In order to successfully help young children, adults must understand their intentions. Similarly, for robots to provide effective support, they must be able to do this. We examined how adults and AI can infer intents from incomplete puzzles. We are interested in the reasoning strategies that emerge in reasoners and whether they predict success.  Furthermore, we seek to understand whether AI has similar success in inferring the puzzles. We hypothesized that the number of correctly connected pieces correlates with the reasoners’ accuracy. Additionally, in the human sample, we speculated that the frequency of some reasoning strategies as well as the reported confidence are associated with their accuracy. Responses from 530 participants were asked to identify the puzzle the child was assembling, explain their reasoning process, and their confidence. We asked four visual language models the same questions and assessed the accuracy of their inferences. This project was supported by funding from F&M's Committee on Grants Program.

Project Mentor:  Professor Willie Wilson, Department of Computer Science