home | about | works | contact

AI Classroom

During Winter quarter of 2023, I took the class Machine Intelligence and Art, which explored the ways AI could be used as a tool, topic, and muse for art. For a project, my classmates and I decided to test particular text-to-image models in their capabilities in generating text as images.

Starting in early January of 2023, the timing of this class corresponded with the rise of ChatGPT and the concerns educators had on its effects on education. Thus, I decided to make my project classroom-themed.

Using DALLE and Nightcafe, I generated posters that are often seen in a classroom, including a world map, a periodic table, and some inspirational quotes.

alphabet poster
alphabet poster generated by DALLE

world map poster
world map poster generated by DALLE

motivational quote poster
motivational quote poster generated by DALLE

periodic table poster
periodic table poster generated by DALLE

As you can see, DALLE did a pretty good job at capturing the aesthetic essence of these posters, but it lacked actual accuracy of text as image content. In comparison, the world map looks better (although we have less familiarity with what the world map exactly looks like, so it is not really a fair comparison).

To present the posters, we generated a classroom wall via DALLE, then extended it from its square dimensions using runway's Infinite Image model. We then projected this image onto the wall and hung the posters, creating the illusion that the posters were in an actual classroom.

poster display
posters on display

We also used a laser cutter and 3d printer to create a puzzle based on the incorrect generated alphabet.

puzzle
alphabet puzzle based on generated alphabet poster

We ended up with a poster that couldn't fit on the projection wall, so we ended up putting it up on the door to the space - will someone notice it?

safety poster
safety poster on display

While we should be wary of AI's role in education, it is clear that these text-to-image models have some room to grow before they are implemented in classrooms in this way. This also serves as a warning to not blindly trust information that is generated through an AI model. Although these models are surprisingly good for many things, there is a limit to their capabilities (at this point in development).