Tom Mitchell
Founders University Professor
- Pittsburgh PA UNITED STATES
Tom Mitchell is a longtime researcher in AI, including new systems that can improve healthcare, education, climate and more.
Biography
Much of his current research focuses around two projects:
Machine learning approaches to analyzing human brain activity. This project uses functional Magnetic Resonance Imaging (fMRI) to capture three-dimensional images of human brain activity at a spatial resolution of 1mm, once per second. This is a wonderful set of data for studying the operation of the human brain, and because it is relatively new, there is a great need for new algorithms to analyze the data. Recently demonstrated that it is possible to train machine learning algorithms to decode mental states of human subjects (e.g., to determine whether the word a person is examining is a noun or a verb) based on their observed fMRI brain activity. Interested in developing new algorithms that will help discover the spatial-temporal patterns of activity associated with a variety of brain processes, and that will help better understand the working of the human brain.
Intelligent workstation assistants that learn to help their users. This is part of a large multi-researcher project to build enduring, personalized, learning assistants for users of computer workstations (like us!). He is working toward a software agent that can understand the user's email, calendar, text files, and actions, and that can learn the user's interests, habits, and tasks, in order to help in a wide range of activities. Specific interest lies in how to make the agent learn. For example, how the agent can learn to automatically extract information from text emails and files, and how it can learn what threads of activities the user is involved in, when, with whom, about what, etc.
Areas of Expertise
Media Appearances
How AI is infiltrating labor union contracts, in Pittsburgh and beyond
Technical.ly online
2025-03-26
As unions consider how they want to cover AI in contracts, Tom Mitchell (School of Computer Science) believes white-collar, college-educated, knowledge-based professions might be the most vulnerable to disruptions from AI tools.
As artificial intelligence surges across daily life, so do concerns AI
TribLIVE.com online
2024-09-08
Tom Mitchell, who founded Carnegie Mellon’s Department of Machine Learning in 2006, gave a simple example: You could show a computer program photos of your mother and then photos of people who are not your mother. With the gained experience, the program would be able to identify the features that distinguish who is a positive example of your mother and who is not.
Why a YouTube Chat About Chess Got Flagged for Hate Speech
WIRED online
2021-03-01
“Fundamentally, language is still a very subtle thing,” says Tom Mitchell, a CMU professor who has previously worked with KhudaBukhsh. “These kinds of trained classifiers are not soon going to be 100 percent accurate.”
Media
Social
Accomplishments
10-Year Outstanding Research Contributions Award, Brain Informatics Conference
2017
Alan Perlis Award for Imagination in Computer Science, Carnegie Mellon University
2018
President’s Medal, Stevens Institute of Technology
2018
Best Paper Award at the User Interface Software and Technology (UIST) Conference
2020
Best Dataset Paper Award at the Learning at Scale Conference
2024
Education
Massachusetts Institute of Technology
S.B.
Electrical Engineering
1973
Stanford University
Ph.D.
Electrical Engineering
1979
Affiliations
- Generative AI Task Force : Chair
Articles
Towards valid student simulation with large language models
arXiv preprint arXiv:2601.05473Zhihao Yuan, Yunze Xiao, Ming Li, Weihao Xuan, Richard Tong, Mona Diab, Tom Mitchell
2026-01-09
This paper presents a conceptual and methodological framework for large language model (LLM) based student simulation in educational settings. The authors identify a core failure mode, termed the "competence paradox" in which broadly capable LLMs are asked to emulate partially knowledgeable learners, leading to unrealistic error patterns and learning dynamics. To address this, the paper reframes student simulation as a constrained generation problem governed by an explicit Epistemic State Specification (ESS), which defines what a simulated learner can access, how errors are structured, and how learner state evolves over time. The work further introduces a Goal-by-Environment framework to situate simulated student systems according to behavioral objectives and deployment contexts. Rather than proposing a new system or benchmark, the paper synthesizes prior literature, formalizes key design dimensions, and articulates open challenges related to validity, evaluation, and ethical risks. Overall, the paper argues for epistemic fidelity over surface realism as a prerequisite for using LLM-based simulated students as reliable scientific and pedagogical instruments.
Read and reap the rewards: Learning to play atari with the help of instruction manuals
Advances in Neural Information Processing Systems2024
High sample complexity has long been a challenge for RL. On the other hand, humans learn to perform tasks not only from interaction or demonstrations, but also by reading unstructured text documents, eg, instruction manuals. Instruction manuals and wiki pages are among the most abundant data that could inform agents of valuable features and policies or task-specific environmental dynamics and reward structures. Therefore, we hypothesize that the ability to utilize human-written instruction manuals to assist learning policies for specific tasks should lead to a more efficient and better-performing agent. We propose the Read and Reward framework. Read and Reward speeds up RL algorithms on Atari games by reading manuals released by the Atari game developers. Our framework consists of a QA Extraction module that extracts and summarizes relevant information from the manual and a Reasoning module that evaluates object-agent interactions based on information from the manual.
Spring: Studying papers and reasoning to play games
Advances in Neural Information Processing Systems2024
Open-world survival games pose significant challenges for AI algorithms due to their multi-tasking, deep exploration, and goal prioritization requirements. Despite reinforcement learning (RL) being popular for solving games, its high sample complexity limits its effectiveness in complex open-world games like Crafter or Minecraft. We propose a novel approach, SPRING, to read Crafter's original academic paper and use the knowledge learned to reason and play the game through a large language model (LLM). Prompted with the LaTeX source as game context and a description of the agent's current observation, our SPRING framework employs a directed acyclic graph (DAG) with game-related questions as nodes and dependencies as edges. We identify the optimal action to take in the environment by traversing the DAG and calculating LLM responses for each node in topological order, with the LLM's answer to final node directly translating to environment actions.
Combining computational controls with natural text reveals aspects of meaning composition
Nature Computational Science2022
To study a core component of human intelligence—our ability to combine the meaning of words—neuroscientists have looked to linguistics. However, linguistic theories are insufficient to account for all brain responses reflecting linguistic composition. In contrast, we adopt a data-driven approach to study the composed meaning of words beyond their individual meaning, which we term ‘supra-word meaning’. We construct a computational representation for supra-word meaning and study its brain basis through brain recordings from two complementary imaging modalities. Using functional magnetic resonance imaging, we reveal that hubs that are thought to process lexical meaning also maintain supra-word meaning, suggesting a common substrate for lexical and combinatorial semantics.