A Dynamic and Dual-Process Theory of Humor

"Don't you hear me? The jokes we laugh at were not made up by any man. Multivac has analyzed all data given it and the one answer that best fits that data is that some extraterrestrial intelligence has composed the jokes, all of them, and placed them in selected human minds at selected times and places in such a way that no man is conscious of having made one up."

-- Issac Asimov, Jokester, 1956

For centuries, the curious cognitive phenomenon known as humor has attracted the attention of philosophers, linguists, cognitive scientists, AI researchers and the like. The earliest theories of humor may have appeared in Plato's Philebus and Aristotle's Poetics. Both philosophers were proponents of the superiority theory, which posits that we laugh at the misfortune of other people. Since then, an abundance of theories have been proposed, ranging from the release of psychic or nervous energy to the formation of an incongruity that is later resolved . Each theory seems to possess some explanatory power, yet none can satisfactorily encompass all theories and provide a unified account. Reviewing these theories, one can easily be reminded of the ancient fable of blind men and the elephant.

In this project, I attempt to provide a single, unified framework for humor, grounded in recent developments on emotion and dual-process cognition. I propose that humor comprehension consists of a four-step dynamic process: surprise, reflection, dismissal and compensation. The proposed theory provides a modern update on existing theories of humor, and is capable of explaining several phenomena that cannot be easily explained by existing theories. Converging evidences from brain imaging results, facial recognition studies, the effect of repetition on humor, and the recently reported frustration smiles support its validity. The current theory highlights the importance of studying complex affects, such as humor and suspense, in the context of interactions between cognitive processes and subsystems. Therefore, it motivates the development of large, comprehensive AI systems and architectures with multiple interacting modules, especially systems that contain self-monitoring and error detection as a major building block.