I Tried to Teach AI Humor, Here’s What Happened l Bob Mankoff

Key Concepts

AI, humor, creativity, averageness, human agency, algorithms, caption contest, large language models, photorealism, existential questions, coping, artificial general intelligence (AGI), vulnerability, understanding vs. answers.

AI and Creativity: A Conflicted Perspective

Bob Mankoff expresses a conflicted view on AI, acknowledging its usefulness in generating ideas but also highlighting its limitations. He compares his feelings about AI to his feelings about phones, recognizing their utility while questioning their overall impact on well-being.

• AI as a Tool: Mankoff sees AI as a tool that can be helpful but should not be granted ultimate judgment, especially in subjective areas like humor, music, or literature.
• The Problem of Averageness: He argues that AI is trained on "averageness," which can be a drawback because individual sensibilities are unique and valuable. "Remember it is being trained on averageness but I'm not average. Each one of us has our own sensibility."
• Loss of Agency: Mankoff warns that ceding too much control to algorithms could lead to a loss of agency in our lives. "If all of a sudden we cede to algorithms completely what it is that we should like, then I do think We lose a certain kind of agency in our lives."

Mankoff's Background and Career

Mankoff provides a brief overview of his career path, highlighting his journey from experimental psychology to becoming a cartoonist and eventually the cartoon editor of The New Yorker.

• Early Interests: He mentions his early interest in drawing and humor, which eventually led him to pursue cartooning.
• Persistence: He emphasizes the importance of persistence, noting that he submitted around 2000 cartoons to The New Yorker before finally getting accepted.
• The Cartoon Bank and Cartoonstock.com: Mankoff describes how he created the Cartoon Bank to monetize rejected New Yorker cartoons, which later evolved into Cartoonstock.com, a business where half of the revenue goes to the cartoonist.

The Caption Contest and AI Research

Mankoff discusses his involvement in AI research related to humor, specifically focusing on The New Yorker's caption contest.

• Crowdsourcing Captions: He partnered with researchers at the University of Wisconsin to crowdsource captions for the contest images, creating a large dataset of ranked captions.
• AI's Initial Failure: The initial attempt to use AI to distinguish funny from unfunny captions failed because the AI, which did not use large language models, could not understand the images. "It could not look at the image and understand what was in it."
• Stimulation of Further Research: Despite the initial failure, the project stimulated further research in the field.

The Limits of AI in Understanding Humor and Human Experience

Mankoff argues that AI is unlikely to achieve true understanding of humor and the human experience, particularly when it comes to existential questions and coping with sorrow.

• Photorealism Analogy: He draws an analogy to photorealism, suggesting that while AI can create impressive imitations, it lacks the depth and meaning of human art.
• Existential Questions: Mankoff points out that AI is useless for addressing existential questions, such as dealing with mortality and loss. He contrasts this with the value of human connection and shared experiences in coping with such challenges.
• Understanding vs. Answers: He emphasizes that humans seek understanding, not just answers, and that AI is ill-equipped to provide this. "We don't actually want answers for anything. We want understanding. And it's only human beings who can understand."
• Humor as Coping: Mankoff cites Mark Twain's observation that humor is rooted in sorrow and hurt, and argues that one of its main functions is coping. He suggests that AI, lacking the capacity for sorrow, cannot truly master humor. "One of the functions of humor is for it to deal with our ultimate sorrow. So one of the main functions of humor is coping. And frankly, AI has nothing to cope with."

Conclusion

Mankoff concludes by expressing his belief that AI can be a useful tool for augmenting human humor, but it should not replace human judgment or creativity. He suggests that AI may achieve mediocre humor in the near future, but deep, meaningful human humor remains beyond its reach. He hopes that AI will not surpass human humor, or at least not in his lifetime.