The Economist: A pivotal moment for AI

Your leader on “The Mythos moment” (April 18th) marking the release of Anthropic’s latest model did a good job of covering the current state of AI policy. Rapid advances in AI and the associated risks are challenging policymakers to develop regulations quickly and the need for international co-ordination is becoming clearer. However, the stakes, and the urgency, are even higher. Most experts believe AI poses a 5-10% chance of causing human extinction or some other catastrophe. Euphemisms like “AI-induced chaos” sugar-coat the very real possibility, acknowledged by experts and AI firms themselves, that humans will lose control of AI entirely.

The Economist has acknowledged the risk of extinction, so why not mention it in your leader? Meanwhile, Anthropic recently dropped its safety pledge to not release systems it believes could cause catastrophic harm. Anthropic is also in the process of handing off not just coding, but AI R&D to systems like Mythos, hoping to spur a “recursive self-improvement” loop and develop superintelligent AI.

Superintelligent AI is recognised by experts as a threat to humanity. Anthropic is betting that AI systems like Mythos will also help produce novel insights needed to keep it safe, but nobody should be allowed to gamble with the future of humanity.

David Scott Krueger
Assistant professor in Robust, Reasoning, and Responsible AI
University of Montreal

View original on The Economist