Reproduce “Friend or Foe: Delegating to an AI Whose Alignment is Unknown” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Drew Fudenberg, Annie Liang
The paper studies how a decision maker can limit an AI’s influence by choosing which patient attributes to reveal. In a theoretical treatment model, an aligned AI reports truthfully, while a misaligned AI can preserve known population rates and induce harmful treatment. For the 40%-need example, expected payoff ranges from +0.4 to −0.6; these are model outcomes, not clinical evidence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.