In response to ChatGPT, there is a lot of temptation to return to ‘old school’ assessments, such as oral exams, in-class essays, or pen and paper exams. These types of assessments would make academic dishonesty such as using ChatGPT very difficult, and so would also present the clearest proof that assessments were completed with academic integrity. Unfortunately, they also heavily penalize any student unable to attend or perform the skill in question at the appointed time. In order to avoid this exclusionary pitfall, I advocate for keeping learner-centred pedagogy and Universal Design for Learning (UDL) at the core of any changes you may make in response to ChatGPT. Ultimately, that will mean maintaining some space for choice and flexibility in your course design.
The learner-centred approach advocates that students should be actively involved in their learning. Time and space allocated for course work should prioritize student learning and student needs, as opposed to lecturer or teacher needs. That means that class time allocated for assessment should prioritize formative over summative assessment, and should be balanced by class time devoted to non-assessed, learner-centred activities. That may include flipped classroom techniques. It may mean a level of choice and flexibility in the assessment designs. Or, it may mean prioritizing activities that help learners to make their own meaning out of the materials of the course – perhaps through discussion, or free-writing, or other creative forms of response. That is, it is unlikely to mean frequent in-class testing.
In Universal Design for Learning (UDL), there is a recognition of the great diversity amongst students: diversity in needs, in capacities, in interests, in perspectives, and in abilities. UDL advocates anticipating as broad a range of learners as possible, so that by the time students sign up for a course, the course will already have been designed to meet their needs. UDL advocates designing a level of flexibility and choice into the course, materials, and assessments from the outset, so that students are empowered to adapt the course to their own needs. For example, in-class writing assignments could be an option and a way to fulfill a particular assessment requirement, but it would be better if one-time, in-class assessments were not the only way to fulfill that requirement.
In advocating for learner-centred course design and UDL, there is a broad category of response to ChatGPT and LLM that has me particularly worried: high-stakes, in-person, one time assessments. In a previous post, I highlighted the ways that prior research about academic dishonesty might give us insight into the ‘new’ world brought about by ChatGPT and other large language model AIs. In short, academic dishonesty is most likely to arise where there is time-pressure, grade-pressure, or peer pressure. High-stakes, one-time assessments increase those pressures rather than alleviating them. If ungrading is not an option (and it is not an option in many contexts), consider a variety of low-stakes and flexible assignments where the time, grade, and peer pressure are alleviated, and the temptation towards academic dishonesty is lower.
Against high-stakes all-or-nothing assessments
Many proposed responses to ChatGPT suggest replacing at-home, untimed assessments (such as essays and take-home tests) with in-class, time-limited assessments. Some proposed responses swap word-processed assignments for handwritten assignments. And some proposed responses replace written tests and assignments with oral presentations or exams. Each of these proposals take relatively flexible written assignment styles and replace them with fixed and unimodal assignment styles, and each of these responses narrows the range of students who can succeed at the assignment. In the process, this narrowing diverges from principles of Universal Design for Learning .
Of course, no one intends to exclude students with diagnosed accommodation needs. Indeed, students with diagnosed and documented accommodations are often legally protected, and will have relatively clear (though nonetheless onerous) procedures for accessing accommodations such as extensions, submission procedures, and alternative assignment formats. But these procedures can only be initiated after the course has been set in motion. They require individual exceptions to be carved out, one at a time, and place a burden on the student – not to mention the instructor who responds to them individually.
However, students with diagnosed accommodation needs are not the only learners whose needs are undermined by high-stakes, in person, one-time assessments. In general, UDL does not limit itself to responding to the diagnosed and delineated needs certified and circumscribed by accommodation gatekeepers, but rather anticipates a wide diversity of learning needs – documented, diagnosed, or otherwise – and designs the course to meet as many of them as possible.
Students with undocumented accommodation needs will have no prescribed procedures available to them, yet their learning needs can still be anticipated by adhering to principles of UDL. The student with a child home sick from daycare, or the student caring for a parent in hospital will have to explicitly disclose their circumstances in order to plead their case for accommodation, and they might still be refused. The student whose emergent medical situation is awaiting assessment by a specialist, or the student whose evolving mental health crisis is in flux, may not be in a position to explain or request the accommodation they need. Indeed, they may not know what they need until they try a few things. But in navigating a course with a level of choice or flexibility, they can still find their own path up to a point, and do so without having to rely on disclosures or instructor mercy.
The student with a broken arm might not have much difficulty pleading their case, but accommodating their in-class, handwritten assignment might nonetheless be onerous for both learner and instructor – requiring out of class dictation, testing centre bookings, or other time-heavy individual accommodations. But allowing flexible modes or occasions of submission for low-stakes assignments, and limiting (if not eliminating) the use of high-stakes, one-time, in-person assessments will help diverse students navigate the course according to their own learning needs.
Indeed, high-stakes, one-time, in-person assessment practices may place unbearable burdens on instructors as well, depending on class size, student population, and the availability of university support. Oral exams of 10 minutes per student would require about 7 hours of exam time for a 42 student class, assuming there were no scheduling issues or time overruns. In-class and timed assessments also presume that the instructor will never have an unexpected emergency, that the university will never have a snow day or tornado warning, and that the fire alarm will never go off during class time. Yet, all of those things and worse have been known to happen.
For these and other reasons, implementing high-stakes, in-person, one-time assessment strategies as a response to ChatGPT will exclude many students, place a burden on instructors, and place particular burdens on vulnerable students. To the extent that you can recognize ChatGPT and also maintain a flexible, student-centred learning environment, everyone involved in the course will benefit.
A learner-centred and non-punitive approach could acknowledge the existence of ChatGPT, recognize it as a temptation, and also offer learners the tools to help resist that temptation. But, in the very least, do not accept excluding students in the name of maintaining academic integrity.