Assessing Generative Artificial Intelligence Resilience in assessments

Critically evaluating existing assessments to determine resilience and rethink approach in a context of Generative Artificial Intelligence (GenAI).

Here are five strategies for critically appraising how resilient your assessments are in a generative Artificial Intelligence (GenAI) context and adapting your approach.

Strategy 1

Identify assessment redesign

Engage with the Rethinking assessment strategies in the age of artificial intelligence (AI) resource from Charles Sturt University to identify the considerations you need to make in assessment redesign, including designing for or designing to limit GenAI usage.

Strategy 2

Considerations for common assessment

Engage with this new Assessment and GenAI resource from Charles Sturt University and/or the University of Wollongong resource which highlight how to improve common types of assessment (e.g., quizzes, essays, reflections, presentations and exams) in light of greater access to GenAI.

Strategy 3

Diagnose AI resilience

Use this AI resilience diagnostic from the University of Technology Sydney, which steps you through specific aspects of assessment design which make assessments vulnerable to generative AI use.

Strategy 4

Critically appraise your assessment

Critically appraise your assessment using the Learning Outcomes – Tasks – Marking heuristic from Monash University to decide if you need to design GenAI in or out:

  • Learning outcomes (LOs) refer to the things you want to claim have been learned by students e.g., knowledge, skills, attitudes, competencies, capabilities. LOs a for a specific lesson/teaching interaction will need to be constructively aligned to subject and course level LOs. Questions to ask about your LOs include:
    • does a student need to be able to demonstrate these outcomes without GenAI assistance? If yes, consider designing AI out of assessment.
    • are these LOs difficult to automate? If yes, then consider designing AI out of assessment.
    • who or what am I attributing the learning to (i.e., the student alone, a group of students, or student-AI tool interaction)? If you cannot be sure that students have met the LOs on their own without using GenAI, consider designing AI in.
  • Tasks refer to the environment that students work in to produce evidence of learning. Questions include:
    • can GenAI be easily accessed under assessment task conditions? If your task conditions (unsupervised, unsecure) make it feasible to access gen AI, consider designing it in. If your task conditions make is unfeasible to access gen AI (secure and supervised), then design it out.
    • can I determine who or what produced the work products associated with these tasks? If your task conditions do not make it possible to determine who produced what, then, consider designing AI in.
  • Marking refers to the processes and tools used to validate learning and achievement of learning outcomes. The main question to be asked is:
    • Do my marking procedures mean that I am at risk of ignoring interactions between students and GenAI and making incorrect attributions of who has done the learning? If yes, consider designing AI in.
Strategy 5

Test your assessment/s.

Use a generative AI tool such as Bing Copilot to test your assessment/s.

  1. Visit Bing Copilot and sign in using your Charles Sturt staff account. On the homepage, choose “sign in with a work or school account” to access the protected version (this is the default setting for Charles Sturt).
  2. Prompt Copilot to “do” your assessment.
  3. Mark the generated response/output using your rubric.
  4. Provide the gen AI tool with feedback on the response/output and incorporate the adjustments gen AI makes to enhance the resilience of your assessment.

Resources: