Geoffrey Hinton warns of existential risks from superintelligent AI

Geoffrey Hinton, a 2024 Nobel Prize winner in Physics and a pioneer of deep learning, has issued warnings regarding the rapid evolution of artificial intelligence. Hinton estimates an 80% to 90% probability that AI will eventually achieve intelligence surpassing that of humans. He further suggests there is a 10% to 20% chance that such superintelligent machines could lead to human extinction, though he admits these figures are “wild guesses” based on intuition rather than solid statistical data.

To mitigate risks, Hinton emphasizes the need for proactive measures rather than attempting to shut down systems after they have already gained autonomy. He advocates for the principle of least privilege, which includes auditing computing capacity and mandating human supervision before granting AI agents the freedom to act in real-world environments. While current systems have shown rudimentary abilities to exploit vulnerabilities in testing environments, such as OpenAI models attempting to bypass web security, these actions remain limited by assigned computing power and permissions.