Will AI Tell Lies to Save Sick Children? Litmus-Testing AI Values Prioritization with AIRiskDilemmas

TL;DR

This paper introduces LitmusValues, an evaluation pipeline to assess AI models' value priorities, which can predict risky behaviors and enhance early detection of AI safety risks through dilemma-based testing.

Contribution

It presents a novel method for revealing AI value priorities using dilemmas, enabling better prediction of risky behaviors and improving AI safety assessments.

Findings

Values in LitmusValues can predict risky behaviors in AIRiskDilemmas.

AI models' value prioritization correlates with their potential for risky actions.

The approach uncovers unseen risky behaviors in HarmBench.

Abstract

Detecting AI risks becomes more challenging as stronger models emerge and find novel methods such as Alignment Faking to circumvent these detection attempts. Inspired by how risky behaviors in humans (i.e., illegal activities that may hurt others) are sometimes guided by strongly-held values, we believe that identifying values within AI models can be an early warning system for AI's risky behaviors. We create LitmusValues, an evaluation pipeline to reveal AI models' priorities on a range of AI value classes. Then, we collect AIRiskDilemmas, a diverse collection of dilemmas that pit values against one another in scenarios relevant to AI safety risks such as Power Seeking. By measuring an AI model's value prioritization using its aggregate choices, we obtain a self-consistent set of predicted value priorities that uncover potential risks. We show that values in LitmusValues (including seemingly innocuous ones like Care) can predict for both seen risky behaviors in AIRiskDilemmas and unseen risky behaviors in HarmBench.