Julian Skirzyński

About

I am a Ph.D. candidate in Computer Science at UC San Diego, advised by Berk Ustun, and a member of ARIA, the NSF AI Research Institute on Interaction for AI Assistants. My research sits at the intersection of machine learning, human-computer interaction, and cognitive science. The main question I study is whether AI helps people solve problems better, and how to build AI that does. To answer it, I work towards a science of human-AI collaboration: I measure how people work with AI, learn from those measurements what people actually need, and engineer systems that meet those needs. For instance, I found that interpretability helps or harms depending on the task. People overrely on interpretable models when they are asked to supervise them, yet an interpretable system we built to teach decision strategies measurably improved people's decisions. I pursue this science wherever people cannot or should not fully defer to AI outputs, e.g., in policy, health, finance, and beyond.

Before my Ph.D., I was a research scientist at the Max Planck Institute for Intelligent Systems, where I worked with Falk Lieder on interpretable reinforcement learning and interventions to improve human planning. I hold an M.S. in Computer Science from McGill University, and an M.S. in Cognitive Science and a B.S. in Mathematics and Cognitive Science from the University of Warsaw. You can access my full CV here and read my research statement here.

Recent news

• Aug 2026Attended the Summer Bootcamp on Trustworthy AI at Brown University.
• Jun 2026New preprint: Measuring What Matters: Synthetic Benchmarks for Concept Bottleneck Models.
• Apr 2026Started a new project on the faithfulness of LLM justifications.
• Dec 2025Cognitive Bias Induction in LLMs accepted at IJCNLP 2025.
• Jun 2025Presented Discrimination Exposed? at ACM FAccT 2025 in Athens.

Selected Papers

• 
  Measuring What Matters: Synthetic Benchmarks for Concept Bottleneck Models
  Julian Skirzyński, Harry Cheon, Shreyas Kadekodi, Meredith Stewart, Berk Ustun
  In Submission, 2026
  [Paper] [Code]
• 
  On the Value of Interpretability in Human Decision-Making
  Julian Skirzyński, Elena Glassman, Berk Ustun
  In Submission, 2025
  paper & code coming soon
• FAccT
  Discrimination Exposed? On the Reliability of Explanations for Discrimination Detection
  Julian Skirzyński, David Danks, Berk Ustun
  ACM Conference on Fairness, Accountability, and Transparency, 2025
  [Paper] [Code]
• Computational Brain & Behavior

  
  Automatic discovery and description of human planning strategies
  Julian Skirzyński, Yash Raj Jain, Falk Lieder
  Computational Brain & Behavior, 2023
  [Paper] [Code]
• Computational Brain & Behavior

  
  Boosting Human Decision‑making with AI‑Generated Decision Aids
  Frederic Becker, Julian Skirzyński, Falk Lieder
  Computational Brain & Behavior, 2022
  [Paper] [Code]
• Machine Learning Journal

  
  Automatic discovery of interpretable planning strategies
  Julian Skirzyński, Frederic Becker, Falk Lieder
  Machine Learning Journal, 2021
  [Paper] [Code]