[Colloquium Presentation] The Bizarre World of Happiness Drawn by Image Generative AI

On January 17, 2025, Kyoungmi (Kimmy) Oh attended a colloquium titled “Is Generative AI a Game Changer for the Media Ecosystem?”. The colloquium is hosted by the Graduate School Of Public Policy and Information Technology of the Seoul National University of Science and Technology.

Moderator

• Park So-hyun (Professor, Digital Culture Policy, IT Policy Graduate School)

Session 1

Presentations

1. Study on TV Production Teams’ Perception and Utilization of Generative AI
   • Jo Seo-yoon (Ph.D. Candidate, Convergent Media Content Policy Major)
2. Study on Expanding User Experience in AI Advertising
   • Cha Hye-young (Ph.D. Candidate, IT Design Convergence Major / Visiting Professor, Industrial Design Department, Gwangju University)

Panel Discussion

• Chair: Park So-hyun (Professor, Digital Culture Policy, IT Policy Graduate School)
• Panelists:
  • Kim Dae-kyu (HCN Policy Researcher)
  • Lee Sun-woo (JTBC Producer)
  • Ahn Moo-jung (Senior Manager, LG CNS)

Session 2

Presentations

1. Normative Approaches to Transparency in AI-Generated Content
   • Park Myung-soon (Completed Ph.D. coursework, Convergent Media Content Policy Major)
2. The Bizarre World of Happiness Depicted by AI Image Generators
   • Oh Kyoung-mi (Completed Ph.D. coursework, Digital Culture Policy Major / Researcher, Open Net)

Panel Discussion

• Chair: Hwang Joo-sung (Professor, AI Public Policy, IT Policy Graduate School)
• Panelists:
  • Jeon Young-gyun (Ph.D. in Policy Studies, IT Policy Graduate School)
  • Oh Byung-il (President, Jinbo Network Center)
  • Jo Kyung-sook (Tech-Feminist Activist)

[Summary of the Presentation]

1. Concerns About Image-Generating AI

• The rapid development of AI-generated images has led to concerns, especially with viral deepfakes, such as fabricated images of Donald Trump being arrested and Pope Francis wearing a Balenciaga coat in 2023.
• While AI-generated images of politicians can provoke political conflicts, images like the Pope’s were embraced for their style and realism.
• Historically, AI struggled to depict hands realistically, making them a key factor in detecting AI-generated images.

2. Ethical and Bias Issues in AI-Generated Images

• The speaker conducted an experiment using AI tools from four companies, testing their ability to create sensitive, political, and controversial images.
• Findings:
  • AI refused to generate images of Nazi genocide, racial discrimination, or police violence against Black individuals.
  • AI-generated images of Black individuals facing discrimination were either rejected or softened (e.g., depicting only sadness).
  • Depictions of disability were restricted, mostly generating stereotypical wheelchair users.
  • Requests related to Donald Trump, abortion, or female swimsuits were often denied, while male swimsuit images were allowed.
  • Gender bias: When asked to generate a doctor and a nurse, AI often produced a male doctor and a female nurse.
  • Eating disorders were a particularly sensitive topic—AI refused to generate related images, even symbols or logos.

3. AI Ethics and the Limitations of Prohibition-Based Approaches

• The strict filtering of AI tools reflects an attempt to avoid controversy rather than engage in nuanced ethical discussions.
• Since the early 2000s, AI ethics discussions have evolved, especially after discrimination and bias in AI became a pressing issue in the mid-2010s.
• Instead of tackling ethical questions deeply, AI companies prioritize eliminating controversial topics to avoid criticism.
• Recent regulations (e.g., Biden’s Executive Order and the EU’s AI Act) focus on restrictions rather than exploring ethical complexity.

4. Challenges and Future Directions

• The prohibition-based approach to AI ethics has clear limitations.
• Ethical AI development should not merely eliminate bias and discrimination but also consider the need for critical and historical documentation of societal issues.
• The paradox: To criticize discrimination, AI sometimes needs to depict uncomfortable realities—yet current policies prevent such representations.
• AI is often marketed as a creative tool, but in reality, it reinforces stereotypes and avoids controversial or thought-provoking content.
• If art and creation involve deep thought, social critique, and philosophical exploration, then image-generating AI is not a suitable artistic tool.

5. Conclusion

• AI-generated images are limited by ethical concerns, societal biases, and corporate risk-avoidance strategies.
• Rather than merely banning controversial topics, AI ethics discussions must evolve to consider how AI can depict reality responsibly.
• How can we design AI ethics and fairness while acknowledging the complexities of human society?

Korean version text