AI and Assessment
On This Page
Evolving your Assessments
Artificial intelligence is changing how U of A students access, create and apply knowledge — and challenging some of our traditional approaches to assessment.
The challenge with AI and assessment goes deeper than preventing students from using AI to shortcut their coursework. Although you could move your assessment to a supervised setting to ensure students are not using AI, this approach comes with significant trade-offs. Time limits restrict what students are able to produce, and this constraint disproportionately impacts students with accessibility needs.
A more complete path forward is to evolve your assessments. Instead of focusing solely on knowledge recall, consider adding authentic or competency-based tasks that mirror real-world challenges students will face in their fields and careers. The Centre for Innovative Teaching and Learning at Indiana University Bloomington highlights some key benefits of authentic assessments, including:
- providing a more accurate measure of student learning, especially for higher-order thinking skills.
- a more engaging and motivating for students because they involve real-world tasks.
- offering specific, actionable feedback on what students have (and have not) mastered.
Keep in mind that AI has become so adept at mimicking human work that few assessment strategies can definitively prevent students from misusing it. The strategies we share here make it less convenient or advantageous for students to take that route. More importantly, these approaches provide your students with opportunities to demonstrate what they have learned in their own authentic voices.
Levels of GenAI use
There will be times in your course when you should limit the use of GenAI because it might allow students to offload the specific skills you want them to develop or demonstrate. On the other hand, there will also be times when using GenAI is appropriate, helpful, and even necessary for learning.
As with any part of your course design, your decision whether to allow students to use AI should be based on the learning outcome for an activity or assessment. If your learning outcomes vary across your activities and assessments, your instructions about using GenAI should vary too.
See the discussion under Setting Expectations for advice on making your GenAI permissions clear and grounded in course learning outcomes.
The Artificial Intelligence Assessment Scale (Perkins, Furze, Roe and MacVaugh, 2024) suggests five levels of AI use that can help you narrow down how you want (or don't want) students using AI for an assessed task.
At Level 1, students complete the assessment without any AI assistance. They must rely solely on their knowledge, understanding and skills.
Examples: In-class exams, supervised writing tasks, oral presentations, clinical assessments, laboratory practicals, performance demonstrations, fieldwork documentation, and reading annotation.
Level 1 assessment activities should be proctored or used only for low-stakes formative assessment because unauthorized AI use may be undetectable.
At Level 2, students are permitted to use AI for brainstorming, organizing their ideas, and research assistance. However, the final product must not include any AI-generated content.
Examples: Research design planning, structured brainstorming, argument mapping, literature review preparation, project scoping, thesis development.
At Level 3, students are permitted to use AI to refine their original work at grammatical, vocabulary, or structural levels.
Examples: Essays, peer review feedback, multilingual compositions and translations, technical or analytical reports, creative writing.
If you consider it important, you can ask students to submit their original work for comparison with the AI-polished version, or request access to a Google doc where you can track changes. Ultimately, most current word processors include Level 3 assistance, making it unproductive to tease out in the writing process.
At Level 4, students are directed to interact or generate content with AI, and then to evaluate or respond to the AI output. Level 4 assessments are useful when AI enables you evaluate learning in a simulated environment, or when critical evaluation of AI outputs is a learning outcome.
Examples: AI-enabled case studies, simulations, AI literacy-building exercises, AI bias detection, and Socratic AI interaction.
Note that the U of A requires instructors to offer a no-AI alternative to students who object to using AI.
At Level 5, students are permitted to use AI at their discretion. This level is appropriate for assessment situations where learning outcomes can be measured regardless of AI use, or where AI use is a presumed aspect of the outcomes.
Examples: Creative work, AI-assisted coding, media production, data analysis, and other AI-enabled workflows.
Making learning visible
Since AI is now widely available, looking only at a student’s final submission might not give you a clear picture of what they actually learned. To get a better sense of their progress, you might want students to document how they developed their work. Whether you need this extra evidence — and what it should look like — depends on your learning outcomes.
Process documents
If your learning outcomes include processes like research, experimentation, planning, consultation, or revision, you need a way to measure those skills. You can build this into your assessment by requiring process documents such as proposals, annotated research summaries, logs, drafts, and staged self-assessments.
Disclosure and reflection statements
Prompting students to document and reflect on their AI use fosters healthy skepticism and proper attribution. However, instructors should not treat disclosure statements as an assurance of academic integrity. As Corbin, Dawson, and Liu argue in their 2025 article, "Talk is Cheap: Why structural changes are needed for a time of GenAI," disclosures alone do not secure an assessment; structural changes to assessment design are essential.
This Responsible use and reflection log is designed to help students think critically about the quality of AI output and its impacts upon their work and thinking.
Scaffolding
Breaking your assessment into stages or checkpoints can be an effective way to support students and make their learning visible. The goal is not to create more work for yourself or your students, but rather to provide strategic opportunities for you to gather meaningful evidence and provide meaningful feedback.
- Make your checkpoints purposeful: Break your assessment into a small number of stages that align with the outcomes you are measuring. Don't add steps to the assessment solely to keep students on track.
- Not everything needs to be graded: If your rationale for scaffolding is to measure a learning outcome at strategic moments in your student's process, it might make sense to have each stage in the work contribute to the final grade. But if a checkpoint is primarily an opportunity to provide feedback, you might consider making it required but ungraded.
- Use feedback strategically: It may not be necessary for you, personally, to give formal feedback at every checkpoint. Peer feedback, self-assessment, one-on-one or group conversations with you or a TA... all of these can be effective ways to support your students' progress.
- Vary AI permissions across stages: Scaffolding affords you the opportunity to define different AI permission levels at different stages in your assessment. At Level 2, for instance, AI is only permitted at the ideation stage. Level 3 is the opposite, permitting AI only at the polishing stage.
Self-assessment
When we talk about self-assessment, we mean more than simply asking students to grade their own effort. We are talking about offering students ways to explain their approach, justify their reasoning and evaluate the outcome. These reflections — alongside the final product of the assessment — enable you, as an instructor, to get a better-informed measure of your students' progress.
It is a happy coincidence that self-assessment also makes AI misuse less convenient and less appealing. While it is true that your student could use AI to fabricate a self-assessment, it raises the ethical stakes. Also, if self-assessment occurs at multiple stages in a project, it reduces the opportunity for students to rely on AI the night before the deadline.
As with any aspect of assessment, self-assessment is most useful when it is directed toward the learning outcomes we want to measure. Therefore, ask students to describe specific aspects of their work or working process. Short, targeted reflections are often sufficient and more manageable to review than extended narratives.
- Reasoning: Ask the student to make decisions under specific constraints, explain their rationale, and justify their trade-offs. This provides you with insight into how they apply concepts and make informed choices.
- Applied problem-solving: Create tasks that introduce complexity, incomplete information or competing goals. Instead of asking for a simple answer, for example, give the student a scenario where they must navigate ambiguity to find a solution. This gives you the opportunity to assess how the student applies their knowledge in context.
- Iteration and growth: Ask students to reflect on their growth. What did they change over the course of the assessment? What aspect of the work did they find most challenging, or still feel uncertain about? What knowledge, skills or attitudes from class did they need to apply to arrive at their final product?
Small changes can make learning more visible
Small adjustments or additions to an existing assessment can give your students the opportunity to demonstrate learning. Here are a few examples:
The learning outcome: Students will be able to analyze course readings by identifying key arguments, accurately explaining core concepts, and making relevant connections to course themes or contexts.
Self-assessment task: Have students use an annotation tool like Canvas Annotations to identify key arguments in a reading, define concepts, ask questions, and make connections to course ideas or their own context.
Learning outcome: Students will be able to analyze and justify problem-solving approaches by explaining their reasoning and evaluating the decisions made in developing a solution.
Self-assessment task: As a supplement to a problem set or code exercise, have students annotate their submission with a description of their approach and the decisions they made.
Learning outcome: Students will be able to develop and refine written arguments by planning, drafting, revising, and reflecting on their writing decisions.
Self-assessment task: Have students complete a writing assignment in phases. Begin with in-class planning and drafting, followed by a peer feedback and revision, followed by reflection after they have submitted the final work. At each stage, have students document the decisions and changes they are making.
Learning outcome: Students will be able to justify their reasoning by explaining and defending their work in response to questions.
Self-assessment task: Have students explain their approach and respond to questions about their work in real time. This could happen during the Q&A following a formal presentation. But it could also happen 1:1 or in small groups with you or a TA. What is key is that the conversation happens in real time.