"Turning Right"? An experimental study on the political value shift in large language models

Abstract Constructing artificial intelligence that aligns with human values is a crucial challenge, with political values playing a distinctive role among the various human value systems. In this study, we developed a standardized method to test political values in AI, utilizing the Political Compass Test alongside rigorous bootstrapping techniques. This methodology was applied to different versions of the notably influential ChatGPT. Our findings reveal that while newer versions of ChatGPT consistently maintain values within the libertarian-left quadrant, there is a significant rightward shift in political values, a phenomenon we term a 'value shift' in large language models. This shift is particularly noteworthy given the deep integration of large language models with human knowledge creation, suggesting potentially profound impacts on societal values. Intriguingly, our analysis indicates that this value shift in ChatGPT may not be directly linked to its training datasets. This research sheds light on the dynamic nature of value alignment in AI, highlighting the need for continuous monitoring and understanding of the underlying factors driving these shifts in AI-generated political values.