Chacha Chen

Chacha Chen

陈诧姹

PhD student
Chicago Human+AI lab
Department of Computer Science, University of Chicago
Email: chacha@uchicago.edu

I am a PhD student in Computer Science at UChicago, advised by Chenhao Tan.

Research Focus

I am passionate about putting humans at the center of AI development, including how to understand the process of human-AI interaction and how we can better build/deploy AI tools that can actually be useful in real-world settings.

Human-AI collaboration dynamics: My work showed that effective collaboration requires accounting for human domain expertise, you can't just deploy AI and expect complementarity. I found laypeople over-rely on AI while experts under-rely. I'm exploring how personalized AI systems can dynamically calibrate their role in the collaboration, adjusting not just to expertise level but to individual working styles, task contexts, and evolving user needs to achieve genuine complementarity.

Improving AI in expert domains where capability still falls short: As AI has become capable in general domains, the natural next frontier is expert domains like medical diagnosis and scientific discovery where current systems still fall short. I study frontier models on highly specialized tasks to understand where they break and how we can improve models for these domains.

Calibration and uncertainty communication: Capability isn't the bottleneck anymore, calibration is. For AI to work effectively and collaboratively with humans, systems must communicate what they don't know. I'm building information-theoretic frameworks for LLM calibration and collaborative decision making, enabling genuine complementarity where humans and AI can both contribute meaningfully.


Selected Publications (Full List)

LLMs in Expert Domains

GPT-4V Cannot Generate Radiology Report Yet
Yuyang Jiang*, Chacha Chen*, Dang Nguyen, Benjamin M. Mervak, Chenhao Tan.
We systematically evaluate GPT-4V's ability to generate radiology reports and identify key limitations in clinical reasoning and structured reporting.
ML4H 2024, NAACL 2025. [Paper]
The Use of Generative Search Engines for Knowledge Work and Complex Tasks
Siddharth Suri, Scott Counts, Leijie Wang, Chacha Chen, Mengting Wan, Tara Safavi, Jennifer Neville, Chirag Shah, Ryen W. White, Reid Andersen, Georg Buscher, Sathish Manivannan, Nagu Rangan, Longqi Yang.
An empirical analysis of large-scale user interactions with Bing Chat, a GPT-4 backed generative search engine.
[Preprint]

Human-AI Collaboration

Can Domain Experts Rely on AI Appropriately? A Case Study on AI-Assisted Prostate Cancer MRI Diagnosis
Chacha Chen, Han Liu, Ziyang Guo, Jiamin Yang, Benjamin M. Mervak, Bora Kalaycioglu, Grace Lee, Emre Cakmakli, Matteo Bonatti, Sridhar Pudu, Osman Kahraman, Gul Gizem Pamuk, Aytekin Oto, Aritrick Chatterjee, Jessica Hullman, Chenhao Tan.
We find that radiologists under-rely on AI assistance in prostate cancer diagnosis, missing opportunities for improved accuracy even when AI predictions are correct.
FAccT 2025 [Paper]
Machine Explanations and Human Understanding
Chacha Chen*, Shi Feng*, Amit Sharma, Chenhao Tan.
FAccT 2023, TMLR 2023, and Best Paper at HMCaT @ ICML 2022
[Paper] [Slides] [Video]
[Machine Explanations Human Studies Literature Survey]
Towards a Science of Human-AI Decision Making: A Survey of Empirical Studies
Vivian Lai, Chacha Chen, Q. Vera Liao, Alison Smith-Renner, Chenhao Tan.
A comprehensive survey of 100+ papers on human-AI decision making, identifying key factors and proposing a framework for future research.
FAccT 2023. [Paper] [Slides] [Survey Website]
Learning Human-Compatible Representations for Case-Based Decision Support.
Han Liu, Yizhou Tian, Chacha Chen, Shi Feng, Yuxin Chen, Chenhao Tan.
We develop a method to learn case representations that align with human similarity judgments for more effective AI-assisted decision support.
ICLR 2023. [Paper]

Applied Machine Learning

UNITE: Uncertainty-based Health Risk Prediction Leveraging Multi-sourced Data
Chacha Chen, Junjie Liang, Fenglong Ma, Lucas Glass, Jimeng Sun, Cao Xiao.
A framework for health risk prediction that quantifies uncertainty when integrating multiple data sources like claims and lab results.
WWW 2021. [Paper] [Slides]
Toward a Thousand Lights: Decentralized Deep Reinforcement Learning for Large-Scale Traffic Signal Control
Chacha Chen, Hua Wei, Nan Xu, Guanjie Zheng, Ming Yang, Yuanhao Xiong, Kai Xu, Zhenhui Li.
A scalable decentralized reinforcement learning approach for coordinating traffic signals across large urban networks.
AAAI 2020. [Paper]

Experience

Apple, AIML

Research Residency, 2025-now

Microsoft Research, Redmond, WA, US

Research Intern, Summer 2023

Amazon, AWS, Santa Clara, CA, US

Applied Scientist Intern, Summer 2022

IQVIA, Analytics Center of Excellence, Boston, MA, US

Machine Learning Research Intern, Summer 2020