Model-Based Soft Maximization of Suitable Metrics of Long-Term Human Power

TL;DR

This paper proposes a new AI objective function that promotes human empowerment and manages power balance, aiming to enhance safety and wellbeing in AI systems through a principled, long-term, and socially aware approach.

Contribution

It introduces a parametrizable, decomposable metric for human power that accounts for social norms and human goals, with algorithms for its computation and analysis of its implications.

Findings

The proposed metric can be computed via backward induction or multi-agent reinforcement learning.

Maximizing this metric can lead to safer AI behaviors that favor human empowerment.

The approach balances safety and wellbeing by explicitly considering human power dynamics.

Abstract

Power is a key concept in AI safety: power-seeking as an instrumental goal, sudden or gradual disempowerment of humans, power balance in human-AI interaction and international AI governance. At the same time, power as the ability to pursue diverse goals is essential for wellbeing. This paper explores the idea of promoting both safety and wellbeing by forcing AI agents explicitly to empower humans and to manage the power balance between humans and AI agents in a desirable way. Using a principled, partially axiomatic approach, we design a parametrizable and decomposable objective function that represents an inequality- and risk-averse long-term aggregate of human power. It takes into account humans' bounded rationality and social norms, and, crucially, considers a wide variety of possible human goals. We derive algorithms for computing that metric by backward induction or approximating it via a form of multi-agent reinforcement learning from a given world model. We exemplify the consequences of (softly) maximizing this metric in a variety of paradigmatic situations and describe what instrumental sub-goals it will likely imply. Our cautious assessment is that softly maximizing suitable aggregate metrics of human power might constitute a beneficial objective for agentic AI systems that is safer than direct utility-based objectives.