Search for: All records

ABSTRACT Artificial Intelligence (AI) methods are valued for their ability to predict outcomes from dynamically complex data. Despite this virtue, AI is widely criticized as a “black box” i.e., lacking mechanistic explanations to accompany predictions. We introduce a novel interdisciplinary approach that balances the predictive power of data-driven methods with theory-driven explanatory power by presenting a shared use case from four disciplinary perspectives. The use case examines scientific career trajectories through temporally complex, heterogeneous bibliographic big data. Topics addressed include: data representation in complex problems, trade-offs between theoretical, hypothesis driven, and data-driven approaches, AI trustworthiness, model fairness, algorithm explainability and AI adoption/usability. Panelists and audience members will be prompted to discuss the value of approach presented versus other ways to address the challenges raised by the panel, and to consider their limitations and remaining challenges.