Research

Our research develops the foundations, methods, and applications of causal artificial intelligence, with a focus on building trustworthy, explainable, and personalized AI through causal reasoning.

We study fundamental questions in causal inference and counterfactual reasoning, develop causal approaches to decision making and machine learning, and translate these advances into real-world applications in education, neuroscience, and health.

Causal Foundations

We develop the mathematical and computational foundations of causal and counterfactual reasoning. Our research focuses on the identification and partial identification of causal quantities, including probabilities of causation and other measures for reasoning about individual-level causal effects.

We are particularly interested in understanding what can be learned when causal knowledge or data are incomplete, and in developing principled methods that combine observational, experimental, and structural causal information.

Topics: Structural Causal Models · Counterfactual Reasoning · Probabilities of Causation · Identification · Partial Identification · ε-Identifiability

Causal Decision Making

We study how causal and counterfactual reasoning can support trustworthy, explainable, and personalized decision making.

Rather than asking only what is likely to happen, causal decision making asks what would happen under alternative actions and which action is most appropriate for a particular individual. Our research develops methods for decision making under uncertainty, incomplete causal knowledge, and limited data, with applications to personalized interventions and policy decisions.

Topics: Causal Decision Making · Personalized Decision Making · Counterfactual Decisions · Unit Selection

Causal Machine Learning

We develop machine learning methods that incorporate causal structure, causal knowledge, and counterfactual reasoning into learning and prediction.

Our work investigates how causal information can improve learning when data are limited, heterogeneous, or collected under changing environments, as well as how machine learning can support causal discovery and causal inference. The goal is to build AI systems that move beyond statistical associations toward models capable of reasoning about causal relations.

Topics: Causal Machine Learning · Causal Representation · Generalization · Learning with Causal Knowledge

Applications

Our foundational and methodological research is motivated by real-world problems in which understanding what causes an outcome, what would happen under an intervention, and which intervention should be chosen is essential.

Causal AI for Education

We develop causal AI methods for understanding and improving student learning, engagement, and educational decision making.

Our research studies heterogeneous and individual-level effects of educational interventions, including emerging AI technologies, and develops methods for personalized educational decisions. We are particularly interested in combining large-scale educational data with causal and counterfactual reasoning to determine not only whether an intervention works, but for whom it works and under what conditions.

Causal AI for Neuroscience

We investigate how causal and counterfactual reasoning can help understand the causal organization of the human brain.

Using neuroimaging and other neuroscience data, we study causal relationships among brain regions and networks and develop methods for reasoning about the effects of interventions when the underlying causal structure is only partially known. Our goal is to complement association-based analyses with causal questions about how changes in neural activity may influence cognition and behavior.

Causal AI for Health

We develop causal AI methods to support personalized and explainable healthcare decision making.

Our research focuses on reasoning about treatment effects at the individual level, integrating multiple sources of causal evidence, and making decisions when data or causal knowledge are incomplete. Ultimately, we aim to help determine which intervention is most appropriate for which individual and why.