Recovering Hidden Human Knowledge: Assistant Professor Haw-Shiuan Chang
College of Information Science Assistant Professor Haw-Shiuan Chang bridges machine learning, natural language processing and information retrieval, focusing on fundamentally narrowing the gap between large language models (LLMs) and human intelligence.
Photo by Michael McKisson.
What excites me is that by understanding how knowledge is formed and organized in humans’ minds, we can improve the learning processes of both humans and large language models.
Haw-Shiuan Chang joined the College of Information Science as an assistant professor in 2026 after completing his PhD at the University of Massachusetts and serving as a postdoctoral scientist at Amazon. By studying how we can recover hidden human knowledge from what people write and use it to teach large language models to think more like we do, he fundamentally asks: What do we actually mean by intelligence?
What brought you to the College of Information Science?
Before joining the University of Arizona, I was a postdoc at Amazon and the University of Massachusetts Amherst. It was a long journey to get here.
In middle school I loved playing online games, like most kids, and I dreamed of designing an educational online game one day. I soon realized how difficult that would be, and one of the main reasons is that we still don't really understand how humans think. That question has guided me for more than ten years, through AI, machine learning and now language models.
Today I believe one of the biggest bottlenecks in AI is data, and data comes from humans. The College of Information Science brings together experts from many fields who care about data, humans, education, games and AI at the same time. That is why I wanted to continue my academic journey here.
What is your current research, and what most excites you about this work?
I am looking for ways to identify and close the fundamental gap between large language models (LLMs) and humans. Today's LLMs learn from internet text, published work and verifiable tasks such as coding and math. What they never see is the thinking that produced those texts (e.g., the reasoning, the intuition, the false starts that a writer rarely puts on the page). I study how we can recover that hidden human knowledge from what people write, and use it to teach LLMs to think more like we do. What excites me is that by understanding how knowledge is formed and organized in humans’ minds, we can improve the learning processes of both humans and LLMs.
What courses are you focusing on?
I am designing a course on reinforcement learning (RL). A typical RL course is full of formulas; I want to teach one that emphasizes the high-level concepts over the mathematical details. I enjoy building a new course because it is a chance to organize and share my own research experience and perspective. My goal is for students to leave with tools they can use to solve the problems they actually care about.
How do you bring your research and experience into your teaching?
I spent two years as a postdoc at Amazon, where I took part in their LLM development effort. One of my ongoing research directions is identifying the fundamental differences between humans and LLMs. I look forward to bringing both into the classroom: what I saw of how these systems get built, and the harder question underneath it: What do we actually mean by intelligence?
Tell us a bit more about your teaching approach.
One advantage of using RL to train LLMs is that the model can reach a goal in whatever way suits it. Our job is to recognize and reinforce the good behavior, which a verifier judges. I would hope my classroom works in a similar way. I hope that students can learn to use the concepts taught in the class to solve the problems they care about in their projects. And I want to be the verifier who tells them when they are on the right track. How to do this well in a classroom is something I am still working out.
Beyond research and teaching, what are your passions?
I enjoy spending time with my family: playing chess and board games, or just watching YouTube together.
What advice do you have for InfoSci students?
Motivation really matters. I have seen some highly motivated students who did not have very good technical skills at the beginning but turned out to achieve something incredible by the end. If you focus on where you really want to go, even if you move slowly, you will often get closer to your goal than you thought at the end.
Learn more about Haw-Shiuan Chang on his faculty page, or explore ways you can support the dynamic, student-invested faculty of the College of Information Science.