Lane 2: Designing AI-robust assignments
Lane 2 assessments are unsupervised assessment tasks in which students may use generative AI, but are not required to do so. In many cases, the learning outcome itself does not need to change. What changes are the assignment design and the assessment criteria: these determine whether, and how, AI may be used while still allowing students to demonstrate the intended outcomes.
When designing AI-robust assignments there are four key aspects to take into account.
1. Learning outcomes
Lane 2 outcomes typically emphasise judgement, evaluation and disciplinary reasoning—capabilities students can demonstrate even when AI tools form part of their workflow. The focus shifts from producing output to making informed, defensible decisions about that output.
In practice assess whether students can:
- evaluate the quality and accuracy of content and competing arguments using disciplinary criteria
- identify limitations, bias, or gaps and address them appropriately
- justify methodological, analytical or design choices
- synthesise (multiple) sources, evidence, and theory in a way that reflects disciplinary standards
- apply concepts to new contexts
- interpret incomplete or conflicting evidence
- evaluate competing theoretical explanations
- explain how disciplinary knowledge informed their decisions
- take responsibility for the final product, regardless of how it was generated
- Science: Students review an AI-generated lab report, identify methodological flaws or incorrect interpretations, and revise it with justification grounded in scientific principles.
- Social sciences: Students compare an AI-generated policy recommendation with course literature and empirical data, evaluating its validity and proposing improvements.
- Humanities: Students critique an AI-generated interpretation of a text, identifying missing perspectives or misreadings and developing a more nuanced argument.
- Methods/statistics: Students assess whether an AI-suggested analytical approach is appropriate for a dataset, justify their choice of method, and explain limitations.
- Professional fields: Students use AI to draft a product (e.g. lesson plan, legal argument, clinical case summary) and then refine it, explicitly justifying decisions in relation to professional standards.
Across these examples, the core question is not “Did the student produce this themselves?” but “Can the student demonstrate sound disciplinary judgement in how the work was developed and evaluated?”
2. Assignment descriptions
To make the learning process visible, students may be asked to document or explain how they worked with AI tools. This is necessary because, in unsupervised settings, the final product alone no longer provides reliable evidence of learning. Without insight into the process, it is difficult to determine whether students engaged in the intended cognitive work or outsourced key steps.
Process evidence
Assignment descriptions should therefore be adapted to explicitly require process evidence alongside the final product. This shifts the focus from what was produced to how and why it was produced, making students’ reasoning, decisions, and use of disciplinary knowledge assessable.
Students might be asked to:
- show development over time (e.g. drafts, revisions, annotated outputs)
- explain how disciplinary knowledge informed their choices
- document their AI use (e.g. which tools were used, for what purpose, and at which stages)
- justify key decisions (e.g. why they accepted, rejected, or modified AI-generated output)
- critically evaluate AI contributions (e.g. identify inaccuracies, bias, or limitations)
3. Grading criteria
To increase the AI-robustness of assessments, grading criteria should focus less on aspects that can easily be outsourced to AI (e.g. spelling and grammar accuracy, fluency and readability of writing, basic structure an coherence, surface-level summaries of literature or theories). Instead, assessment criteria should focus on aspects such as:
- quality of reasoning
- justification of decisions
- depth of critical engagement
- integration of disciplinary knowledge
- ability to evaluate AI-generated output
- transparency about how AI tools were used (for example through brief explanations, logbooks, or documentation of prompts and revisions).
4. Instructions
Instructions should move beyond “AI allowed/not allowed” and instead:
- specify which uses are appropriate or even encouraged (e.g. brainstorming, editing, generating alternatives)
- clarify which aspects must reflect students’ own reasoning (e.g. analysis, conclusions, justification)
- indicate what must be documented (e.g. prompts, outputs, decisions) and in what format (e.g. copy of prompts, link to AI chat, etc.)
- how AI use will be considered in the assessment criteria and, if applicable, link AI use explicitly to assessment criteria (e.g. critical evaluation of AI output)
- what remains the student’s responsibility when using AI tools
Ask students to respond to the following questions:
- Which AI tools did you use (if any)?
- For what purposes did you use them?
- What parts of the output did you revise or reject?
- How did you verify the reliability of the information?
- What decisions in the final work were made by you?