AI models extrapolate human biases. In our work, we (with @hplisiecki.bsky.social ) show how easily they catch the bias of judges in sentiment analysis. Not a very surprising result (but large, very significant effect sizes), but it's an important step towards unbiased text analysis tools.