Here’s How an AI Slowdown Could Actually Be Enforced
wired ·
AI researchers and leaders of major AI companies are increasingly discussing a slowdown or pause in frontier-model development because of concerns that advanced systems could become dangerous. The article argues that enforcing such a pause would be difficult, especially if companies or countries tried to advance secretly. Proposed measures include more independent model inspections, improved evaluations, monitoring cloud-computing activity, and tracking or restricting the use of advanced GPUs. Some researchers have suggested cryptographically secured chip records or remote “off switches,” though these ideas remain largely theoretical. International cooperation, potentially including a treaty with China to limit hardware growth, is considered essential but politically challenging. The uncertainty surrounding recursive self-improvement makes it important to monitor whether AI can increasingly conduct AI research beyond human understanding. Experts also warn that rushed or poorly designed controls could become politicized or captured by industry, producing results worse than doing nothing.
wired