A University of Washington professor has played down warnings that artificial intelligence could wipe out humanity by the end of the decade, saying the more immediate danger is the way the technology is already influencing people’s lives.
Chirag Shah, an adjunct professor at the university’s Information School, was speaking after former Anthropic researcher Jacob Coxon resigned and warned that people developing AI “earnestly believe” it could “kill us all by the end of the decade”.
Shah said a takeover scenario remained highly unlikely, but argued that concerns over poorly tested systems, unchecked automation and the erosion of human judgement deserved far more attention.
“I think AI killing us all is a very far-fetched scenario. I don’t think that’s very likely,” he said.
AI risks are already affecting people
According to Shah, artificial intelligence is increasingly being used to assist with decisions involving job applicants, university admissions and claims, sometimes without sufficient testing or human oversight.
“There are a number of things where AI is being used in a way without any oversight or without enough testing, and it’s starting to impact real human lives,” he said.
He said the risks became harder to manage as AI moved beyond producing text and images and began carrying out tasks on behalf of users.
Shah’s research includes “agentic AI” – systems designed to pursue goals by taking actions such as opening accounts, shopping, completing forms and making decisions.
Such systems may perform correctly most of the time, he said, but people cannot always predict when they will make a harmful decision.
“This is the thing that maybe 95% of the time they do the right thing. Five percent of time, something bad happens,” Shah said. “The problem is we don’t know when that 5% happens.”
He added that AI systems could not necessarily be fully predicted, controlled or made safe after something went wrong.
Warning over dependence on artificial intelligence
Shah said the debate should also focus on what happens when people rely on AI for skills that once required education, professional training and critical thinking.
He was particularly concerned about schools and universities, where students and teachers are adapting to tools capable of producing written work and completing assignments.
Comparing the issue with learning multiplication tables despite the widespread availability of calculators, he said some abilities remained important for human development even when technology could perform a task more efficiently.
“AI doesn’t need to destroy us, right?” Shah said. “We are slowly hacking away our foundations of education, foundations of our professional training, foundation of democracy.”
“Critical thinking is more important than ever before,” he added.
Shah’s comments come amid a wider argument over the pace of AI development. Coxon, who has worked in research at Anthropic and OpenAI, has accused leading companies of pressing ahead with increasingly powerful systems while failing to resolve fundamental safety questions.
Anthropic has said it is transparent about the potential benefits and unprecedented risks of AI and that it continues to develop safeguards and test its models for dangers including cybersecurity and biological misuse.
US-China competition complicates calls for caution
Shah said efforts to slow the development of artificial intelligence were also complicated by competition between the United States and China.
Both countries regard AI as a national security priority, he said, giving them little incentive to pause if they believe doing so could allow a rival to gain an advantage.
“If they want to win this race, why slow down?” Shah said.
Despite his concerns, he said the future of AI was not predetermined and that governments, companies and the public still had a role in deciding how the technology should be used.
“AI is not inevitable, even though yes, you know, the genie is out of the bottle, we can’t put it back,” Shah said. “But what happens with AI? It’s still in our hands.”
He said the key questions were who should be held responsible when an AI system caused harm, how its decisions could be scrutinised and how people could preserve their ability to learn, think and make decisions for themselves.
