AI Humor Generation: Cognitive, Social and Creative Skills for Effective Humor (2502.07981v1)

Abstract: Humor is a social binding agent. It is an act of creativity that can provoke emotional reactions on a broad range of topics. Humor has long been thought to be "too human" for AI to generate. However, humans are complex, and humor requires our complex set of skills: cognitive reasoning, social understanding, a broad base of knowledge, creative thinking, and audience understanding. We explore whether giving AI such skills enables it to write humor. We target one audience: Gen Z humor fans. We ask people to rate meme caption humor from three sources: highly upvoted human captions, 2) basic LLMs, and 3) LLMs captions with humor skills. We find that users like LLMs captions with humor skills more than basic LLMs and almost on par with top-rated humor written by people. We discuss how giving AI human-like skills can help it generate communication that resonates with people.

• The paper demonstrates that embedding humor-specific skills in AI enables the generation of meme captions that closely rival human creativity.
• The methodology employs a multi-stage process integrating visual detail extraction, narrative ideation, and fine-tuned caption generation tailored for Gen Z humor.
• Empirical evaluations reveal that AI-generated captions are competitively rated against human entries, confirming the potential of computational humor.

An Analysis of AI Humor Generation for Gen Z Audiences

The paper explores the potential of equipping AI systems with human-like cognitive, social, and creative skills to generate humor effectively, specifically targeting Gen Z audiences through Instagram meme captions. Authored by Sean Kim and Lydia B. Chilton from Columbia University, the research undertakes a systematic comparison of AI-generated humor to human-authored captions to assess the feasibility of computational humor.

Core Hypothesis and Methodology

The paper posits that LLMs, when imbued with specific humor-oriented skills, can produce captions that are competitive with human-generated content. Given the complexity of humor, which often requires audience awareness, cultural context, and nuanced cognitive abilities, the research focuses on captions that resonate with a Gen Z demographic—a group marked by their affinity for digital content and social media engagement.

The methodological framework is a multi-stage process:

1. Visual Detail Extraction: This phase utilizes a visual LLM (VLM) to analyze the image, capturing elements that could serve as humorous triggers.
2. Humor Ideation via Narratives: AI is prompted to generate potential humorous angles by introducing narratives or conflicts related to Gen Z experiences, thus broadening the humorous context beyond the immediate visual content.
3. Caption Generation: The process involves generating captions through a fine-tuned LLM trained on Gen Z humor, which explores both direct image-based humor and metaphorical narratives.
4. Humor Evaluation: Captions are evaluated by a Gen Z humor expert agent—an LLM trained to assess the relevance and appeal of humor for this demographic, ensuring the selection of optimal captions.

Empirical Evaluation

In an empirical setting, the paper administers a humor rating survey encompassing three categories of captions: human-authored, those generated by a basic LLM, and those generated using the HumorSkills process. The survey concluded with the AI-produced captions being rated nearly as highly as the human-written entries, suggesting a significant narrowing of the performance gap between human and machine humor. While HumorSkills captions marginally fell short of top human captions, they were notably superior to simple AI-generated humor.

Implications for AI Development

The results indicate that incorporating human-like skills into AI systems can considerably enhance their communicative capabilities, addressing the traditionally human domain of humor generation. This improvement not only enlarges the scope of AI's creative applications but also raises questions about the role of AI in social contexts traditionally mediated by humans.

Future Directions

The research points towards several exciting trajectories for future exploration. There is potential for refining the Gen Z humor model with more extensive datasets and diverse humor styles, possibly extending beyond captioning into other forms of humor delivery such as stand-up comedy or interactive storytelling. Additionally, the application of multi-step reasoning and refined narrative techniques could further improve the sophistication of AI-generated humor.

Conclusion

The paper presents a compelling argument for the viability of AI-generated humor, specifically through the structured application of human-like cognitive and social skills. By effectively blending computational capabilities with nuanced understanding of audience demographics, it opens a path for AI to emulate complex human communication skills, such as humor, and challenges the traditionally human-centric domain of creative expression. The findings contribute to the broader discourse on AI's evolving role in creative industries and interpersonal communication.