AI Researcher Says Humanity May Have Less Than a Decade
Published on October 30, Soares’s warning centers on the possibility that increasingly capable AI could exceed human control. He said he stopped contributing to his retirement savings about a decade ago because he believes humanity may not have much time left. He still has a 401(k) from his years at Google, he said, but has not added to it in years.
Soares, president of the Machine Intelligence Research Institute, said he does not consider human extinction inevitable. He described trying to prevent it while also living a good life, comparing the fear to earlier generations’ anxieties about nuclear destruction, war and disease. He is not primarily concerned that AI will eliminate jobs; his stated fear is that it could kill everyone.
During the podcast, host Nicholas Thompson questioned the prediction, pointing to limitations in current AI systems. Soares responded with an analogy to horses witnessing the development of cars: early limitations did not prevent the technology from becoming more capable. He said AI systems are taking on longer tasks as they improve.
Soares distinguished between present-day uses, such as seeking cancer treatments, which he described as acceptable, and efforts to scale AI to imitate and surpass human intelligence. He argued that this kind of development should not be pursued. His proposed safeguard is to limit that scaling, rather than to abandon every use of AI.
Soares Warns AI Race Could Put Human Survival at Risk
Nate Soares is warning that the drive to make AI more powerful could carry a cost far beyond lost jobs: human extinction. The former Google and Microsoft engineer puts the chance of AI wiping out humanity within ten years above 50 percent, and says he stopped contributing to retirement savings about a decade ago because he believes time may be short.
Soares, who leads the Machine Intelligence Research Institute, does not say catastrophe is inevitable. He says he is working to prevent it, and argues that the danger lies in pushing AI to imitate and surpass human intelligence. He distinguishes that ambition from current applications such as seeking cancer treatments, which he considers acceptable.
Host Nicholas Thompson challenged Soares by pointing to the limitations of current AI. Soares countered that early shortcomings did not stop cars from replacing horses, and said AI is improving and handling longer tasks. That is precisely why the choice to scale these systems matters: limitations today are no guarantee of safety tomorrow.
Soares’s answer is to limit AI’s scaling, not to pursue ever more powerful systems and assume control will follow. He says the technology’s development need not end in human extinction, but that preventing it requires changing course before imitation and surpassing become the goal.
A Dire AI Forecast Is Not Proof of an Impending Apocalypse
Nate Soares has made an extraordinary prediction: AI may wipe out humanity within a decade, with the odds, in his view, above 50 percent. He says the fear has shaped his personal finances, as he stopped contributing to retirement savings about ten years ago. But a stark forecast remains a forecast, not a demonstrated outcome.
Soares, a former Google and Microsoft engineer who now heads the Machine Intelligence Research Institute, argues that AI’s growing capabilities could become existentially dangerous if developers scale systems to imitate and surpass human intelligence. He says present-day uses such as seeking cancer treatments are acceptable, and that extinction is not inevitable. His proposed answer is to limit scaling.
Nicholas Thompson pushed back on the prediction by noting that current systems still have shortcomings. Soares replied that cars once had limits too, and that AI is improving and taking on longer tasks. The analogy explains his concern, but the podcast also captures the gap between a warning about future capabilities and what current systems can do.
Soares says society should avoid building AI that surpasses human intelligence. That case deserves to be weighed against the evidence and the technology’s present limitations, rather than treating the most alarming prediction as settled fact.
Reported by New York Post, Daily Mail US.
