Published at 02/03/2025 17:49
464 words - 2.32 min of reading time
While watching a brief lecture on the history of kombucha in the USA, a few images in the presentation caught my attention.
These humorous images were created by the presenter, Hannah Crum.
These images sparked a thought: are artificial intelligences already capable of making memes?
Currently, AI-generated images and videos are awe-inspiring. However, I wondered if they could create "an idea, behavior, or style that spreads by means of imitation from person to person within a culture and often carries symbolic meaning representing a particular phenomenon or theme". [1]
Spoiler: probably not yet.
Obviously, understanding what makes content "memetic" is quite complex, and not exactly my area of expertise.
Still, I decided to test Chat-GPT to generate some potentially viral texts about kombucha, taking advantage of the current topic.
After several low-potential textual attempts, we managed to produce one that was a bit funnier, summarized below:
An image in two panels: in the first, someone asks "Why does this taste like vinegar?"; in the second, surrounded by kombucha bottles, they say "I need my probiotic elixir."
Next up was image generation, since a picture is worth a thousand words, and the sharing potential (self-propagation) would be greater.
Taking advantage of Chat-GPT's integration with DALL-E, I made a few attempts, shown below.
The images where generated with Portuguese text ("Por que isso tem gosto de vinagre?", and "Preciso do meu elixir probiótico").
With no more free attempts left for the day, my current conclusion is that the latest models' ability to understand what might lead to the self-propagation or spread of an idea in the human mind is still limited.
Additionally, DALL-E struggled to grasp the concept of "two panels", and to handle text within images.
Maybe because there is no full understanding of humor, persuasion, or even how to create coherent comic strips.
Or maybe I was just a bad comic writer. I do not know.
I decided to edit the last image, aiming to make it somewhat more "memetic."
Perhaps adding text to the bottles in the last panel (or even replacing them altogether) would improve spreadability.
As unlikely as it may be, it still seems better than the initial suggestion made by Chat-GPT.
