next decade.
Other researchers have outlined scenarios in which advanced AI systems could cripple electrical grids, commandeer military technology or help engineer deadly biological weapons.
