Can AI understand humor ?

The ability of AI to understand humor is a topic of interest in the field of natural language processing and cognitive science. This would involve recognizing and interpreting the complex social and cultural cues that underlie human humor, as well as generating humorous responses. The understanding of humor would have significant implications for human-computer interaction, entertainment, and social bonding. It would require a deep understanding of the cognitive and social mechanisms that underlie human humor, as well as the ability to generate novel and contextually appropriate humorous responses. The potential applications of such a capability are vast, ranging from chatbots to comedy writing. However, it also raises important questions about the potential impact on human creativity and the role of AI in shaping human culture.

Current AI humor systems rely on pattern-matching: they can recognize witty wordplay or culturally common jokes by training on large corpora, but they do not truly “get” why something is funny. Benchmarks like the Joke Explainer or the New Yorker Caption Contest show models scoring above chance in labeling jokes or filling in punch-lines, yet their explanations remain shallow, often paraphrasing surface features rather than invoking shared social or emotional understanding. Systems still struggle with novel, absurdist, or culturally specific humor that depends on nuanced world knowledge and pragmatic inference. Consequently, AI humor is best described as assistive—helping writers generate options—rather than genuinely comprehending humor.

— Enriched May 11, 2026 · Source: best-effort summary, no public reference