Understand Your Options

Here’s a key insight from the framework: you don’t have to choose between ‘all AI’ and ‘no AI.’ Instead, think about a two-track approach where your programme includes both secured (AI-restricted) and open (AI-integrated) assessments.

Track 1: Secured / AI-Restricted

These are assessments where students must demonstrate knowledge or capability independently, without AI support. Examples:

• In-person exams (no resources, supervised)
• Clinical skills demonstrations (real-time, observed)
• Foundational knowledge assessments (verifying core concepts students must know)

Why Track 1 matters: Some capabilities truly must be demonstrated without AI. A surgeon needs to know human anatomy independently. A teacher needs to understand pedagogical theories without a search engine. Not everything can be outsourced to AI.

Track 2: Open / AI-Integrated

These are assessments where students are encouraged to use AI ethically and critically, reflecting authentic professional practice. Examples:

• Analysis and critique of AI outputs (students use AI, then evaluate what it got right/wrong)
• Complex problem-solving in realistic contexts (students might use AI as a thinking partner)
• Projects that mirror real-world work (where professionals actually use AI tools)

Why Track 2 matters: In most fields, the future workplace will involve AI. Students need to learn how to use it responsibly, critically, and ethically. An assessment where students apply AI tools and reflect on the results develops genuine capability.

The balance: A coherent programme includes both tracks. Some high-stakes assessments are Track 1 (secure the foundational knowledge). Other assessments are Track 2 (develop AI fluency). Students experience a learning progression where they first master core material, then learn to apply AI thoughtfully.

This isn’t either/or. It’s both/and.

Whether you’re designing Track 1 or Track 2 assessments, the framework suggests five practical strategies that make learning visible and make shortcuts much harder:

1. Multi-Stage Tasks

Instead of one submission, break the work into stages. For example:

• Week 5: Submit a proposal or plan (shows initial thinking)
• Week 8: Submit an annotated draft or progress report (shows refinement)
• Week 10: Submit final work + reflection on how thinking changed

Why it works: You can see the evolution of learning. It’s much harder to pretend understanding if you had to show your thinking at multiple points.

2. Oral Components

Add presentations, vivas, Q&A sessions, or discussions. For example:

• 10-minute presentation of key findings (shows understanding)
• 15-minute defence where you answer challenging questions (can’t fake it if you don’t understand)
• Seminar discussion where students explain and critique each other’s work

Why it works: Real-time human interaction. AI can’t show up to defend ideas. You learn a lot about understanding from how students respond to tough questions.

3. Reflective Elements

Ask students to think aloud about their process. For example:

• ‘What was your thinking when you made this choice?’
• ‘What feedback surprised you? How did it change your approach?’
• ‘If you used AI, where did you use it and why? What limitations did you notice?’
• ‘What would you do differently if you had more time?’

Why it works: Reflection reveals understanding. A student who can’t articulate their reasoning probably hasn’t done it themselves. Also signals: ‘I care about your thinking process, not just your final answer.’

4. Collaborative Work

Include group projects with clear individual accountability. For example:

• Group project with peer evaluations of individual contributions
• Collaborative problem-solving where each person’s role is documented
• Peer feedback sessions where students comment on each other’s work

Why it works: Collaboration is harder to fake. When people actually work together, you see negotiation, disagreement, synthesis. Plus, collaboration mirrors real-world professional practice.

5. Authentic, Real-World Problems

Design tasks around genuine challenges, not artificial scenarios. For example:

• Real client projects (non-profit needing marketing strategy)
• Current events analysis (applying concepts to what’s happening now)
• Community partnerships (students solving problems that matter to local stakeholders)
• Discipline-specific professional scenarios (what would a real engineer/nurse/architect actually do?)

Why it works: Real-world problems have constraints and complexities that resist template answers. They require contextual judgement. They’re harder for AI to pattern-match because the situation is novel and specific.

Here’s the thing: these aren’t isolated tricks. They work best together.

You might think: ‘That sounds great, but I have over 100 students in my class. How do I add all this?’

The framework directly addresses this. You don’t have to mark everything the same way.

For early stages (multi-stage tasks):

• Pass/fail marking instead of detailed criteria
• Checklists (‘Did they submit? Is it clear?’) instead of rubrics
• Sampling: grade a subset, spot-check others

For collaborative work:

• Peer feedback does some of the evaluation work (students assess each other)
• Structured peer evaluation forms can be quick to complete

For oral components:

• Group presentations (5 students at a time) instead of individual vivas
• Class discussion instead of individual Q&A
• Student recordings (video submission of explanation) instead of live oral

For reflective elements:

• Structured prompts (‘In 100 words, explain one key decision’) instead of open essays
• Rubric focused on presence of reflection, not depth (easier to assess at scale)

The point: You don’t redesign everything simultaneously. Start with your highest-risk assessments. Start with strategies that feel most manageable for your context. Iterate.