The State of AI in Teaching – Challenges and Solutions

This newsletter is based on a series of workshops on "AI, Examinations and Teaching" held for a number of departments at SDU throughout 2026.

It is evident that AI is a transformative force that is in the process of changing the professions, research, the content of programmes and disciplines, teaching, supervision and examinations, students’ use of AI, and teachers’ use of AI in planning – and preparing for – their teaching. Any notion of ‘containing’ AI or ‘keeping AI out’ is unrealistic.

It is equally evident that these changes differ considerably from programme to programme, from profession to profession, and from one field of research to another.

Furthermore, it is clear that AI in the hands of academically trained researchers, i.e. researchers who (i) have encountered the resistance of the subject matter – have toiled through their reading, spent time taking notes and condensing material, struggled to write and revise texts, translated between languages, submitted texts to peer review (again and again), and given feedback and peer-reviewed countless times; (ii) have built up in-depth specialist knowledge in a field; (iii) are able to put the right, academically incisive and relevant questions to chatbots (i.e. to prompt); and (iv) are able to assess the academic soundness and quality of answers and sources – can increase the productivity and quality of their own research.

The major pedagogical challenge for us in our role as teachers, however, is how students who have not yet acquired specialised knowledge and academic analytical skills can develop these with AI at hand (only a few keystrokes away) and as a constant temptation. How can we design their programmes so that they encounter sufficient resistance in the subject matter to acquire academic skills? And sufficient academic skills to become knowledgeable and critical users of AI?

This is a formidable challenge, and we are still far from having all the answers. The good news is that we are not alone in facing these challenges. The same discussions are taking place at universities around the world, and research has begun to provide useful answers. For this reason, the newsletter also refers to a small selection of articles and online resources where further inspiration may be found.

Below are some reflections, suggestions and sources of inspiration.

1. AI separates text from thinking – the validity of examinations is under pressure

A text written by a student can no longer simply be taken as an expression of what the student thinks, understands and is able to argue. As a result, a fundamental premise of many teaching activities and forms of assessment – in particular graded take-home assignments, bachelor’s projects and master’s theses – has come under pressure.

At the workshops held in various departments, a recurring element has been to ask a powerful chatbot with a Deep Research function to complete examination assignments at both bachelor’s and master’s level. As a rule, the chatbots perform rather well within 15 minutes – without anyone having to read or write anything, or indeed learn anything.

Assignments that measure the ability to retrieve, structure and present information are exceedingly easy for chatbots (these three skills have become ‘cheap’). Competences such as local interpretation, theoretical decision-making and documented source criticism are ‘more expensive’ competences, which chatbots find it more difficult to emulate.

We therefore propose the following actions:

• Stress-test your own examination papers. Ask an advanced chatbot to answer the most recent assignment, and discuss within your subject group what grade the answer would receive.
• Introduce oral defences or random oral verification of written take-home assignments. This tests whether the student has ownership of the text.
• Map the assessment profile of your programme. Which examinations are vulnerable, and is there reasonable variation and progression in the forms of assessment?

Further reading:

• Dawson, P., Bearman, M., Dollinger, M. & Boud, D. (2024). Validity matters more than cheating. Assessment & Evaluation in Higher Education, 49(7), 1005–1016.
• Corbin, T., Bearman, M., Boud, D. & Dawson, P. (2026). The wicked problem of AI and assessment. Assessment & Evaluation in Higher Education.
• Online resource: TEQSA (2025), Enacting assessment reform in a time of artificial intelligence.

2. Cognitive offloading and the "performance paradox" – better products, less learning

Perhaps the most serious challenge posed by AI is – as described above – that it makes it easy to let the machine perform precisely the cognitive work that learning requires. José Bowen compiles the research on We Teach with AI, and the message is clear: when students’ use of AI is unguided, their take-home assignments improve, but their examination results decline. Moreover, students overestimate how much they have learned. Leslie Loble and Jason Lodge refer to this phenomenon as a "performance paradox": AI provides a short-term boost in performance, but students do not retain the material when they are tested later (EducationHQ).

Suggested actions:

• Distinguish explicitly between beneficial and harmful offloading. Tell students what they may delegate to AI (e.g. language proofreading) and what they must think through themselves, because that is where learning takes place.
• Use purpose-built AI tools that withhold answers, e.g. a course chatbot that provides hints rather than solutions.
• Build independent self-testing into your teaching. Short quizzes and oral checks show students what they are able to do on their own, without the machine.

Further reading:

• Bastani, H. et al. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. PNAS, 122(26).
• Fan, Y. et al. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. British Journal of Educational Technology.
• Online resource: José Antonio Bowen, Design Principles for Teaching in the Age of AI.

3. Students’ declarations of AI use do not work as intended

Most universities, including SDU, require students to declare their use of AI. Teachers consistently report that these declarations are imprecise and often uninformative. Most teachers also report that they do not address the use of AI with students in their courses or teaching. Students are uncertain about their teachers’ positions and still do not know whether the use of AI will count for or against them in the assessment, and they presumably under-declare. Teachers develop "police eyes", looking for cheating rather than assessing the academic contribution – which often takes its toll on job satisfaction.

The major problem with the current declaration rules is that they are unrealistic, for the following reasons:

1. The process of scientific analysis is complex – and students have only just begun to learn about the analytical steps involved in a scientific analysis.
   
   
2. AI tools and the use of AI in scientific/academic analysis are complex – and students lack the necessary competences for this – so we need to teach students to use AI tools for scientific/academic purposes through a step-by-step approach.
   
   
3. Transparency about AI is fundamentally not about rules and compliance with guidelines – but about restoring the connection between text and thinking – so we need to set new requirements for written assignments.
   
   
4. It is no longer self-evident to students why and how they should engage with the subject matter, including in connection with take-home assignments – so we must take particular care to create motivation for the work.

Declaration is not merely a matter of complying with rules; it is ultimately a matter of accountability. Declaration means that one affirms – and signs one’s name to the fact – that one is able to stand behind one’s work intellectually and professionally, i.e. that one can explain, reproduce, evaluate and defend one’s work in an academic conversation. Rules that leave the assignment itself unchanged depend on voluntary compliance, which cannot be verified (Corbin, Dawson & Liu 2025).

Suggested actions:

• Expand the AI declaration into a "methods section on technology". It is not sufficient to write "I used ChatGPT". Students should describe how they used AI in each step of the analysis and what judgements they made, e.g.: "I asked the AI to summarise X but judged that it had overlooked Y, so I supplemented it with Z myself."
• Require process traces for larger assignments. For bachelor’s projects and master’s theses, students should be able to present saved chat logs, logbooks or versioned documents if the assessors are in doubt.
• Assess both product and process, e.g. with a 70/30 weighting or with pass/fail on process elements.

Further reading:

• Walton, J., Bearman, M., Crawford, N., Tai, J. & Boud, D. (2026). How university students work on assessment tasks with generative artificial intelligence: matters of judgement. Assessment & Evaluation in Higher Education.
• Corbin, T., Dawson, P. & 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.
• Online resource: SDU UP, Newsletter June 2025: Generative AI and Examinations (in Danish).

4. Unclear expectations – teachers should provide AI declarations for their courses

Declaration is not the responsibility of students alone. The role that AI is to play in a subject is now an unavoidable didactic decision for the teacher. The quick fix – a "two-lane road", in which AI is either completely prohibited in supervised examinations or entirely unrestricted in all others – is, according to Guy Curtis (2025), untenable. Good teaching builds up skills step by step – much like learning to drive a car.

"Use AI responsibly" gives students nothing to act upon. An explicit formulation links the permission to the learning objective: "You may use AI for language feedback on text you have written yourself. You may not use AI to interpret data, because that is precisely what the assignment assesses."

Suggested actions:

• Provide AI declarations for your courses based on the AI Assessment Scale, so that the course description makes clear what role AI plays in teaching activities, learning objectives and examinations.
• Use a simple translation model: learning objective → student activity → role of AI → student’s responsibility → explicit expectation. Write two or three sentences to students about what is encouraged, permitted and not permitted – and why.
• Ensure alignment. If the assignment can be completed with extensive use of AI, AI competences should also be included in the learning objectives and in the grading.

Further reading:

• Perkins, M., Roe, J. & Furze, L. (2025). Reimagining the Artificial Intelligence Assessment Scale: A refined framework for educational assessment. Journal of University Teaching & Learning Practice, 22(7).
• Curtis, G. J. (2025). The two-lane road to hell is paved with good intentions: why an all-or-none approach to generative AI, integrity, and assessment is insupportable. Higher Education Research & Development, 44(8).
• Online resource: The AI Assessment Scale (AIAS).

5. How do students learn to write – and to have a voice of their own?

In the future, all writing will be "blended writing", as José Bowen points out. Yet learning to write remains important. Writing is a craft, a window into thinking, and a learning process in its own right. When a chatbot delivers a polished text in seconds, the student’s academic voice is at risk of disappearing.

For many years, we have trained students to avoid writing in the first person. Perhaps we should now move in the opposite direction.

Suggested actions:

• Require first-person statements that make the student visible as a decision-maker: "Drawing on inspiration from my teacher, two texts from the syllabus and four suggestions from my chatbot, I chose this research problem because …", "I have chosen to take theory 1 and theory 2 as my point of departure because …".
• Divide writing assignments into phases (problem formulation, research, outline, draft, revision, reflection), and provide feedback on the process along the way.
• Design new types of assignments that are local, personal, topical or interview-based, or that require students to take a position on ethical dilemmas and justify personal choices. Alternatively, let AI write a draft that students then critique, improve and explain their own contribution to.

Further reading:

• Barrett, A. & Pack, A. (2023). Not quite eye to A.I.: student and teacher perspectives on the use of generative artificial intelligence in the writing process. International Journal of Educational Technology in Higher Education, 20.
• Tai, J., Ajjawi, R., Boud, D., Dawson, P. & Panadero, E. (2018). Developing evaluative judgement: enabling students to make decisions about the quality of work. Higher Education, 76(3).
• Online resource: Teach GenAI (AAU, AU and SDU), Learn GenAI – How to use generative AI in your studies.

6. Motivation and the teaching community – why "bother" with the subject matter?

When a chatbot can solve the exercise, and the temptation to let it do so is ever-present, we as teachers must take particular care to motivate the hard work. José Bowen proposes that – in order to support this care and transparency – individual programmes introduce shared templates for written rationales for teaching activities and exercises, and that, as a result of this motivation, students should always be able to answer yes to: I care (does the assignment make sense?), I can (can I manage it?) and I matter (does my contribution matter – is anyone listening?).

AI can enrich – and undermine – all learning activities in teaching. For example, in group work, AI can be used to stimulate and enrich the group’s discussions by providing additional information, clarifying doubts and giving feedback. But AI can also stifle group discussions if it is used as an answer key that shuts down all debate. Group work, too, must be motivated, explained and instructed anew.

Suggested actions:

• Use an assignment template that motivates: what are students to learn, why is it valuable, and why is the effort necessary?
• Try out course-specific tutor bots, and demonstrate them in the first teaching session.
• "Use AI in your preparation, use your brain in the classroom", as Harvard Business School puts it. Spend the time together on analysis, critique and taking a position.

Further reading:

• José Bowen and C. Edward Watson, “Designing Assignments and Assessments for Human Effort”, chapter 12 in Teaching with AI: A Practical Guide to a New Era of Human Learning, Baltimore: Johns Hopkins University Press, 2026, pp. 231–245.
• Kestin, G., Miller, K., Klales, A., Milbourne, T. & Ponti, G. (2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15.
• Online resource: Harvard Business School, Teaching in the Age of AI.

Back to basics

Although AI is a powerful transformative force in the education sector and in university teaching, a number of things remain unchanged, including:

1. Good and motivating teaching is still characterised by strong alignment between learning objectives, examination requirements and formats, and learning activities.
   
   
2. It is still those who put in the work who learn something.
   
   
3. Scientific analysis – and the competent handling of knowledge – rests on having a firm grasp of the source(s) of one’s knowledge and an assessment of the nature and truth value of those sources. Or, as the core of research-based teaching has been defined:

”Every science distinguishes between its subject matter and its method, regardless of the close connection between them … But it must be remembered that method is so essential to all science that the treatment of the subject matter may with good reason be described as an ongoing methodological discussion …” (Erik Rasmussen, Komparativ Politik 1 [Comparative Politics 1], Copenhagen: Gyldendal, 1971, p. 11; our translation).

In many respects, the answer to the major challenge posed by AI is likewise: back to basics.