GenAI Creativity in Narrative Tasks
Author Information
Author(s): Florent Vinchon, Valentin Gironnay, Todd Lubart
Primary Institution: Laboratoire de Psychologie et d’Ergonomie Appliquée (LaPEA), Université Paris Cité and Univ Gustave Eiffel
Hypothesis
How does ChatGPT perform on standard creativity tasks, as assessed by the EPoC framework?
Conclusion
ChatGPT shows remarkable fluency in generating content but its creative output is often average and lacks true originality.
Supporting Evidence
- ChatGPT demonstrated higher fluency in generating ideas compared to human norms.
- GPT4 outperformed GPT3.5 in creativity tasks.
- AI-generated stories often showed patterns of plagiarism from well-known narratives.
- Significant differences were found in the performance of GPT3.5 and GPT4 on the AUT task.
Takeaway
This study looked at how well an AI can create stories, and while it can come up with a lot of ideas quickly, those ideas aren't always very original.
Methodology
The study used the EPoC verbal test to assess the creativity of ChatGPT across different narrative tasks.
Potential Biases
The AI's outputs may reflect biases present in its training data, leading to potential plagiarism and lack of originality.
Limitations
The study lacked a human comparison group, which would have provided a better context for the AI's performance.
Participant Demographics
The study involved 100 'new chat' individuals, split evenly between GPT3.5 and GPT4.
Statistical Information
P-Value
p<0.001
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website