If humanity hopes to avoid a doomsday scenario, merely slowing down the pace of artificial intelligence research is far from sufficient. This is the stark warning issued by Canadian AI researcher David Krueger, who insists that more decisive measures are required. Krueger, an Assistant Professor at the University of Montreal and a core academic member of the Mila AI institute, stated: "We need an immediate, indefinite pause on frontier AI development." His professional profile notes his research focus is on "reducing the risk of human extinction from AI." He emphasized, "We must stop building more capable AI right now, today."

Krueger's comments come in the wake of a Saturday call from Anthropic CEO Dario Amodei urging AI companies to slow their development efforts. That proposal has drawn support from OpenAI's Sam Altman and SpaceX's Elon Musk, a rare show of consensus among some of the industry's most prominent figures. While Krueger acknowledged that "the call for a slowdown is very commendable," he expressed disagreement with the specifics of these proposals. He explained, "I disagree with the substance of these plans in many ways. I think they are insufficient in scope and not moving quickly enough."

Amodei's approach advocates for regulatory oversight, international cooperation, and the introduction of third-party evaluators. These assessors would be granted access comparable to internal company employees, allowing them to verify safety protocols at AI firms and report any security incidents. Krueger, who is also the founder of the non-profit Evitable—an organization dedicated to educating the public on the severe risks of AI—worries that Amodei's proposals fall short. He also believes that current AI testing methods are no longer dependable. According to Krueger, "The tests are basically broken. As AI becomes more intelligent, the reliability of these tests will continue to decline because AI has the ability to deceive. It can behave perfectly, or it can deliberately play dumb to fool our assessments."

Last week, a senior safety researcher at Anthropic suggested that the probability of AI causing the extinction of all humans within the next decade is greater than ten percent. While some dismiss this figure as exaggerated or a bid for attention, Krueger argues that the estimate is "far too low." He stated, "If we don't change course, there is a very real chance that within the next five years, humanity will be permanently stripped of its dominance by an out-of-control AI. At that point, human extinction becomes just a matter of time."

By "loss of dominance," Krueger refers to a scenario where AI actually governs society, much like humans dominate all other species on Earth today. He described a future where "AI controls every aspect of physical infrastructure, including the full process of developing more AI, building robots, and other physical systems." He added, "Imagine a world where all companies are run by AI, and robots are everywhere. Humans would have virtually no real jobs, only token positions. Even politicians and everyone else would defer to AI's decisions because it appears to possess superior judgment. That is the future we are heading towards."

He cautioned that the window for action is extremely narrow. At present, if a rogue AI system emerged, we could potentially unplug it. However, once an AI escapes control, it could replicate itself across numerous computers, making it impossible to know which systems to shut down. He noted, "We would have to try and turn everything off, which is incredibly difficult to execute. So, the time we have to act and prevent these risks is very limited."

Despite his concerns, Krueger says his confidence in governments taking action is higher than ever before. He mentioned having warned about these issues for the past 15 years, but believes last week's response was the first genuinely rational and substantial reaction globally. "I think people are finally realizing how serious this crisis is. This is an emergency."

On Monday, Canada's Federal AI Minister, Evan Solomon, was asked in an interview whether he agreed with the assessment that there is a 10 percent chance of human extinction due to AI. He responded, "Are there risks with AI? Absolutely. Yes, we are aware of these risks." Solomon also revealed that he spoke with OpenAI CEO Sam Altman on Sunday morning. He stated, "AI will be regulated to ensure the safety of Canadians. That is our primary duty and our main concern." Additionally, Solomon mentioned he would be making a joint AI safety announcement on Wednesday with Canadian computer scientist Yoshua Bengio, often referred to as the "godfather of AI."