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
In search of a biological crux for AI consciousness – Bradford Saad (Global Priorities Institute, University of Oxford)
Whether AI systems could be conscious is often thought to turn on whether consciousness is closely linked to biology. The rough thought is that if consciousness is closely linked to biology, then AI consciousness is impossible, and if consciousness is not closely linked to biology, then AI consciousness is possible—or, at any rate, it’s more likely to be possible. A clearer specification of the kind of link between consciousness and biology that is crucial for the possibility of AI consciousness would help organize inquiry into…
In defence of fanaticism – Hayden Wilkinson (Australian National University)
Consider a decision between: 1) a certainty of a moderately good outcome, such as one additional life saved; 2) a lottery which probably gives a worse outcome, but has a tiny probability of a far better outcome (perhaps trillions of blissful lives created). Which is morally better? Expected value theory (with a plausible axiology) judges (2) as better, no matter how tiny its probability of success. But this seems fanatical. So we may be tempted to abandon expected value theory…
Population ethics with thresholds – Walter Bossert (University of Montreal), Susumu Cato (University of Tokyo) and Kohei Kamaga (Sophia University)
We propose a new class of social quasi-orderings in a variable-population setting. In order to declare one utility distribution at least as good as another, the critical-level utilitarian value of the former must reach or surpass the value of the latter. For each possible absolute value of the difference between the population sizes of two distributions to be compared, we specify a non-negative threshold level and a threshold inequality. This inequality indicates whether the corresponding threshold level must be reached or surpassed in…