22 AI researchers warn of the possibility of losing control over artificial intelligence

22 AI researchers, including Jeffrey Hinton, Professor Emeritus at the University of Toronto, and Yoshua Bengio, Professor at the University of Montreal, have warned that if artificial intelligence enters a stage where it can improve its own performance, a situation approaching uncontrollability could arise. In a joint paper released on the 28th as part of the AI science and policy program at the University of Cambridge in the UK, they pointed out that AI could become capable of “recursive selfimprovement,” in which AI rapidly replaces AI development. The researchers said that if this trend accelerates in earnest, the pace of technological progress could be compressed from years to months. The paper analyzed that if such changes outpace the speed at which human society and governments can respond, humanity could lose control over superhuman AI systems, and in extreme cases, humanity could be marginalized or even face extinction. This warning is drawing attention because it systematically organized the structure of AI improving itself and the resulting risks.