"Not Human, Funnier": How Machine Identity Shapes Humor Perception in Online AI Stand-up Comedy

Paper Title

'Not Human, Funnier': How Machine Identity Shapes Humor Perception in Online AI Stand-up Comedy

Publication Info

• Topic area: AI-generated humor and its perception in human-AI interactions.
• Keywords: AI humor, machine identity, stand-up comedy, generative AI, audience perception, humor strategies, computational authenticity, identity-driven design, interactive systems, HCI.

Background and Problem

• Problem / challenge: Existing AI humor systems often mimic human comedic styles but lack originality and fail to leverage AI’s unique computational identity as a comedic resource.
• Significance: Understanding how machine identity can enhance AI humor could lead to more engaging and authentic human-AI interactions, with applications in entertainment, education, and customer service.
• Motivation and related work: Prior research has focused on scripted or human-mimicking AI humor, neglecting the potential of AI’s computational traits as a source of humor. This study builds on gaps in identity-driven humor creation and explores how machine identity can shape audience perceptions.

Solution

• Proposed approach: A machine-identity-based AI comedian that uses its computational traits as a comedic resource in an online stand-up comedy format.
• Novelty:
  1. Introduces machine identity as a core comedic framework, emphasizing computational traits over human mimicry.
  2. Develops a structured prompting framework for humor generation, incorporating timing, rhetorical techniques, and ethical safeguards.
  3. Implements a real-time multimodal interaction system to simulate live performance dynamics and audience feedback.
  4. Evaluates the impact of machine identity on humor perception, personality traits, and audience engagement.
• Procedure and key techniques:
  • Conducted formative studies, including expert interviews, video coding of stand-up performances, and a focused literature review.
  • Designed a comedy-specific prompting framework with identity-driven humor strategies (e.g., irony, absurdity, self-deprecation).
  • Built a live performance interface with real-time audience interaction and adaptive humor logic.
  • Conducted a within-subject study (N=32) comparing the machine-identity-based system to a baseline AI comedian.

Results

• Concrete findings:
  • Machine-identity-based humor was rated significantly higher in perceived humor (W = 54.0, corrected p = .030).
  • Improved perceptions of personality traits such as Agreeableness (W = 34.0, corrected p = .047) and Warmth (W = 42.0, corrected p = .030).
  • Enhanced user perception metrics like Anthropomorphism (W = 48.0, corrected p = .030) and Animacy (W = 44.5, corrected p = .030).
• Advantage over baselines:
  • Machine-identity humor was seen as more original, engaging, and relatable compared to generic, human-mimicking jokes.
  • Audience interaction was more frequent and sustained due to better timing and thematic coherence.
• Experiments / evaluation:
  • Participants experienced two AI comedy performances (baseline vs. machine-identity) in a counterbalanced within-subject design.
  • Quantitative measures included humor ratings, personality perception, and user engagement.
  • Qualitative feedback highlighted the novelty, relatability, and ethical considerations of machine-identity humor.
• Limitations and future work:
  • Limited cultural diversity in the participant sample, predominantly Asian.
  • Online setting lacked the social presence of in-person comedy.
  • Simplified audience interaction (binary feedback) and short performance durations.
  • Future work should explore multi-party platforms, hybrid reality environments, and longer performances.

Summary

This study demonstrates that leveraging machine identity as a comedic resource can enhance the originality, engagement, and perceived personality of AI-generated humor. By shifting from human mimicry to computational authenticity, the proposed system achieved higher humor ratings and audience interaction compared to a baseline AI comedian. These findings have implications for designing AI systems in entertainment, education, and customer service, emphasizing the value of identity-driven approaches. Future research should address cultural diversity, richer interaction modalities, and extended performance contexts to further refine AI humor design.