Go deeper · from Step 5 · about 15 minutes
What AI means for the grade
In Step 5 you saw that universities are rewriting assessment for AI while the grade at the end stays almost untouched, and you met five questions for rethinking it. This page follows Juuso Henrik Nieminen and colleagues further, into why grades stay put and what the questions could mean for your own course.
The summary on this page is ours. Quotations are in the article’s own words. Read the article in Assessment & Evaluation in Higher Education.
§01 Grades as relations
What a grade does
Technically, a grade is a brief label for what a student achieved in a course, usually on a categorical and hierarchical scale such as A to F or pass/fail, and it ends up on transcripts and diplomas. Nieminen and colleagues find that definition too thin to explain what grades do, so they take a relational view. A grade gets its meaning and function from the ways it is put to use, and those uses reach far beyond your course.
The student and themselves
Students learn to see themselves through their grades, as high achievers, average students or failures, often measured against what counts as normal.
Student and teacher
The relationship between them is shaped by the plain fact that one of them grades the other.
The university and the world
Grades select candidates for exchange programmes and rank job applicants. For public universities they are also part of the social contract with the state.
That last link is sensitive now that people ask whether universities can still assure what their graduates have learned. Generative AI enters the same networks, and the authors ask whether it will do what so many technologies in higher education have done before and leave the relations as they were.
§02 Two responses
AI-powered and AI-proof grading
Research on AI and grading is mostly about marking single tasks and says little about the course grade on the transcript. The article separates two approaches at that task level and asks whether either of them disturbs the grade itself.
AI-powered grading
AI automates parts of the marking while the teacher stays in the loop. The promises are faster marking and less human bias, but the aim is to imitate human judgement as closely as possible. That may settle grading as a routine task, further from pedagogy, and tie it more firmly to the idea of objective measurement.
AI-proof grading
Graded tasks are shielded from AI, for example by invigilated pen-and-paper exams or by lab work done in front of an assessor. The purpose is to protect the values and practices that grades rest on. There are good reasons for that, not least the university’s social mandate, but the grade itself stays as it was.
AI-proofing could still unsettle grades if it secured a different kind of record, such as a portfolio with hurdle tasks (parts a student must pass) that assure learning without AI. Nothing like this has been put into practice at scale yet.
Our addition. The assessment twins in Step 4 show the same pattern on a small scale. Roe, Perkins and Giray (2026) let the twin confirm the take-home task or set a threshold for it, and in their worked example the outcome is still one mark.
If you are thinking of marking with AI
Where policies mention grading at all, it is mostly to regulate teachers who use AI to mark. In Europe the GDPR requires a human to be accountable for grading. The University of New South Wales generally bars its markers from using AI platforms to mark or grade, and the University of Newcastle requires that students are told clearly when AI is used in marking and lets them opt out.
Our addition. Before you use AI in your own marking, check your institution’s rules on AI and on data protection.
§03 Why grades stay put
What holds grades in place
Why does the grade stay the same when so much around it changes? Because, the article answers, grading systems make many other systems work.
A measurement legacy
The belief that assessment should be rational, quantitative measurement runs deep in universities. It yields comparable data for selection, however thin.
Numbers that travel
Grade point averages move easily to scholarship, exchange and doctoral applications in other institutions and countries. Numbers can be compared whether or not the comparison makes sense.
Roles
Grading has long set the social order between teachers and students, so a new system would ask both to redefine their roles.
Technology
Higher education has often taken up learning technologies that reinforce established practice. Generative AI may turn out to be one more.
Or is it the grade that holds higher education in place?
Nieminen and colleagues also turn the question round. Grades are an infrastructure for ranking and selection, and they stabilise the sector in turbulent times. How else would universities pick students for scholarships and honours? The meaning can also outlast a change of practice. Pass/fail grading does not necessarily end competition between students, and it may produce new forms of it.
Grades can be seen as the bedrock of education
McCloskey’s (2014) typology of educational rituals helps explain why grades sit so deep. Students and teachers perform grading visibly and again and again. Grades work as symbols, a language for who students are (an A+ student, a GPA of 3.2). And they are built into the academic year through exam periods and grade release dates.
Critiques of the capitalist and neoliberal logic of grading are familiar, and the authors think AI is a reason to revisit them. Generative AI has been said to favour productivity and clarity at the expense of depth and uncertainty, and grading already prizes productivity, objectivity and clarity. Together, they suggest, the two could move the sector a step towards the dystopia where students use AI to write papers and professors use AI to grade them.
§04 The five questions
Five questions, taken further
Step 5 gave the five questions in a line each. The article fills them out, and the detail is what makes them usable with colleagues. They are meant to challenge technological solutionism. Simply removing grades won’t do unless it changes the networks grades belong to, and while a university can’t control where achievement data goes, it does decide how achievement is recognised and represented.
To develop new solutions, the sector needs new problematisations.
For what, and for whom?
Sorting and selection have been the main reason for grading. Should universities do that work, and how much of it? The authors tie the question to three missions, preparing students for work, bringing them into relationship with bodies of knowledge and developing them as people. Which of these needs data on what students can do, and should that data be public or private?
What should be recorded?
Should a transcript say something about generic skills such as AI literacy? Institutions should also discuss what to leave off, such as internships and placements they are poorly placed to validate, and tie achievement data more closely to graduate outcomes.
What can different data do?
Grades are imprecise, decontextualised, monomodal and homogeneous. Trading numbers for narratives is not enough on its own. Ask what each kind of data lets people do, and keep inclusion at the centre so that Indigenous and disabled knowledges are not pushed aside.
Who constructs the data?
Grading is often treated as a burden and handed to others. Could it become a meaningful academic task that affirms the identities of both student and teacher? Students deserve a part too, through contract grading and their own evaluative judgement (Ajjawi and Boud, 2024).
The questions reach the programme level too. The authors point to the programme-wide reform that TEQSA encourages in Australia, which you met in Step 5, and ask whether traditional grading can sit meaningfully alongside it.
§05 Hybridity
With AI, without AI, and the record
The fifth question comes closest to your own course. The line between what people and machines can do keeps blurring, yet grading has barely taken part in that conversation. Current grades can’t show what a student can do with generative AI or without it, and they give students no way to present who they are and what they can do.
Students have always been graded together with their tools, the authors note, whether the tool is AI, a sewing machine, a laptop, the internet or a pencil. The metadata you met in Step 5 follows from that view. It treats AI as one of the tools and materials in an assessment situation, to be described along with the rest.
Our addition. Metadata about AI use can only report what someone was in a position to observe. As Step 1 showed, a rule about AI use can’t be enforced where nobody watches (Corbin, Dawson and Liu, 2025). “No AI permitted” on an unsupervised task records the rule and says nothing about what happened. An honest label describes the conditions, such as whether the work was supervised and which tools were available.
Our addition. Students’ use of AI is hybrid. They work alone at some points, let AI help at others and hand some tasks over, in no fixed order. A record with two sides, what the student can do without AI and what they can do with it, would describe that better than one grade can, and neither side would be the lesser achievement. The label activity in Step 5 is a first draft of such a record.
§06 Read it for what it is
A conceptual argument and a narrow review
The article is conceptual. Its policy review covers only public documents from three well-resourced systems in March 2026, and two of the Hong Kong universities kept their AI policies internal. It can’t capture change at faculty, programme or classroom level, and the authors expect larger shifts to show only in the years to come.
Our addition. Where the grading scale is fixed nationally, your course has little room to change the scale itself. The questions still apply to what your grade says and to the information that goes with it.
Why read it anyway? It moves the debate about AI from the task to the record. For the critical debate around it, Nieminen (2025) asked students in Finland and Hong Kong to imagine university without grades. Nieminen, Bearman, Ajjawi and Boud (2026) argue that criticism of grades, which goes back at least a century, has run out of steam, and they call for post-critical approaches. Ajjawi and Boud (2024) make the case for new ways of representing achievement.
Your own assessment
Choose one course you teach.
- Have you changed the assessment because of AI? If so, have you also changed how the grade is set, or what it says?
- What does the course grade tell a reader about what the student can do with AI and without it?
- Who uses the grades from your course, and what do they actually need to know?
Sources
- Ajjawi, R. and Boud, D. (2024). Changing representations of student achievement: the need for innovation. Innovations in Education and Teaching International, 61(3), 597–607. doi.org/10.1080/14703297.2023.2192513
- Corbin, T., Dawson, P. and Liu, D. (2025). Talk is cheap: why structural assessment changes are needed for a time of GenAI. Assessment & Evaluation in Higher Education, 50(7), 1087–1097. doi.org/10.1080/02602938.2025.2503964
- McCloskey, A. (2014). The promise of ritual: a lens for understanding persistent practices in mathematics classrooms. Educational Studies in Mathematics, 86(1), 19–38. doi.org/10.1007/s10649-013-9520-4
- Nieminen, J. H. (2025). What if there were no grades? Educational Assessment, Evaluation and Accountability, 38(3), 293–324. doi.org/10.1007/s11092-025-09468-7
- Nieminen, J. H., Bearman, M., Ajjawi, R. and Boud, D. (2026). Beyond a critique of grades. Higher Education. doi.org/10.1007/s10734-026-01627-z
- Nieminen, J. H., Bearman, M., Boud, D., Corbin, T., Dawson, P., Tai, J. and Walton, J. (2026). Grading in an age of assessment reform: the elephant in the room. Assessment & Evaluation in Higher Education. Open access (CC BY-NC-ND 4.0). doi.org/10.1080/02602938.2026.2694451
- Roe, J., Perkins, M. and Giray, L. (2026). Assessment twins: an approach for strengthening assessment validity in the age of generative AI. Journal of Applied Learning & Teaching, 9(2). doi.org/10.37074/jalt.2026.9.2.3