Tom Kerby
Assistant Professor of Statistics, Brigham Young University
Learning, reasoning, and agency in humans and AI
Contact
Email: tkerby@stat.byu.edu
Office: WVB 2196
Phone: 801-422-7054
Research
I study learning, reasoning, and decision-making in humans and AI from a statistical perspective. I view modern AI as a form of learned, probabilistic computing whose behavior emerges from data rather than explicit programming. Understanding such systems—and designing them to support human judgment—raises fundamental questions about dependence, uncertainty, generalization, measurement, and decision-making.
My work spans both machine and human learning. I develop statistical and information-theoretic methods for understanding learned representations and generalization in AI systems, and I develop new approaches to measuring how people learn, reason, and make decisions through interactions with AI. Across both, I am interested in how uncertainty, information, and statistical evidence should inform decisions without unnecessarily replacing human judgment or agency.
Understanding AI Systems
What do complex learned systems learn, and what properties persist?
Statistical and information-theoretic methods for characterizing dependence, structure, and generalization in learned representations and generative models.
Human-AI Learning and Measurement
What can we infer about human knowledge and reasoning from rich interactions with AI?
New approaches to measuring latent knowledge, reasoning, and decision-making through adaptive AI-mediated assessment, simulation, and experimentation.
Trustworthy Human-AI Systems
How should uncertain information be used when humans and AI make decisions together?
Statistical and empirical approaches to information, risk, and appropriate delegation in AI-assisted decisions, with attention to human agency and trust.
Selected Current Projects
Local CorEx
Characterizing high-dimensional dependence structure in learned neural network representations using total correlation, with applications to interpretability and robustness.
AI-Mediated Assessment & Measurement
Using adaptive, interactive AI to measure latent knowledge and reasoning from rich behavioral traces—treating assessment as a statistical measurement problem.
RedactaRAG
A policy-aware retrieval and redaction framework for LLM workflows, studying how information access, uncertainty, and competing objectives should shape AI-assisted decisions.
Diffusion Model Generalization
Identifying the statistical regularities that persist across independently trained generative models—a question about what structures generalize under the randomness of data and training.
Interested in doing research with me?
I work with students interested in understanding AI systems and in studying how humans learn, reason, and make decisions with AI. Projects range from mathematical and computational work on dependence, representation, and generalization in neural networks to statistical and experimental work on measurement, assessment, and decision-making with AI. These projects provide opportunities for students to apply and extend statistical thinking to some of the emerging challenges created by modern AI.
Apply to Join Our Research →Background
I earned my PhD in Statistics & Data Science from Utah State University in 2025, where I developed algorithms for learning higher-order interactions from high-dimensional data and extended training-free guidance techniques to discrete diffusion models. My dissertation, “Graph-Based Machine Learning: Higher-Order Interactions, Guided Generation, and Knowledge-Graph Tools,” was advised by Kevin Moon.
Prior to joining BYU, I gained industry experience through internships at Idaho National Laboratory (anomaly detection for cybersecurity), Enveda Biosciences (molecular property prediction), and FamilySearch (graph-based methods for assisted family tree construction).
I received my undergraduate degree from BYU in Statistics with minors in Computer Science, Mathematics, and Molecular Biology.
Latest News
Our work "Temporal Geometry of Diffusion Model Generalization" was accepted at TAG-DS 2026 as a Spotlight Oral Presentation in the 4-page Non-archival Extended Abstract Track; Non-archival because the work is currently under review at a different venue.
Our research group had an outstanding conference at the Red Rock Data Science Conference (RRDS) in St. George, Utah! Our students presented cutting-edge research including: Jun Kim on the generalizability of diffusion models; Timothy Hardy on RedactaRAG, a policy-based security framework for LLM interactions; Ridge Atwood and Cooper Maughan on Local CorEX for analyzing neural network robustness; and Patrick Willmott presented a poster on practice problem generation for automated decision-based assessment. Beyond excellent presentations, the group enjoyed memorable outdoor experiences hiking The Narrows at Zion National Park, exploring lava flows and tubes, and discovering a not well known beautiful slot canyon in Snow Canyon.
Our paper, "Nexarag: Democratizing Reproducible Knowledge Graph Contexts for LLM Research," was published in the Journal of Open Source Software. The paper is available here: https://doi.org/10.21105/joss.09910
Congratulations to Ridge Atwood and Tim Hardy on their acceptance to the BYU MS Statistics program.
Our research group gave outstanding presentations at the BYU Student Research Conference. Special congratulations to Tim Hardy, who won 1st place in his session for his presentation, "RedactaRAG: Strengthening Policy-Based Security in LLM Interactions."
Our paper, "Compilation of a Nationwide River Image Dataset for Identifying River Channels and River Rapids via Deep Learning," was published and featured in Remote Sensing. The paper is available here: https://www.mdpi.com/2072-4292/18/2/375
Officially joined the Department of Statistics at Brigham Young University as an Assistant Professor.
Presented research on NexaRAG at the Utah AI Summit.