Towards shutdownable agents via stochastic choice
Elliott Thornley (Global Priorities Institute, University of Oxford), Alexander Roman (New College of Florida), Christos Ziakas (Imperial College, London), Leyton Ho (Brown University) and Louis Thomson (University of Oxford)
GPI Working Paper No. 16-2024
The POST-Agents Proposal (PAP) is an idea for ensuring that advanced artificial agents never resist shutdown. A key part of the PAP 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 provide some initial evidence that DReST reward functions could train advanced agents to be USEFUL and NEUTRAL. Our theoretical work suggests that these agents would be useful and shutdownable.
Other working papers
A Fission Problem for Person-Affecting Views – Elliott Thornley (Global Priorities Institute, University of Oxford)
On person-affecting views in population ethics, the moral import of a person’s welfare depends on that person’s temporal or modal status. These views typically imply that – all else equal – we’re never required to create extra people, or to act in ways that increase the probability of extra people coming into existence. In this paper, I use Parfit-style fission cases to construct a dilemma for person-affecting views: either they forfeit their seeming-advantages and face fission analogues…
Can an evidentialist be risk-averse? – Hayden Wilkinson (Global Priorities Institute, University of Oxford)
Two key questions of normative decision theory are: 1) whether the probabilities relevant to decision theory are evidential or causal; and 2) whether agents should be risk-neutral, and so maximise the expected value of the outcome, or instead risk-averse (or otherwise sensitive to risk). These questions are typically thought to be independent – that our answer to one bears little on our answer to the other. …
In Defence of Moderation – Jacob Barrett (Vanderbilt University)
A decision theory is fanatical if it says that, for any sure thing of getting some finite amount of value, it would always be better to almost certainly get nothing while having some tiny probability (no matter how small) of getting sufficiently more finite value. Fanaticism is extremely counterintuitive; common sense requires a more moderate view. However, a recent slew of arguments purport to vindicate it, claiming that moderate alternatives to fanaticism are sometimes similarly counterintuitive, face a powerful continuum argument…