A starting point for discussion
The idea
Two independent bodies that check the AI executive should each be able to pause it if it causes harm. Each should have its own administrator and separate credentials. This is a tentative proposal, not a proven solution.
Place in the reasoning
AI assessment · Connected to reasoning
Proposes two pause-capable bodies with separate administrators and credentials, while leaving shared powers and dependencies unresolved.
Separate checking bodies may still share a weakness
This is the tentative separation design whose sufficiency is qualified by possible shared dependencies.
Rules for stopping AI need clear powers and limits
Proposes pause authority but does not specify triggers, evidence standards, or adjudication.
What this means and why it matters
What this means
Each of two bodies that check the AI executive would have its own administrator and separate credentials, and each could pause the executive.
Why it matters
This avoids the specific choice to have one administrator serve both bodies. It does not show that other shared systems or decision-making steps are safe.
Conditions & assumptions
Which other functions the bodies that check the AI executive can safely share, and whether separate administrators and credentials remove all important shared dependencies, remain unresolved.
Questions this connects to
AI-suggested relationships. Authors have not necessarily responded to one another.