Through qualitative methods, Assistant Professor John P. Nelson explores the possibilities of how AI can help and hurt technological progress
John Nelson
By Colin Bowyer, Communications Manager - December 23, 2025
In recent years, improvements in artificial intelligence (AI) have driven increasing attention to AI’s applications and implications for scientific research and technological development. It’s presumed that AI will help to accelerate scientific advancement, leading to economic growth and the development of solutions to global problems. But how plausible are these expectations for AI?
A new study lead-authored by School of Public Policy Assistant Professor John P. Nelson sheds light on the real-world impact of AI and machine learning (ML) on technological progress in manufacturing and materials science. Contrary to popular narratives of AI as a revolutionary force, the research suggests that AI is best understood as a powerful research and development tool, but one in keeping with the historical arc of scientific methods development.
Published as a preprint, titled "Can Artificial Intelligence Accelerate Technological Progress? Researchers’ Perspectives on AI in Manufacturing and Materials Science," the study draws on 32 in-depth interviews with U.S.-based academic researchers actively using AI/ML in their work.
“AI and machine learning allow scientists to search for promising candidate designs or technologies more cheaply and quickly, but those candidates still need to be validated by more conventional theory and empirical trials,” said Nelson. “For now, they’re tools, not magic bullets.”
The authors’ key findings include:
- Surrogate modeling: AI/ML can be used to create surrogate models for physics-based simulations, significantly reducing the cost and time of computational modeling and exploration of design spaces for materials and manufacturing processes.
- Phenomenological modeling: AI can model poorly understood phenomena, offering predictive capabilities where traditional theory falls short.
- Conditional benefits: While AI can solve previously intractable problems and identify patterns in large datasets, its reliability depends on dense, high-quality training data.
- Not a substitute for theory: Interviewees expressed skepticism about AI’s ability to generate disruptive scientific theories. AI complements traditional methods rather than replacing them.
- Risks and limitations: Concerns include over-reliance on AI, interpretability issues, high energy consumption, and the possibility that shortcuts enabled by AI could hinder long-term scientific discovery.
“We hope our study may also illustrate the value of contextualizing study of AI/ML within extant theories of technological change,” writes Nelson. “A detailed understanding of knowledge creation and technological development clarifies what AI/ML will need to do to genuinely transform these processes; and what its consequences may be.”
The study calls for balanced investment strategies that integrate AI/ML with conventional research methods, ensuring continued progress in both incremental and potentially disruptive innovations.