The panelists, including myself, didn’t see what was being projected on the screen. The goal was not to influence the discussion but to raise awareness among the audience about how generative AI models visually interpret speech. The origin of the AI images was revealed only after the first hour, giving the audience some time to ponder their meanings. Most assumed someone was manually typing prompts somewhere in the backstage. Realizing the process was actually fully automated caused an unsettling feeling in the room.

Some images were expected, some surprising. The most hilarious examples were DALL-E 3’s attempts to include buzzwords mentioned in the discussion, sometimes even adding a critical undertone when mentioning entities like Google. The reactions of the audience, from surprise to amusement, highlighted the power and limitations of AI in capturing and representing human discussions.

This approach not only raises awareness of AI’s imperfections but also shows that advanced prompt engineering is still necessary to harness AI effectively in creative contexts.