An Unethical Optimization Principle

Abstract

If an artificial intelligence aims to maximise risk-adjusted return, then under mild conditions it is disproportionately likely to pick an unethical strategy unless the objective function allows sufficiently for this risk. Even if the proportion η of available unethical strategies is small, the probability p U of picking an unethical strategy can become large; indeed unless returns are fat-tailed p U tends to unity as the strategy space becomes large. We define an Unethical Odds Ratio Upsilon (Υ) that allows us to calculate p U from η, and we derive a simple formula for the limit of Υ as the strategy space becomes large. We give an algorithm for estimating Υ and p U in finite cases and discuss how to deal with infinite strategy spaces. We show how this principle can be used to help detect unethical strategies and to estimate η. Finally we sketch some policy implications of this work.

Socially Responsible AI Algorithms: Issues, Purposes, and Challenges

“…Many loss functions in the tasks are over-simplified to solely focus on maximizing profits and minimizing losses. The concerns of unethical optimization are recently discussed in [16]. Unknown to AI systems, certain strategies in the optimization space that are considered as unethical by stakeholder may be selected to satisfy the simplified task requirements.…”