Dr. Karen Chen is an Assistant Professor in the department and leads the lab for Informatics for Human Flourishing. Chen received an MS in Statistics, and an MS and PhD in Information Systems all from Carnegie Mellon. Chen’s research is guided by an overarching framework in promoting human flourishing by leveraging the power of data and human-centered computing under the umbrella of “Informatics for Human Flourishing.”
Monday, April 15, 12-1pm, Dr. Chen will give a virtual presentation titled, “The GPT Dilemma: Navigating the Pros, Cons, and Impacts on Learning.” Chen talks about what to expect from the presentation, how this work came about and how to leverage GPT as a productive tool in the classroom. The presentation is part of the AI, Privacy, and Ethics Symposium sponsored by the Albin O. Kuhn Library.
Information Systems: Can you give a preview of your upcoming virtual presentation, “The GPT Dilemma: Navigating the Pros, Cons, and Impacts on Learning?”
Karen Chen:I will share an experiment conducted during the Fall 2023 semester within two data science courses: the undergraduate course IS 428 and the graduate course IS 733. In both classes, I not only permitted but encouraged the use of ChatGPT. This approach was designed to promote the productive and critical application of tools like GPT in learning and practicing data science. I revised the learning objectives and overhauled numerous class activities and assignments to align with this pedagogical strategy. This reworking allowed for the free (students were encouraged to use GPT as they saw fit) and transparent (students were instructed to openly disclose and accurately document their use of GPT) incorporation of GPT into the learning experience. I plan to share reflections from the students on their experiences with this approach, alongside insights gathered from my perspective as an instructor. These reflections are crucial for understanding the pedagogical impact and potential of integrating AI tools like GPT into data science education.
Furthermore, the presentation will highlight an interesting GPT-related study led by IS PhD student Maryam Alomair, co-advised by Dr. Shimei Pan and myself. This study evaluated various versions of GPT for their ability to solve data science problems. Using a specialized case-based learning tool, the Caselet—for which won the DARPA AI Tools for Adult learning competition—we examined GPT’s efficacy in generating high-quality, insightful, and engaging explanations.
Overall, this presentation will give my perspective on the evolving role of GPT in “doing data science” and “teaching and learning data science.” I hope to spark a broader conversation on how these advanced AI tools can augment our teaching and empower students to be more self-regulated learners, working productively with a strange and imperfect AI companion like GPT.
Information Systems: What are the most substantial pros and cons of GPT in education today?
Karen Chen: As far as data science as a discipline, GPT is changing how we practice data science, which has implications for what we teach about data science (i.e., learning objectives) and how we teach it (i.e., pedagogy). This requires a rethinking of data science education. One apparent trend is that GPT performs increasingly well when given specific, detailed instructions; in data science, this could involve writing the code to run a logistic regression, as observed by students experimenting with GPT in the classroom. However, to tackle real-world problems, data scientists must engage in higher-order tasks such as abstracting the model from messy real-world problems, decomposing the problems into sub-programs, debugging and refining the pipeline if it does not work as expected, and explaining the analysis product to a non-technical audience. All these require skills well beyond the so-called “calculator” level skills. I believe that GPT is pushing data science education towards these higher-order tasks at a much faster pace. As for using GPT in data science education, GPT offers opportunities for students to learn concepts, receive feedback on analysis, assist in debugging, and many other benefits. Students in my class experimented with many of those tasks with a little bit of guidance and inspiration here and there. However, GPT may also give students a false sense of learning; without robust self-regulation, GPT can become a crutch that deters learning rather than enhancing it. Because GPT can make mistakes, sometimes simply making things up, it should be treated as an imperfect learning companion. It requires students to be vigilant of this important fact, always applying a critical lens when using it. I have seen many students gain these important insights through many hours of wrestling with GPT and develop their own taste of the GPT pros and cons through concrete hands-on experience.
Information Systems: What are some important resources for those wanting to learn more about GPT in the classroom as a productive tool?
Karen Chen: The most effective resource is GPT itself. Take a task (such as a homework assignment for which you already know the solution), discuss it with GPT to explore alternative approaches, and critically assess the solution. Ask for explanations, brainstorm, summarize, quiz, or role-play in a job interview, and many more. In the early days, I often referred to Ethan Mollick’s blog https://www.oneusefulthing.org/ for inspiration. There are many more resources out there these days.
Information Systems: You recently presented at a Machine Learning Seminar for the Department of Mathematics called “Multimodal Machine Learning and Analytics for Characterizing Parent-Child Interactions: Applications in Education and Social Works.” How does this work relate to your research in GPT?
Karen Chen: This presentation focuses on work that, while not directly related to GPT, paints a picture of the vision of different kinds of AI assistants in the education and social work domain. It explores how AI/data-driven methodologies can be leveraged to uncover the nuances of human-parent interaction within the context of math coaching and attachment-based interventions. For both studies, we employ multimodal data gathered from camera sensors to quantify the fine-grained dynamics between parent and child in natural settings, such as the home environment. This analytical framework establishes the foundation for AI-supported coaching strategies, which I aim to further develop through interdisciplinary collaborations with those with expertise in human-centered computing, learning science, social work, and others.
Information Systems: You are also the director of the Informatics for Human Flourishing lab at UMBC. What type of research are you and your students pursuing in the lab and can you talk about the concept of “human flourishing?”
Karen Chen: Our lab focuses on developing and studying data-driven systems and technologies—broadly categorized under informatics—that promote human flourishing. Our primary interest lies in systems and technologies designed to gather, analyze, and model human data. The overarching goal of our research is to empower individuals to thrive and lead fulfilling lives in a sustainable manner. We believe that to be able to flourish is a fundamental human right for everyone. Human Flourishing, our “North Star,” is deeply rooted in ancient Greek thought. However, we adopt the modern framework outlined by VanderWeele in his 2017 paper “On the promotion of human flourishing”. This paper identifies five determinants of human flourishing (i.e. Happiness and Life Satisfaction, Physical and Mental Health, Meaning and Purpose, Character and Virtue, Close Social Relationships) and four pathways to promote it (i.e. Family, Work, Education, and Religious Community). This framework not only directs our work but also gives us a profound sense of purpose in our daily work.