Why the Funniest AI Joke Will Not Be Funny to Us

Discover a groundbreaking computational theory of humor in this insightful podcast, formally equating jokes with cognitive bugs. Explore how humor emerges from the sudden detection and resolution of epistemic errors within intelligent agents, with laughter serving as a public signal of successful model correction.This discussion delves into:• The revolutionary idea that humor is a form of self-debugging, serving as a proxy indicator for general intelligence.• Crucial implications for AI safety, arguing that AI systems unable to recognize joke-like incongruities might overlook critical misalignments with human values.• A unique taxonomy of joke types mapped to software bug categories, including semantic, logical, and syntax errors, as well as runtime, off-by-one, memory leak, and security bugs.• Why the funniest AI joke will not be funny to us, touching on the concept of "superjokes" that are too complex for humans to understand.• The role of comedians as society's debuggers, adept at identifying and highlighting contradictions and absurdities.• Insights into how humor is AI-complete, meaning true comprehension and generation require human-equivalent cognitive abilities.

Roman V. Yampolskiy proposes a novel computational theory of humor, equating jokes to cognitive bugs or unexpected twists within an intelligent agent's predictive models, and views laughter as a public signal of successful model correction. It argues that the ability of artificial intelligence (AI) to generate and comprehend humor can serve as a proxy for general intelligence and is crucial for AI safety, as systems failing to recognize humorous incongruities may also miss critical misalignments with human values. The paper categorizes joke types by software bug categories, discusses the subjective nature of funniness, and suggests that humor may be an AI-complete problem, implying it requires comprehensive human-level cognitive abilities for true mastery. Ultimately, the text explores how humor can function as a self-debugging mechanism for both human and AI minds, fostering learning, social bonding, and critical thinking.