Hayastan Avetisyan, Parisa Safikhani, and David Broneske

In this study, we conduct a thorough comparative analysis between artificial intelligence (AI)- generated humor and human humor. The objective is to acquire a more profound understanding of AI’s present capabilities in generating humorous text. We investigate the structural, sentiment, and linguistic patterns in jokes created by AI and humans, evaluating ’funniness’ and ’originality’ via a comprehensive annotation process. Our findings indicate that AI can produce humorous and occasionally novel content. Additionally, we employed the RoBERTa model for humor detection on a dataset consisting of 500 entries, including both human and AI-generated humor. This model demonstrated its proficiency in accurately categorizing a large dataset encompassing up to 200,000 entries with remarkable accuracy of up to 98%. Nonetheless, it lacks the emotional depth and originality commonly seen in human humor. The study underscores the challenge involved in developing AI models that can generate humor equivalent to human communication. Future research should focus on enhancing AI’s ability to create humor and further examine AI’s potential to adopt human humor strategies. Despite some limitations, this study contributes significantly to improving the humorous capabilities of AI models and the expandability of AI-generated humor.

Artificial Intelligence (AI)-generated humor, Human humor, Linguistic patterns, Funniness evaluation, Originality evaluation, RoBERTa model