Grading is a critical aspect of higher education, connected closely with student learning, credentialing, and institutional accountability. More recently, the widespread use of generative AI (GenAI) among university students has introduced new layers of complexity to teachers’ grading practices. For example, do students who complete their work without GenAI assistance deserve a higher grade? How can teachers disentangle and properly weigh the respective contributions of GenAI versus the unique insights students bring to their work? If a student declares GenAI assistance in their work, does this make their work less original or less independent? These questions are now arguably central to every teacher’s grading decision but come with no straightforward answer.

One key reason for this is that the rise of GenAI has complicated and called into question many fundamental concepts underlying teachers’ grading decisions, including what signifies the quality of student work and their academic achievement. As teachers make their grading decisions, they are also at the same time making value judgements heavily mediated by their own assumptions about what is important, ethical, or desirable in the grading (Sun & Cheng, 2013). By teacher value judgement, it refers to any decision that reflects a teacher’s subjective prioritisation of values in the grading process. These values may be implicit and often not openly discussed, such as student effort and behaviour. Values may also be externally prescribed, such as those formalised in rubrics – however, making the value judgement would still involve teachers bringing their own interpretations to the rubric, which are influenced by their individual priorities and the practical constraints they experience. 

At present, however, little is known about how university teachers navigate and make these value judgements, how they balance various, sometimes even competing values, and what they prioritise in the grading of GenAI-assisted work. While much of the scholarly discussion in higher education has centred on academic misconduct, the reality faced by teachers is far more complex than making a binary decision on whether particular uses of GenAI are acceptable in student work. 

Therefore, this study addresses an important gap in higher education literature by investigating university teachers’ grading practices in a time when GenAI can mediate students’ work quality to varying degrees. We propose to address the following research question:

What value judgements do teachers make when determining the grades of students who (may) have used GenAI to assist in their work?

This study engaged 33 university teachers in scenario-based interviews to investigate their grading practices in a time when GenAI can mediate students’ work quality to varying degrees. The interview consisted of two parts and typically lasted between 45 to 60 minutes. The first part asked general questions such as the participants’ beliefs about GenAI and grading, the types of assignments and any GenAI-related policies in their courses. However, these questions, although necessary to set the scene, often result in general statements from teachers that contribute little to elucidate their value judgements. Therefore, the second part invited teachers to respond to several controversial grading scenarios related to the use of GenAI in student work. These grading scenarios were derived from recent research on the challenges around students’ GenAI use in assignments and informed by the format (e.g., multiple-choice questions) suggested by the two earlier research which used controversial grading scenarios to elicit teachers’ value judgements around grading (i.e., Brookhart, 1993; Sun & Cheng, 2013). The purpose is not to elicit “correct” answers from teachers or to quantify their answers, but to provide a grounding for teachers to articulate their judgement process in a more concrete, contextualised manner. We acknowledge that teachers’ decision-making captured in this study did not carry the same stakes as real-world grading. However, interviewing teachers about their real-world grading decisions may not effectively reveal the implicit assumptions teachers hold about student work and grading, as many are reluctant to open their grading to scrutiny due to fear of critique. By using these hypothetical scenarios, this research design creates a low-pressure environment that allows teachers to share their genuine thoughts.

Data show that teachers do make value judgements of student work, which extend beyond the assignment itself to encompass teachers’ conjecture about who the student is (person-oriented values), what they are capable of (capability-oriented values), how they relate to others (relation-oriented values), and whether the grading decision leads to good outcomes (justice-oriented values). Many non-academic values are prioritised in the grading (e.g., honesty, diligence, trust), and there are significant variations across teachers’ value judgements. The study points to a messy grading space full of tension and inconsistency. If left unaddressed, this will likely result in many unintended consequences, such as distrust from students, biased grading and weakened credibility of academic certifications. We foreground validity as an important concept (Dawson et al., 2024) to help teachers navigate this complex grading landscape and call for greater transparency about how students’ GenAI use will be factored into teachers’ grading decisions. The study seeks to move beyond the binary debate of whether GenAI use in student work constitutes cheating, towards a nuanced investigation into the subjectivities of grading in the age of GenAI.

Brookhart, S. M. (1993). Teachers' grading practices: Meaning and values. Journal of Educational Measurement, 30(2), 123-142. Dawson, P., Bearman, M., Dollinger, M., & Boud, D. (2024). Validity matters more than cheating. Assessment & Evaluation in Higher Education, 1-12. Sun, Y., & Cheng, L. (2013). Teachers’ grading practices: meaning and values assigned. Assessment in Education: Principles, Policy & Practice, 21(3), 326-343.