Thoughts on AI pause

Abstract

This post discusses the arguments for and against advocating for a pause in artificial intelligence development. Proponents of a pause emphasize the need for AI safety and governance to catch up with the rapid advancements in AI capabilities. However, the tractability and potential benefits of a pause are uncertain, especially compared to alternative efforts such as focusing on worst-case AI safety or mitigating s-risks (risks of astronomical suffering). A common flaw in the AI pause discourse is its simplistic “doom vs. non-doom” framing, which overlooks the potential for severe suffering in human-controlled futures. From a suffering-focused perspective, it is unclear whether a human-controlled future with advanced AI is preferable to a future where control is lost. While ensuring the avoidance of catastrophic AI-controlled outcomes is a moral imperative, equal consideration should be given to the potential for catastrophic human-controlled outcomes. Therefore, it is not evident that advocating for an AI pause is the most effective approach to reducing future suffering. – AI-generated abstract.