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

Consciousness makes things matter – Andrew Y. Lee (University of Toronto)

This paper argues that phenomenal consciousness is what makes an entity a welfare subject, or the kind of thing that can be better or worse off. I develop and motivate this view, and then defend it from objections concerning death, non-conscious entities that have interests (such as plants), and conscious subjects that necessarily have welfare level zero. I also explain how my theory of welfare subjects relates to experientialist and anti-experientialist theories of welfare goods.

Economic inequality and the long-term future – Andreas T. Schmidt (University of Groningen) and Daan Juijn (CE Delft)

Why, if at all, should we object to economic inequality? Some central arguments – the argument from decreasing marginal utility for example – invoke instrumental reasons and object to inequality because of its effects…

AI alignment vs AI ethical treatment: Ten challenges – Adam Bradley (Lingnan University) and Bradford Saad (Global Priorities Institute, University of Oxford)

A morally acceptable course of AI development should avoid two dangers: creating unaligned AI systems that pose a threat to humanity and mistreating AI systems that merit moral consideration in their own right. This paper argues these two dangers interact and that if we create AI systems that merit moral consideration, simultaneously avoiding both of these dangers would be extremely challenging. While our argument is straightforward and supported by a wide range of pretheoretical moral judgments, it has far-reaching…