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 does not 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
Altruism in governance: Insights from randomized training – Sultan Mehmood, (New Economic School), Shaheen Naseer (Lahore School of Economics) and Daniel L. Chen (Toulouse School of Economics)
Randomizing different schools of thought in training altruism finds that training junior deputy ministers in the utility of empathy renders at least a 0.4 standard deviation increase in altruism. Treated ministers increased their perspective-taking: blood donations doubled, but only when blood banks requested their exact blood type. Perspective-taking in strategic dilemmas improved. Field measures such as orphanage visits and volunteering in impoverished schools also increased, as did their test scores in teamwork assessments…
‘The only ethical argument for positive 𝛿’? – Andreas Mogensen (Global Priorities Institute, Oxford University)
I consider whether a positive rate of pure intergenerational time preference is justifiable in terms of agent-relative moral reasons relating to partiality between generations, an idea I call discounting for kinship. I respond to Parfit’s objections to discounting for kinship, but then highlight a number of apparent limitations of this…
How should risk and ambiguity affect our charitable giving? – Lara Buchak (Princeton University)
Suppose we want to do the most good we can with a particular sum of money, but we cannot be certain of the consequences of different ways of making use of it. This paper explores how our attitudes towards risk and ambiguity bear on what we should do. It shows that risk-avoidance and ambiguity-aversion can each provide good reason to divide our money between various charitable organizations rather than to give it all to the most promising one…