Calculating chuckles: When algorithms go for laughs

Abstract: I examine the intersection of artificial intelligence (AI), humor comprehension, and mathematics education through the lens of a recurring classroom assignment. Over several years, my students in a beginning algebra course were asked to find and explain math jokes, yielding a wide variety of responses. However, after I integrated AI into the assignment, the diversity of jokes sharply declined, with AI frequently repeating the same well-known joke. Drawing on research and classroom experiments, I explore the limitations of current AI models in processing and generating humor, particularly puns, in contrast to human performance. Case studies—including AI's struggles with puns and cartoon caption contests—underscore AI's reliance on statistical patterns rather than genuine understanding. I conclude that while AI can mimic humor's surface structure, its deficiencies are instructive: recognizing these limitations may help prevent overconfident outputs and foster the development of more reliable, emotionally intelligent AI tools.

Economics

Audience: researchers in the topic

DVU Data Science Seminar Series

Series comments: First Presentation: National Center for Civil and Human Rights