AI as Boundary Object
Motivation
Accelerated materials discovery is an interdisciplinary research team comprised of computer scientists and chemists who work together to develop generative AI models to accelerate the discovery of polymers. However, this interdisciplinarity causes challenges in algorithm design and use.
Method
2-phased study:
8 unstructured interviews with computer scientists and chemists working in AMD interdisciplinary team. Interviews were analyzed using thematic analysis and were used to develop problem scenarios.
4 participatory design sessions with 1 computer scientist and 1 chemist in each session. Problem scenarios were used as prompts to facilitate discussion.
Depiction of GPT and Scenarios as Boundary Objects
Findings
There is a difference in tacit knowledge, terminologies, concepts, and data structures between computer science and chemistry. This manifests in algorithm design and makes it difficult to capture features for AI from chemical data. Implementing factsheets or datasheets in model development can help with explainability and interpretation.