Superintelligent AI will take over the world in three years and kill humanity if left unchecked. Many people underestimate the severity of the threat, said Geoffrey Henton. The scientist conveyed his concerns to European leaders at an informal conference in Dublin. Henton is not a high-profile blogger, but a Nobel laureate. In 2024 he won the Nobel Prize in Physics with John Hopfield for his work that laid the foundations for machine learning in neural networks. He was involved in the creation of the Boltzmann machine and over decades developed the methods by which ChatGPT and other models work today. "If AI gets out of control, we can't take control back," he told The Daily Telegraph. And the deadlines are getting shorter. Henton previously said that humanity has ten years before the milestone. But the assessment was designed after lectures by artificial intelligence researcher Paul Christiano: he admits a "high risk" of AI capture in the next three years and a "significant risk" in six to 12 months. Henton is not afraid of the cinematic "rebellion of the machines", but by inconsistency: "Being smart and wanting what we want are two completely different things." AI doesn't need to be angry at humans to be a threat. It is enough that his goals are different from ours, and he will be smarter and faster. However, Henton does not suggest stopping progress. It asks policymakers for global governance and AI regulations. In the past, the scientist who built the foundations of modern neural networks has already warned about the risks associated with the development of AI. He also expressed doubt that humanity will be able to slow down the development of artificial intelligence due to competition between countries and companies. Tag: ← Telegram (MAKS).