Towards shutdownable agents via stochastic choice

Elliott Thornley (Global Priorities Institute, University of Oxford), Alexander Roman (New College of Florida), Christos Ziakas (Independent), Leyton Ho (Brown University) and Louis Thomson (University of Oxford)

GPI Working Paper No. 16-2024

PDF download

Some worry that advanced artificial agents may resist being shut down. The Incomplete Preferences Proposal (IPP) is an idea for ensuring that doesn’t happen. A key part of the IPP is using a novel ‘Discounted REward for Same-Length Trajectories (DREST)’ reward function to train agents to (1) pursue goals effectively conditional on each trajectory-length (be ‘USEFUL’), and (2) choose stochastically between different trajectory-lengths (be ‘NEUTRAL’ about trajectory-lengths). In this paper, we propose evaluation metrics for USEFULNESS and NEUTRALITY. We use a DREST reward function to train simple agents to navigate gridworlds, and we find that these agents learn to be USEFUL and NEUTRAL. Our results thus suggest that DREST reward functions could also train advanced agents to be USEFUL and NEUTRAL, and thereby make these advanced agents useful and shutdownable.

Other working papers

Existential Risk and Growth – Leopold Aschenbrenner and Philip Trammell (Global Priorities Institute and Department of Economics, University of Oxford)

Technology increases consumption but can create or mitigate existential risk to human civilization. Though accelerating technological development may increase the hazard rate (the risk of existential catastrophe per period) in the short run, two considerations suggest that acceleration typically decreases the risk that such a catastrophe ever occurs. First, acceleration decreases the time spent at each technology level. Second, given a policy option to sacrifice consumption for safety, acceleration motivates greater sacrifices…