Prototyping social computing systems through generative AI modeling

PI: Michael Bernstein, Percy Liang


Prototyping is a core epistemic activity to design, enabling the designer to experience how their system might be adopted or co-opted and refine the design in reaction to this. Unfortunately, techniques for prototyping of social systems such as social networks and collaborative software are stuck at small scale, and are unable to help a designer understand how a space might behave with diverse participants, or what kind of antisocial behavior, disagreements, or harassment might arise as many people join. We will draw on recent advances in generative NLP models (e.g., GPT-3) to create a tool that empowers a designer to specify a distribution of user personas and populate a social system with these personas and their interactions with each other. This prototyping tool will enable the designer to understand how a social system might get used or abused, and iterate on the design of the platform to tune norms more tightly toward the designer’s goals.

We will evaluate this tool by tracking the changes that social computing designers make to their designs in response to our tool, in comparison to alternative social computing best practice prototyping approaches.


Joon Sung Park