2026 ASEE Annual Conference & Exposition

Following the emergence of ChatGPT in November of 2022, higher education has grappled with how to craft rules and policies surrounding permissible student generative artificial intelligence (GAI) use within their classes. Despite being three years on from the first public version of ChatGPT, higher education policies have largely remained high level, and GAI use has been left to the instructor to determine. GAI policies are not only how students gain awareness for what they can/cannot use GAI for within the class, but often impact how they see the ability to use GAI beyond the classroom as well. Differences in GAI policies between classes, and even between sections, also make it difficult for students to know when GAI use is permitted, the cases in which it is not permitted, and the wider nuances that led to the decision of why GAI can/cannot be used. This decentralized approach to GAI use policies raises the questions: what are the GAI policies that engineering faculty members have been adopting for their classrooms? What types of variation do we see now in GAI policies?

As part of a larger National Science Foundation CAREER grant, 151 higher education engineering faculty members and instructors at 15 universities have been interviewed about their mental models surrounding GAI and assessment in STEM higher education. Interviewees were asked to discuss both their current GAI policy for a class they were teaching and how these policies were created.

The policies were clustered using semi-supervised learning into three initial predefined clusters based on GAI restriction categories: no restriction, semi restricted, and fully restricted. “No restriction” policies are those in which GAI use is fully allowed on all class assignments and activities. “Semi-restricted” policies are those in which GAI use is allowed for some class assignments and activities, but GAI usage is prohibited for other assignments or activities. “Fully restricted” policies are those in which GAI use is not allowed for any class assignment or activity. Clustering was also performed using K-means clustering with three clusters set to evaluate for overlaps and differences between the two methods. Each cluster and cluster method was evaluated to identify common themes across the clustered groups.

From this clustering and analysis of GAI policies, groups of similar policies are created. Additionally, descriptions of why we think the K-mean clustering algorithm clustered the way it did are also created. This clustering and analysis helps instructors more directly compare their GAI use policy to those we collected. The dissemination of these findings, among other GAI resources, across institutions, will help to better inform instructors of the different ways to navigate student assessment in STEM higher education in the age of GAI, and promote dialog on different policy recommendations for GAI usage.