Vitalik Buterin says AI 2040 debate hinges on superintelligence timing and governance

Vitalik Buterin said the divide in the AI 2040 debate stems from conflicting assumptions about how quickly advanced AI could progress and whether superintelligence would become a concentrated source of power. In a thread on X, the Ethereum co-founder said he is unsure which worldview is closer to reality, but remains open to slowing or pausing AI development if risks become significant. He framed one side as expecting some form of superintelligence by 2040 unless strong measures intervene, while critics fear restrictions on freedom and human decision-making more than superintelligence itself. The discussion drew in Meta chief AI scientist Yann LeCun, author Daniel Jeffries, policy analyst Adam Thierer, AI researcher Richard Ngo and others. LeCun argued AI safety is an engineering problem that can be improved iteratively and said current large language models are limited systems that do not yet reason like humans. Buterin countered that this view rules out the possibility of an AI powerful enough to perform any task on its own, which he labels ASI. Buterin outlined a "plan A" centered on mandatory openness about AI development and an emergency off-switch that could slow or stop large training runs if systems become dangerous. He acknowledged there is no perfect approach to an ASI transition. The wider argument also touches on whether open source models reduce power concentration or whether tighter approval systems would choke open development. Daniel Kokotajlo’s AI Futures Project report proposed U.S.-China cooperation to delay superintelligence until 2040, including shared research and a deterrence model it calls mutually assured compute destruction.