Summit Reflection: Generative AI in Duke Courses

In early October, Learning Innovation hosted the inaugural Emerging Pedagogies Summit for Duke faculty and staff invested in teaching and learning. The Summit’s goal was to promote ongoing conversations about the future of education in hopes of shaping a future vision of pedagogy. In this blog series, we will revisit each session to reflect on the lessons shared by the presenters and envision how these emerging pedagogies may take root at Duke.

There are no clear cut answers on how to engage with generative AI in education. However, between banning entirely and complete adoption there is hopefully a middle road. Finding sensible ways to integrate AI in teaching that furthers learning takes both purposeful planning and attention to the shortcomings of the content AI produces. One of the panels at the summit showcased how a diverse set of faculty around campus are experimenting with generative AI in their teaching, while also addressing the benefits of and problems with AI with students. This post outlines two themes that emerged from their experiences. The participants were:

  • Andrea Lane, Assistant Professor of the Practice, Social Science Research Center
  • Craig Hurwitz, Executive in Residence, Pratt School of Engineering
  • John Reifschneider, Executive in Residence, Pratt School of Engineering (Facilitator)
  • Mark Olson, Associate Professor of the Practice of Art, Art History & Visual Studies

Click the image below to watch the entire session:

Watch "Generative AI for Teaching and Learning at Duke"

Learning from Generative AI’s Limitations

Although generative AI can create human-like, natural language content, it does not always convey accurate information and fails to develop strong arguments. This means users of AI always need to double check sources and sometimes improve content, unless it is a relatively direct task. Two of the panelists discovered that the shortcomings of AI were surprisingly opportunities for learning.

For an exhibit at the Nasher Museum, Mark Olson and his students trained a ChatGPT bot to think like a museum curator (or attempted to). His team researched the technologies needed to create an AI bot that could process the metadata and images from the museum and then interact with museum staff to curate an exhibit. The bot made some curious choices of pieces based on the theme (dystopia/utopia) for the exhibit, which led the human curators to reflect on the types on descriptions and metadata that ChatGPT was trained on and wondered why the content wasn’t interpreted as they intended. Their experience with ChatGPT led to discussions about the quality of descriptions of art being written by humans and how to improve them.

In his FinTech course, Craig Hurwitz asked students to use ChatGPT to write a draft of an Executive Summary to pitch an idea to a venture capitalist. Students were directed to track their changes and answer reflection questions about the assignment. The majority of the students reported that they did not trust the content to be accurate and this pushed them to find alternate sources and/or write improved prompts to refine the output from generative AI. Interestingly, although they had to fix ChatGPT content, they still reported that the draft saved them time. This points to the ways in which generative AI (despite issues with its output) and students can combine forces to create a better final product. Craig wrote a blog post earlier this year about his experiment, if you are interested in additional details.

Generative AI as an Equalizer

One of the benefits of generative AI is that it has the ability to reduce barriers to learning for students. Generative AI can be a tutor for students who need explanations of concepts or other types of homework help. It can also serve as an alternate point of access for students who do not feel comfortable engaging with an instructor with questions. Two of the panelists spoke to this aspect of generative AI in their classes.

In the FinTech course, Hurwitz’s international students pointed to the ability to overcome a language hurdle with the help of ChatGPT that would normally slow down their work as a benefit. The use of AI to improve their writing allowed the international students to express their ideas in an equivalent manner to native speakers.

For her data science course, Andrea Lane used a course system that leverages AI to promote student learning (the product – Classwise – was developed by the facilitator of the panel, Jon Reifschneider). She trained the AI with her course materials, which allowed her students to quickly search through content in texts and videos while studying. There were strong pedagogical benefits related to assessments created by the AI. It could produce additional examples, extra problems, and comprehension checks for students. All students had an equal chance to level up on concepts they didn’t understand. This was especially important in Lane’s class because there were a wide-range of learners, from students who had never encountered data science concepts to advanced users.

Reifschneider echoed this benefit when he commented, “One of the great things about AI is that its strengths are largely complimentary to our own strengths as human educators. It’s always on. It never sleeps. It has infinite patience.” When an instructor is sound asleep and a student has a problem, AI may be able to help.

Further Resources

Interested in exploring how to use generative AI in your courses? Check out: