Why AI Safety Requires Uncertainty, Incomplete Preferences, and Non-Archimedean Utilities

TL;DR

This paper argues that ensuring AI safety and alignment with human values necessitates AI systems capable of reasoning under uncertainty, managing incomplete preferences, and utilizing non-Archimedean utilities to handle complex decision-making scenarios.

Contribution

It introduces the importance of non-Archimedean utilities and incomplete preferences for designing safe and aligned AI agents, linking these concepts to AI assistance and shutdown problems.

Findings

AI safety requires reasoning under uncertainty.

Handling incomplete preferences is crucial for alignment.

Non-Archimedean utilities enable better decision-making in AI safety.

Abstract

How can we ensure that AI systems are aligned with human values and remain safe? We can study this problem through the frameworks of the AI assistance and the AI shutdown games. The AI assistance problem concerns designing an AI agent that helps a human to maximise their utility function(s). However, only the human knows these function(s); the AI assistant must learn them. The shutdown problem instead concerns designing AI agents that: shut down when a shutdown button is pressed; neither try to prevent nor cause the pressing of the shutdown button; and otherwise accomplish their task competently. In this paper, we show that addressing these challenges requires AI agents that can reason under uncertainty and handle both incomplete and non-Archimedean preferences.