Teaching with Ocean Data in the Age of ChatGPT

How do you engage students in ocean science, data literacy, and programming skills–especially when your students have access to AI models like ChatGPT?

In this fast-changing world, each instructor has come up their own tricks. For two Data Labs members, Dr. Mikelle Nuwer and Atticus Carter of the University of Washington, it meant tapping into their students’ creativity. Along the way, they found this strategy helped unlock their own creativity, as they created new educational tools to engage their students in oceanography in new ways.

The challenge of AI in large, online courses

Dr. Mikelle Nuwer is a teaching professor at the University of Washington. She’s been teaching there for over 15 years and is primarily responsible for teaching large introductory undergraduate courses for the School of Oceanography.

A woman pointing at a decorated cupcake on a table in front of her. She is pointing out different features to an audience off-screen.

Nuwer showcases her creativity, using a cupcake decorating challenge to explain her approach to data literacy.

In one of her online courses, Nuwer noticed that many of her students were simply copying and pasting ChatGPT answers for their homework questions. She wondered, how do you combat machine learning in a learning management system you’ve always used with a course you’ve perfected through the years?

Nuwer decided to try a novel solution: You enlist the expertise of those around you to help create exercises that will get students to focus on the “enduring understandings” of the course.

Searching for expertise

Nuwer found this expertise among her teaching assistants (TAs). Her courses are fashioned to have graduate and undergraduate TAs working together to support student learning in an asynchronous online course. “It’s kind of this mentored ladder approach where there’s the faculty supervisor instructor, a graduate TA, and a team of undergraduate instructors,” explained Nuwer.

For one of Nuwer’s courses, her undergraduate TA was Atticus Carter. A current undergraduate at the University of Washington studying oceanography and educational studies, research, and policy, he spends a lot of his time designing educational materials for oceanography. Recently, he has been building a collection of computer vision tools to support oceanographic research.

While Nuwer has limited coding experience, she has extensive experience with data literacy pedagogy. So, she worked with Carter to create a scaffolded set of activities and questions focused on graphing ocean data. “Initially Atticus identified question types and strategies that other instructors were using to encourage students to engage with the material and think critically about a topic and then he adapted the assignments in the course to make the homework assignments more meaningful and less ChatGPT-able.”

Working around the problem of ChatGPT

Nuwer and Carter met weekly to figure out the best way to fashion the online homework needed for their students to engage with the real-time data problem sets they were being presented with and build in the good habits they needed to engage fully with their learning processes.

Two people sitting next to each other around the corner of a table. They are in casual discussion, both with their laptops open.

Carter and Nuwer, in discussion during the Princeton 2024 Design Workshop.

As Carter described it, they created scaffolded question sets that involved higher-level thinking —questions that required interpretation and comparison. The questions required students to refer to charts and graphs and not just respond to text-based prompts.

“It was basically a lot of higher-level thinking questions that AI couldn’t do yet. Like, ‘look at this graph and interpret two trends’ or ‘which of the following makes the most sense when you’re taking into account this graph.’” Carter elaborated that this was during a time when AI was not advanced enough to read pictures and graphs and make meaningful interpretations, which is an active area of AI development now (so, future adaptations may be necessary).

Together, they guided their students’ work to a creative final project–an infographic on what data means to them. “So even if they did use AI to get that info, at least they’re creating something,” Carter reasoned. “They’re still making more connections with it.”

In Carter and Nuwer’s view, it’s not a fight to stop AI. It’s a strategy to show students its inner workings–the good, the bad, and the ugly.