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

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

Economic growth under transformative AI – Philip Trammell (Global Priorities Institute, Oxford University) and Anton Korinek (University of Virginia)

Industrialized countries have long seen relatively stable growth in output per capita and a stable labor share. AI may be transformative, in the sense that it may break one or both of these stylized facts. This review outlines the ways this may happen by placing several strands of the literature on AI and growth within a common framework. We first evaluate models in which AI increases output production, for example via increases in capital’s substitutability for labor…

Against the singularity hypothesis – David Thorstad (Global Priorities Institute, University of Oxford)

The singularity hypothesis is a radical hypothesis about the future of artificial intelligence on which self-improving artificial agents will quickly become orders of magnitude more intelligent than the average human. Despite the ambitiousness of its claims, the singularity hypothesis has been defended at length by leading philosophers and artificial intelligence researchers. In this paper, I argue that the singularity hypothesis rests on scientifically implausible growth assumptions. …

Aggregating Small Risks of Serious Harms – Tomi Francis (Global Priorities Institute, University of Oxford)

According to Partial Aggregation, a serious harm can be outweighed by a large number of somewhat less serious harms, but can outweigh any number of trivial harms. In this paper, I address the question of how we should extend Partial Aggregation to cases of risk, and especially to cases involving small risks of serious harms. I argue that, contrary to the most popular versions of the ex ante and ex post views, we should sometimes prevent a small risk that a large number of people will suffer serious harms rather than prevent…