Sorry! De informatie die je zoekt, is enkel beschikbaar in het Engels.
This programme is saved in My Study Choice.
Something went wrong with processing the request.
Something went wrong with processing the request.

Creating AI-proof assessments using the two-lane approach

Back to the didactic tips overview page
Last updated on 19 November 2024
With the advent of generative AI, many programmes and lecturers face the challenge of determining how to make their assessments 'AI-proof'. This is particularly relevant for take-home assignments such as essays, programming tasks, laboratory reports, case studies and dissertations where monitoring students' use of generative AI is impossible.

But is it effective to try making these assessment forms 'AI-free' or 'AI-partly-allowed'? Or is a broader approach needed? The University of Sydney offers a useful perspective. Read more about their two-lane approach here. 

When students work on assignments at home without supervision, it's difficult to verify whether they're working entirely independently or using generative AI. This isn't a new problem; students could previously receive unauthorised help from friends, parents, or ghostwriters. Although this problem is sometimes common (Jones, 2011; Christensen Hughes & Eaton, 2022), it appears manageable. These assessment forms were never completely prohibited. It remains a manageable risk for programmes, provided the usage isn't widespread and systematic, and lecturers and examiners maintain vigilant oversight. 

Monitoring generative AI use without supervision is impossible
However, generative AI applications such as ChatGPT, Dall-E, Copilot, or Claude belong to a different category. This AI can generate excellent, well-thought-out, well-formulated and structured content based on a good prompt. These applications are widely available and are already being used extensively by students, including for other tasks such as graphic design, creating summaries, asking questions about the material, or reviewing papers for structure and spelling. 

It's virtually impossible to monitor how much generative AI students use in their assignments. This applies generally and even when lecturers explicitly permit different levels of AI use, such as only for linguistic checks or idea development. Therefore, AI-proof assessment of the aforementioned assignment forms that students must complete at home or independently doesn't exist.

A two-lane approach
Researchers at the University of Sydney (Liu & Bridgemen, 2023) devised an approach to address this challenge. The guiding principle is that education must equip students to participate ethically and actively in a society permeated by AI. They therefore suggest that it's unnecessary to choose between, for example, abolishing homework assignments or letting everything slide. Instead, they propose that a good curriculum or course balances the two assessment perspectives - Assessment of Learning and Assessment for/as Learning (Black & William, 1999) - where generative AI can play a role in both assessment perspectives. Liu and Bridgemenn describe curriculum and course design as a combination of elements from two lanes (PDF).

Lane 1 – Assessment of learning 

On one hand, programmes, lecturers, and the professional field need to be able to determine what students can do independently (cognitive knowledge accumulation and the meritocratic principle). This involves Assessment of Learning, or: extra safeguarding of learning verification. This forms the first 'lane'. This requires supervision, where students must present and respond to questions from lecturers or examiners directly from their accumulated (ready, broad, deep) knowledge base. These assessment forms include examinations, presentations, interviews, defences, papers, and oral assessments. The use of generative AI isn't necessarily prohibited here. A student might need to demonstrate their competence in using generative AI, whether in combination with an assignment or test. 

Lane 2 – Assessment for/as learning 

On the other hand, the premise is that students must learn to use the opportunities of generative AI and understand its limitations and act accordingly. This is about Assessment for or as Learning. Here, generative AI is seen as an inseparable part of students' and future professionals' thinking and working tools. In this process, through guidance, learning to prompt, critical analysis and discussion, and attention to academic integrity, lecturers and fellow students maximise the stimulation of student learning through assignments. We provided ideas for this in an earlier teaching tip. The activities students perform and results they produce in this 'lane' of a course or programme are also counted in a final assessment of students' acquired competencies, including the use of generative AI. 

Combining lane 1 and lane 2 

In a curriculum, and in every course, programmes or lecturers must make a well-considered choice about implementing methods from both lane 1 AND lane 2. Each course contains elements to directly assess students' knowledge development under controlled conditions AND elements where AI use is unlimited. In the second lane's implementation of lessons, workshops, projects and lectures, AI is purposefully taught, used and critically evaluated. The choice is determined by the learning objectives or final terms that stem from AI's place in life and work. For a thorough final assessment of a student's competency, you include judgements from both lane 1 and lane 2.

Notes and implications
The strength of the two-lane approach perspective lies in the model's clarity. It instantly shows that there isn't a dilemma, but rather that with a good educational and assessment approach, the concept of AI-proof assessment becomes a manageable task. 

Of course, the model is a simplification of reality. Upon deeper reflection, you discover that much education is already designed according to the two-lane approach principles. And that's good news: we don't need a revolutionary different educational setup. However, it does require programmes and lecturers to continuously recalibrate their learning objectives, and new technical developments can quickly overtake the design. 

Note that examination boards and programme committees also face challenges, and good coordination with each other is important here. The same applies to providing a level playing field for all students. The variation in AI competencies and availability is enormous: students who are less proficient or have less access to AI must not be disadvantaged compared to students with more access. In lane 2, clear monitoring and control by programmes and IT services is needed on the rapid development of AI in professional software and plug-in applications within every conceivable application. Standard packages are a thing of the past, and safely offering only ChatGPT, eduGenAI or Copilot to students isn't sufficient. This demands flexibility from future-proof IT support organisations and financial space. 

Let us know what you think of Liu and Bridgemen's perspective by sending a message to onderwijswerkplaats@vu.nl.

References

Quick links

Research Research and Impact Support Portal University Library VU Press Office

Study

Education Study guide Canvas Student Desk

Featured

VUfonds VU Magazine Ad Valvas

About VU

About us Contact us Working at VU Amsterdam Faculties Divisions
Privacy Disclaimer Safety at VU Amsterdam Colofon Cookies Web archive

Copyright © 2024 - Vrije Universiteit Amsterdam