Part 1: Shorter-term approaches

If redesigning your assessments is not feasible for you, you might want to consider shorter-term strategies to tweak existing assessments. It’s important to recognize that these are ‘short-term’ and that they may not be as effective or sustainable as AI capabilities improve (and they are improving at a fast pace!). We’ve included the following options that are more easily integrated into your existing assessments:

  • Invigilated / Observed In-class work
  • In-class work that integrates AI
  • Revising assessments to emphasize tasks AI cannot perform well
  • Revising grading schemes and rubrics

Be mindful of how an in-class assessment might present barriers for students – particularly those that may have academic accommodations in place. A well-structured, timed, writing exercise, for example, may cause significant concern for students who struggle with cognitive load or focus issues.[1]

Invigilated/observed in-class work

Traditionally, postsecondary has leaned heavily on invigilated tests and exams. A well-intended effort to move away from these to more authentic assessments has required much thought and labour on the part of the instructor, who perhaps now are grappling with how it may have unintentionally created new ways for students to use AI to complete the assignment. One option to circumvent a possible plagiarism risk is to introduce space during in-class time for assessment of learning. This may involve shifting some of the content delivery to an asynchronous environment (e.g., recorded lectures, assigned readings). Some ideas may include:

  • Real-time in-class discussions and reflections – an in-class group discussion, individual written reflection or oral response to a topic-based prompt demonstrates critical thinking and personal reflection.
  • Group case studies – tapping into the proven practice of Problem-Based Learning (PBL), present students with real-life scenarios or case studies and use in-class time to collectively discuss, apply knowledge and problem-solve to analyze, propose solutions and back-up choices. Even as an ‘out of class’ activity, this approach requires the human judgement and contextual understanding that makes it less susceptible to AI-shortcuts.
  • Presentations and debates – assigning topics or issues to research in class and present in the form of short presentations or paired/group debates can provide an opportunity to assess information literacy, argument structure and persuasive communication skills in addition to knowledge of the topic in a way that circumvents tasks that might be more easily automated by AI. Including a Q and A as part of a presentation also invites an opportunity for dialogue and engagement on what the student has learned.

In-class work that integrates AI

Integrating AI into low-stakes in class assessments will help communicate to your students that you recognize how AI can be used, and at the same time foster a better understanding for your students around how it can be used intentionally, ethically, and in support of your teaching goals. Here are just a few ideas for in-class activities:

  • Instead of starting the class with a question on a key course topic, include the AI-generated answer and invite the class to critique and revise – either independently or in small groups.
  • Hold a ‘humans vs AI’ debate where students pit their answers to topic prompts against those generated by AI. This discussion will help students organize their arguments’ points of view and also discover new perspectives and strengthen critical analysis skills.
  • Divide the class into three groups and have each group evaluate the AI-generated output to a course topic prompt for either factual accuracy, artificial empathy or bias.

Emphasizing tasks that AI cannot perform well

In the same vein that educators try to design authentic assessments that are valuable learning opportunities with a side benefit of being not easily plagiarized through the affordances of the pre-AI Internet, we have focused on trying to come up with alternative assessments that ChatGPT cannot easily perform. This is risky, as the technology is evolving at a quick pace, with massive amounts of prompts being continually added, and thereby improving the outputs being generated. However, while AI-proofing assessments may be nearly impossible, we can choose to focus on tasks in assessment expectations that encourage personalized and localized connections. An example of this might be to demonstrate the learning through links to local context, current events which may not be well represented in LLMs.

Revising grading schemes and rubrics

You may want to revise your current grading scheme, rubrics and the criteria outlined to reweight and emphasize the less-mechanical (aka easily AI created) competencies. For example, with a writing assignment you may choose to focus more on skills such as creating a good argument, including and evaluating evidence and critical analysis rather than grammar, and essay structure. You may also require that rough planning notes be included as part of their assignment submission to indicate assessment of the process rather than the finished product.


Icon for the Creative Commons Attribution 4.0 International License

Generative Artificial Intelligence in Teaching and Learning at McMaster University Copyright © 2023 by Paul R MacPherson Institute for Leadership, Innovation and Excellence in Teaching is licensed under a Creative Commons Attribution 4.0 International License, except where otherwise noted.

Share This Book