About the AI and Science Initiative

Overview

Artificial intelligence (AI) and machine learning (ML) will fundamentally change the nature and pace of scientific discovery.

Widespread adoption of AI in the sciences has the potential to integrate scientific inquiry with modes of hypothesis generation, data analysis, simulation, and testing that will transform our capacity to address scientific problems that currently appear intractable.

Successfully applying AI to science requires a radical rethinking of how best to pose and answer research questions. The University of Chicago is ideally suited to accelerate this transformation because of its expertise in training scholars in a mode of inquiry that is intensely rigorous, theory-driven, systematic, interdisciplinary, and intrinsically open to critique. The UChicago Approach has always been to question fundamental assumptions, frame problems at as deep a level as possible, and develop novel theoretical and methodological foundations that push the boundaries of new disciplines.

The rise of AI in science and engineering presents both a remarkable opportunity and a profound challenge to human-centered modes of inquiry. Beyond training scientists to effectively use existing AI tools, the UChicago vision of accelerating AI+STEM research includes systematically reimagining the role of AI in the scientific discovery process.

AI uncovering new laws of nature: Given observations of a system, use AI to uncover the governing physical laws. For example, physicists are developing novel tools to learn equations that describe physical forces exerted by and within cells using videos of cells, overturning classical biophysical models. Similar efforts focus on learning the dynamics of bacterial communities so we can better manage resilient ecologies in a changing environment.

AI guiding scientific measurement: Use AI to design better experiments, simulations, and sensors. Particle accelerators and sky surveys are just two examples of measurement systems that collect vastly more data than can be preserved. With AI-enabled real-time data analysis, we can rapidly filter data with dramatically higher accuracy. AI methods can also be used to help navigate a broad landscape of possible molecular structures to design efficacious and environmentally sustainable materials (such as batteries) or new proteins.

Physics-informed machine learning: Optimally leverage physical models and experimental or observational data. Examples include learning new climate forecasters, using machine learning together with physics-based models to uncover material structure and properties, and performing medical image reconstruction with unprecedented speed and accuracy.

Scientific discovery advancing AI frontiers: The AI+Science Initiative will develop new AI theories, methods, and open-source software tools inspired by scientific settings that will be able to simultaneously solve fundamental challenges across many disciplines. Examples include machine learning on graphs, neural networks for studying differential equations, generative models for science, biologically-inspired AI approaches informed by neuroscience, joint representations of data from disparate sources, and calibration and inference methods for quantifying the uncertainty associated with learned models.