Title: ‘Godfather of AI’ Warns of Superintelligence and Structural Risks

Geoffrey Hinton admitted he initially struggled to accept his Nobel Prize victory. When the notification arrived in 2024, the 77-year-old computer scientist performed a mental probability check. He questioned the likelihood of a theoretical psychologist, who had retreated into the field of computer science, receiving the Nobel Prize in Physics. “I calculated it as perhaps one in two million,” he recalled addressing the audience at the Sana AI Summit in New York last week. He then considered the possibility of a dream, estimating the odds similarly at one in two million. This led to a grim deduction: “That implies it is a million times more probable that this is a dream than reality.” The room erupted in laughter, but Hinton was not finished. He revealed that for days following the announcement, he half-anticipated waking up from the event. The only solace he could find was a dark joke about his political anxieties: “If it were a dream, I would wake up, and that nightmare involving Trump’s presidency would not be real.” After a pause, he added, “I would gladly sacrifice the prize for that,” a remark that resonated deeply given his stated belief that there is a 10% to 20% probability that AI will lead to human extinction within three decades, and that artificial intelligence will surpass human intellect before he dies.

Hinton engaged in a dialogue with Joel Hellermark, the 29-year-old founder and CEO of Sana AI, who had previously promised him the Nobel Prize during an earlier appearance on the same stage. This time, Hellermark posed more challenging questions. Hinton responded in his signature leisurely manner, interspersed with humor, delivering answers that were both technically exact and profoundly expansive. The central theme of their discussion was that humanity is not merely constructing new entities, but ones that will vastly exceed human cognitive capabilities. Furthermore, Hinton emphasized that time is running out to determine the nature of these beings. “My estimate is that they will become significantly more intelligent than we are,” Hinton stated. He predicted that humans would never again compete with AI in games like Go or chess, pointing to the technology’s rapid advancements in mathematics as evidence.

Hinton’s comments were grounded in recent developments. To his clear satisfaction, an AI system had successfully proven one of Paul Erdős’ mathematical theorems by utilizing a branch of mathematics that researchers had previously overlooked. Hinton viewed this as a pivotal moment. He explained that in closed systems such as mathematics, AI possesses the ability to formulate its own conjectures, test them, learn from errors, and accumulate knowledge continuously. This mirrors the evolution of AlphaGo, which moved from copying expert strategies to destroying them once it began generating its own training data. Hinton argued that language models are following a similar path. The crucial mechanism, he noted, involves providing a model with initial beliefs; if the model reasons toward a conclusion that contradicts those beliefs, an inconsistency arises. This inconsistency serves as a training signal that does not require additional data to leverage. “This suggests that language models can achieve massive increases in intelligence without needing substantially more data,” he said, noting that fellow Nobel laureate Demis Hassabis shares this view. Hinton forecasts that AI will surpass the world’s top mathematicians within ten years. Looking further ahead, he believes the gap between current AI and figures like Albert Einstein will close within the next two decades, even if it does not happen immediately.

This marks a shift in Hinton’s narrative from mere prediction to urgent warning, a stance he has refined over nearly three years. When he departed Google in 2023, expressing regret over his contributions, the primary concern focused on malicious actors and the potential loss of human control. By 2025, however, his perspective had evolved to highlight structural economic issues. He argued that AI would trigger widespread job losses while concentrating wealth, not due to inherent flaws in the technology, but because of the economic framework utilizing it. “What will actually occur is that wealthy individuals will employ AI to displace workers,” he stated in September. “That is not a fault of AI. That is the capitalist system.” In August, he suggested that technology firms should program AI with “maternal instincts,” deliberately embedding safety features.

Source: Yahoo News Generated at: 2026-06-02 10:41:29 UTC