While testing the robustness of humans and AI models on story-based analogy problems, the study found the models were susceptible to answer-order effects — differences in responses due to the ...