University of Melbourne graduates emerge as well-rounded, thoughtful and highly skilled professionals who make a positive impact across the globe. As part of our dedication to delivering outstanding teaching and learning, we’re committed to safeguarding a culture of academic integrity, ensuring our graduates have the skills, knowledge, and understanding built through our curriculum and that our degrees continue to represent genuine educational achievement.
The increasing capability and prevalence of Generative Artificial Intelligence (GenAI) tools, alongside an ever-growing number of cheating services and websites, mean some traditional assessment types can no longer assure that students have achieved the intended learning outcomes required from a task.
To address this risk, the University is introducing two new assessment principles. These require that either 50 per cent or more of the total marks in a subject are based on secure assessment, or that a form of secure ‘programmatic’ assessment is adopted within a degree.
Read more about the University’s Assessment Principles
Over the next few years, the University is undertaking institution-wide assessment reform to ensure our curriculum is in line with these principles. Transforming our approach will take time and the University will implement this work in stages. Faculties will lead the approach within their discipline areas, coordinating where assessment practices are shared, and guided by whole-of-University targets and standards.
A balance between open assessment types, which provide important opportunities for learning and feedback, and secure assessment types will be maintained. Work is underway to consider how these assessment types can be combined within subjects and tasks. Teams across the University are also working to design innovative assessment types to expand the range of secure assessment options, and disciplines are encouraged to consider what practices might be appropriate for their learning goals.
Explore the current guide to secure assessment types and tasks
Generative artificial intelligence (GenAI) is a form of artificial intelligence that is distinct due to its ability to produce novel outputs in response to natural-language prompts that require no programming knowledge. It is different to machine learning, an area of AI in which there have also been significant recent advances, particularly in its availability and ease of use, and in the types of inputs and outputs that can be used. Popular GenAI tools include image generators, such as Dall-E, and chatbots, such as ChatGPT. ChatGPT belongs to a type of GenAI called a Large Language Model or LLM, and was released to the general public in late 2022.
Powered by LLMs, a proliferation of GenAI tools has followed, which are able to ‘read’ images, texts, data files and audio inputs and output in all of these formats and more. Embedded GenAI tools are appearing across common consumer software, including Microsoft Office (Copilot), many have features or access levels that remain free to the public, and a number of open-source LLMs are also now available. Rapid advancements and widespread competition mean that the limitations of today’s tools are swiftly overcome, and any available niche quickly filled.
Widely available tools that can almost instantaneously generate novel outputs have clear implications for many forms of assessment. Early limitations of GenAI tools such as the inability to reference recent events, use accurate citations or respond to custom datasets have been swiftly overcome and current limitations are likely to disappear just as quickly.
This is particularly challenging for those styles of assessment where learning has been evaluated through students’ ability to produce original outputs, usually written pieces but inclusive of design responses, visual art, animation, video, music, podcasts, and presentations. Where such material is produced without observation it is now very difficult to determine if a GenAI tool could have been used to generate part or all of these products.
One response to this challenge is to suggest that the process of producing these outputs without GenAI tools is no longer important, and that assessment should pivot to considering the quality of the output students can produce when using GenAI. Having students engage with GenAI is an important part of producing graduates who are AI-literate and prepared for a future of work where GenAI tools are likely to be embedded in productivity tools. Industry uses of GenAI tools are still nascent in most cases, but embedding authentic teaching, learning and assessment activities in the curriculum, which gives students hands-on experience with Gen AI, and challenges them to demonstrate how they add value above and beyond the capabilities of a tool, are a critical part of responding to the GenAI challenge.
A key skill in using GenAI tools, however, is the ability to evaluate the quality of the outputs of these tools. This ability to critique and evaluate is only possible after a degree of expertise has been developed in the related topic area, and experience with quality products. The challenge for educators then is not just how to build students’ expertise and understanding by asking them to undertake certain activities, but how to judge when this expertise has been developed independently of Gen AI tools. Recognising that the artefacts students produce for evaluation and assessment of learning may no longer entirely reflect the development of an individual’s foundational knowledge and understanding, means we must consider alternative ways in which to observe or determine this.
GenAI tools are increasingly embedded within the personal and professional lives of our students and graduates. These tools can be useful, and students need to learn how to use them appropriately and effectively as a key skill for professional practice in the future. As GenAI tools produce novel outputs it is also very difficult to enforce such a ban, especially as tools become more sophisticated and ‘solve’ many of the original limitations.
While AI detector tools have been on the market from early 2023, they have a very wide range of efficacy and quality. The best-in-market tools are under continuous development to keep up with evolution both in the GenAI tools themselves and in the behaviour of those who seek to ‘fool’ the tools with patchwriting, washing and other techniques.
No detector is entirely able to eliminate either false negatives (where the tool fails to detect AI-written material) or false positives (where the tool incorrectly flags human-written material as AI-written). The poor quality of many publicly available tools has led to significant negative sentiment on their use and high levels of student anxiety. This is preyed upon by unscrupulous online companies who use falsified ‘detection’ scores to sell students ‘washing’ products, further magnifying concerns.
While responsible use of high-quality detector tools can play a role in securing assessment, they must be used with care and are only one part of a broader approach to maintaining academic integrity.
No, not all assessment needs to be secure and the Guidelines outline two types of assessment: secure assessment and open assessment. The first principle of the guidelines stipulates that 50% of the assessment in a subjects must be secure.
Open assessments will play a critically important part in our curriculum and in students learning. There is no implication that open assessments will necessarily allow the use of GenAI tools. That said, students will participate in many types of open assessment activities as part of the discipline, in related employment or similar and these may never be able to be ‘secured’. For these reasons, the University is seeking to strike a balance between the types of assessment students undertake in their studies, such that we can be assured that students who graduate from our degrees have achieved the intended learning outcomes, while at the same time not removing opportunities for engaging, authentic experiences that may not be secure.
Not necessarily. Invigilated exams are one type of secure assessment, but academic staff are encouraged to explore other assessment types that may be appropriate for the learning outcomes of their subject. The current guide to secure assessment types can be found here. Work is underway across the University to test and develop more secure assessment tasks, and to build out case studies and advice on implementing these. The guide is expected to continue to be updated and expanded as this work progresses.
An Academic Adjustment Plan (AAP) is a critical resource for students who have circumstances which require some form of reasonable adjustment to study. Adjustments are made on a case-by-case basis with the student in advance of assessments and may include provision of alternative assessment arrangements. As we explore and implement new assessment practices across the University, AAPs will remain an important tool to ensure we are an institution where all students have the opportunity to succeed.
Work is also already underway through the Advancing Students and Education Strategy (ASE) and the Diversity and Inclusion Strategy 2030 to improve the inclusive design and delivery of teaching, learning, and assessment such that fewer adjustments are required. We encourage all staff to explore and integrate guidance in the Assessment Principles to Support Student Wellbeing and the Accessibility in teaching, learning and assessment webpage.
Please do! The University has a number of resources to support you in experimenting with and embedding GenAI in your subject, including in assessment. While teaching and learning uses of GenAI are still emerging, the most promising use cases appear to be:
Whether or not assessment is secure is based on the degree to which an academic can reliably judge whether or not the student has met the intended learning outcomes. This will typically mean that some element of observation is included in the assessment. However, it may be that in some discipline areas and for some intended learning outcomes the use of GenAI tools is permitted (or even required) in completing an assessment and meeting the intended learning outcomes.
To assure that our graduates have met the learning outcomes we certify, the University is introducing two assessment principles. These require that either 50% of the assessment in a subject is secure assessment, or that a form of ‘programmatic’ assessment is adopted within a degree or program to assure learning. It is anticipated that most coursework programs will adopt a subject-based approach, and that programmatic approaches will be approved by exception by the Academic Board.
Associate Deans Teaching and Learning are working through plans to support the implementation of assessment changes throughout the curriculum over the next few years, with a view to prioritising subjects with the highest student numbers. This work will be supported with expert learning design assistance, assessment guidance and strategic funding. For more information on how this will work for your discipline and in your faculty, please contact your Associate Dean Teaching and Learning (or equivalent).
Cadmus is an online assessment tool currently in use at the University. It allows subject coordinators to create, and students to complete, written assessment tasks in a custom digital environment.
Simply using Cadmus will not ensure that an assessment is secure. However, the University is actively exploring the ways in which the use of Cadmus can be integrated as part of a wider assessment design approach to support the assurance of learning.
Further information and guidance will be made available on Cadmus and its role in the assurance of learning process following the completion of this work.
Your insights and contributions are helpful to shaping assessment transformation across the University. If you have any considerations we should keep in mind, or any recommendations or examples of secure assessments we should consider, please send them though to us.
While we will review all responses carefully, we might not be able to respond individually to each one.