Reproduce “Incentives for Responsiveness, Instrumental Control and Impact” 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
Ryan Carey, Eric Langlois, Chris van Merwijk, Shane Legg, Tom Everitt
The authors present a causal-graph framework distinguishing response, instrumental control, and impact incentives in optimizing agents. Graphical criteria identify when a causal structure admits each incentive. In a recommender model, evaluating hypothetical clicks after inert content removes the modeled incentive to influence user opinions while retaining an incentive to provide desired content. Admission indicates compatibility, not actual occurrence or strength.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.