AI voter models treat groups differently and can be unfair
AI Alignment for Digital Democracy: Fairer Voters Representation against LLM Biases
Computers and SocietyArtificial Intelligence
Summary
Large language AI models are used to mimic how voters might choose in elections, but they don't represent all voter groups equally well, which can be unfair. The authors studied how these AI models perform on different demographic groups using past U.S. elections and found that removing some personal details from voter profiles can reduce errors for some groups without hurting accuracy. However, common methods to fix fairness problems often don't work and can even make things worse. This points to the challenge of making AI voter models both accurate and fair to everyone.
What this means in practice
- •For political data analysts: Improve voter simulation methods by adjusting voter information inputs to reduce bias without losing accuracy in election modeling.
- •For online platform designers: Design fairer AI-driven decision support tools in digital democratic systems by avoiding ineffective bias-reduction prompts and strategies.
Authors
Yi Qi, Srijoni Majumdar, Evangelos Pournaras
Abstract
AI representation is emerging in digital democracy, assisting in voter emulation and decision-making. Large language models (LLMs) behave differently depending on the model, the prompt context, and the voter information they receive, representing some groups more accurately than others and potentially making the emulation unfair to certain groups. We investigate demographic biases in LLM-emulated electorates: how accurately LLMs represent individual votes, whether some voter groups are represented more accurately than others, and whether AI alignment can reduce these disparities without sacrificing overall accuracy. Using three U.S. presidential elections, we compare a closed-source and an open-source LLM across voters' demographic features and test AI alignment strategies toward fairer voter representation. Removing gender from the voter profile reduces the largest group error of the models by 22-42%, and removing education reduces it by 64%, whereas removing self-reported race increases the mean group error by over 40%, all without lowering accuracy. Common alignment strategies, such as repeated queries or fairness instructions, do not reduce these errors and can double them.