MIRI Criticizes OpenAI Safety Post for Confusing Messaging and Unproven Assumptions
MIRI researcher Harlan Stewart challenges OpenAI's recent safety and alignment webpage, arguing that it relies on confusing timelines and unproven assumptions. Stewart urges the AI developer to take greater responsibility for proving its superintelligence safety strategy.
The Machine Intelligence Research Institute (MIRI) publishes a critique of OpenAI's recent webpage on AI safety and alignment. Author Harlan Stewart appreciates that OpenAI participates in the public debate about existential AI risks but expresses deep disagreements with both the company's strategy and its communication style.
Stewart points out that OpenAI's messaging about AGI development timelines is confusing and contradictory. While OpenAI claims progress will be a "continuous" process rather than a sudden leap, the company also predicts AI will transform the world to be more different from today than today is from the 1500s within just a few years.
The critique also highlights that OpenAI's discussion overlooks important alignment considerations the company previously acknowledged. Furthermore, Stewart argues that OpenAI's plan to prevent superintelligence from causing harm relies on unproven assumptions, and he insists that OpenAI takes direct responsibility for proving those assumptions before it is too late.