SofAI

What does it mean to be a generative entity?

SofAI is an event where participants act as part of a speculative machine learning system. This project provides alternative ways of understanding labeling, text prediction, image generation, and user profiling through analog and embodied activities. 

All acts of creation connect us to material and social histories, reminding us how knowledge is co-created (not just emanating from our screens). The activities highlight the places where humans are involved in the machine learning process and help us consider how these processes could be different if executed fully by humans.

More broadly, this project hopes to collectively situate our place as humans in the grand picture of technological progress and start thinking about what future we want to create.

“breaking out of the black box”

+ AI butthole logo

physical invitations!

Bringing attention to data workers’ invisible labor and the immense amount of human input that goes into training datasets.

Asking what gets left out of the bounding boxes by encouraging participants to label not just objects, but also to identify emotions and actions in the picture, leading to more subjective ways of seeing.

Large language models lack context, and create synthetic text using next-word prediction: what tokens or “chunks” are statistically likely to follow.

Using this idea, participants add one word in response to different prompts, building on ideas of crowdsourced knowledge to create an almost collaborative poetry.

Images start as static noise, becoming clearer through many many iterations. For the activity, images are created through 30-second rounds

Original “noise” images made in touchdesigner with Tabi Cass

As guests arrived, they were all asked to fill out a card, including a user’s name and three data points about them

As the last activity, cards were randomly distributed and participants acted as the LLM to respond to the same prompt, but tailor their answers for their user.