Catch the Pig! Understanding Interaction Between People and AI
Author:
Akil Atkins ’22
Co-Authors:
Faculty Mentor(s):
Chris Dancy, Computer Science
Funding Source:
NSF Grant
Abstract
Our study sought to investigate the ways in which people would interact with an AI agent, based on how the agent was racialized. To investigate this we modeled a game after the stag hunt task, where participants were tasked with gaining as many points as possible. Participants could gain points by either cooperating with an AI agent to capture a pink game piece, which represented a pig or exiting the game through the black squares on either side of the board. The study was a true experiment in which participants were randomly assigned to one of three conditions. Participants were either assigned to a condition where the AI was racialized as Black, where the AI was racialized as white, or a condition where the AI wasn’t racialized at all, which represented the control condition. After completing the game participants were asked three survey questions to assess how they perceived the AI’s strategy when playing the game. So far our results have shown that participants in the control condition were more likely to believe the AI was working with them to capture the pig than participants in both the Black and white treatment groups. Moreover, the participants in the white treatment group were more likely than those in the Black treatment group to believe the AI agent was working them to capture the pig. The results do suggest that there is a relationship between the racialization, or lack thereof, of AI and how people interact with AI Agents.