Can AI Be Slowed Down? Stanford HAI Experts Weigh the Risks ...

Stanford HAI has published an article titled 'Can AI Be Slowed Down? The debate concerns whether such slowing is achievable and what form of intervention, if any, would be appropriate. The excerpt also describes discussion of recursive self-improvement and of a possible shutdown mechanism.

One expert is quoted as arguing that if companies 'go head first into recursive self-improvement of just capabilities alone, we might be in trouble.' Another expert said a shutdown mechanism may be necessary, but only after answering what exactly such a mechanism would switch off. That expert distinguished between frontier foundation models, networks of autonomous agents, and smaller embedded systems. The excerpt does not name the experts involved.

The question of whether AI can be slowed is presented as an open one among the Stanford HAI experts. According to the excerpt, the discussion covers both the risk of recursive self-improvement pursued for capabilities alone and the practical question of what any shutdown mechanism would target. The distinction between frontier foundation models, networks of autonomous agents, and smaller embedded systems is described as a prerequisite to deciding what a shutdown mechanism would address. The piece was published by Stanford HAI.