The Meaning of Understanding in AI-Laden Science

Speakers

Siyu Yao, Dept. of Philosophy, Shanghai Jiao Tong University
André Curtis-Trudel, Dept. of Philosophy, University of Cincinnati

Abstract

Deep learning (DL) is a powerful scientific tool for classification, inference, and data emulation, but its opacity raises concerns about undermining epistemic virtues such as understanding and objectivity. Is it possible for scientists to retain understanding as DL permeates science? If so, what form of understanding is relevant? Throughout history, scientists often use complex “alien” tools effectively without full technical knowledge, relying on practical sense-making strategies.

Recent philosophy frames this as “pragmatic understanding,” where scientists learn how to apply tools without fully grasping their inner workings. Drawing on interviews with astronomers using AI, we identify strategies of establishing pragmatic understanding, such as embedding AI into existing methods, interpreting outputs through domain reasoning, developing design heuristics, testing performance, tracking errors, and fostering interdisciplinary collaboration. These practices show that understanding is plural and context-dependent. We argue that high-quality pragmatic understanding depends on building a diverse, interconnected set of practices, enabling reliable and generalizable use of DL through ongoing methodological learning across domains.

A philosophical examination of what it means to understand in an era of AI-laden science. Explore the relationship between AI, explanation, and understanding, and consider what scientific pursuits remain worthy in this new landscape.

Topics Covered

• AI-laden science and the relevance of philosophy
• AI, explanation, and understanding
• Empirical study of understanding
• What's worthy of pursuit now?
• Conclusion and discussion

Meeting Connection

Join the Meeting