BEL+T Guidance on Generative AI

GenAI has significantly impacted architecture, policy planning, and urban design, redefining the boundaries of creativity and efficiency. It offers the ability to generate countless design options and automate repetitive tasks, aiding architects and urban designers in developing optimised, innovative solutions.

However, concerns over the potential loss of creativity, homogenisation of designs, and insensitivity to unique local factors persist. There's a fear that genAI could overshadow human intuition, local cultural nuances, and personal identities within design. Privacy and data security are also crucial issues as urban planning becomes increasingly data-driven.

Despite these challenges, genAI is opening up remarkable opportunities. The technology can aid sustainable, equitable, and resilient urban development, simulating design intervention impacts on a range of factors, from climate adaptation to socio-economic growth. It also democratises the design process, allowing for genuine citizen engagement and interaction.

The emergence of genAI has significant implications for future graduates. While traditional skills remain relevant, there's an increasing demand for proficiency in AI, data science, social and environmental systems, and cross-disciplinary collaboration. Additionally, future professionals must be well-versed in the ethical application of AI, coupled with critical thinking abilities to responsibly navigate its complexities.

With genAI gaining ground, the urgency for upskilling has never been higher. Graduates who can wield this technology will have a competitive edge in the market. Moreover, the novelty of these techniques presents a significant opportunity for academic and commercial advancements, potentially sparking an influx of startups and monetisation possibilities. The future of these disciplines, therefore, converges at the intersection of technology, creativity, societal considerations, and a thorough understanding of genAI.

Authors:

• Dr Thanh Ho - University of Melbourne
• A/Prof Jason Thompson -  Senior Research Fellow, Univeristy of Melbourne
• Dr Sachitch Seneviratne - Research Fellow in Computer Vision and Health, University of Melbourne

To understand the effect of genAI in the BE disciplines, it is useful to first distinguish between genAI for design visualisation and communication (image-to-image applications) and the design concept and design development tools (text-to-image models).

Image-to-image tools are useful to generate photorealistic images from a reference image and sketches or clay renderings. These tools are expected to speed up the process of rendering production, and users won’t need to be as skilled as current experts in design visualisation.

Text-to-image tools allow designers to produce images of design ideas through textual prompts.

The development of text-to-image and image-to-image tools has already created a new professional figure, called ‘prompt engineer’. Prompt engineers currently work in visual arts and advertisement, and can likely be employed in architecture in future years, even without any specific formal design training. For the moment, we have seen a number of academics and architects promoting themselves online as AI experts because they use genAI tools. In fact, we believe it is more appropriate to consider such designers and academics simply as users and designers, testing and exploring the potential and limits of new digital tools.

GenAI is seen as a novelty and is used as a cutting-edge technology. The myth of the ‘novitas’ has always been appealing to architects and designers, and it was previously seen in digital design developments, including parametric design and optimisation. However, we believe that designers will have to go back to the roots of AI development (theory and technology) to fully explore how AI developments will unfold in the near future. Designers will need to rediscover Negroponte’s work and associated studies of the 60s and 70s. The same applies to many other research projects of the 90s, which focused on intelligent CAD systems and extensively discussed issues associated with intelligence, creativity, and human-machine interaction.

With this in mind, we can begin to theorise what this might mean for future graduates' knowledge or skills. Students can use genAI in design studio settings and seminar-based subjects without further elaboration. We believe that future graduates will need to develop critical thinking skills to reflect on the products of genAI. To enable a fruitful collaboration with such tools, future graduates must develop their communication and teamwork skills, which are the basis of a successful human-machine interaction. It sounds like a paradox, but the knowledge and skills future graduates will require can be found in the basic principles of architectural design thinking and design processes.

Authors:

• Dr Alberto Pugnale - Senior Lecturer in Architectural Design, University of Melbourne
• Dr Gabriele Mirra - University of Melbourne

The impact of genAI on the discipline of landscape architecture is still emerging and somewhat uncertain. There's an anticipation of a surge in AI-generated images that communicate design concepts more evocatively, albeit without the detailed design work typical of the field.

A challenge presented by the implementation of GenAI is the inherent recursiveness and potential bias in the process. While AI can access the vast repository of the internet, this does not always present an accurate or comprehensive representation of landscapes. Many landscape sites, even urban ones, lack exhaustive documentation or robust datasets. The nuances and intricacies that landscape designers often need to work with may not be adequately captured or represented by AI.

Nevertheless, the advent of GenAI opens up intriguing opportunities in the field. Landscape architecture has had a long-standing, occasionally tense relationship with representational tools, particularly living materials like plants that are highly variable in structure and growth outcomes. Predicting their appearance as part of a design can be a challenge because the eventual look of a design isn't always known. GenAI provides an accessible way to rapidly explore, visualise, and communicate these future states. Yet, it's important to note that this forms part of creative speculation and should not be regarded as absolute or a replacement for a comprehensive design process.

In terms of what this means for the skills and knowledge required of future graduates, criticality is of paramount importance. GenAI can be an exciting and engaging tool for designers to explore and visualise ideas, but students must understand its limitations. Given the language-based nature of AI prompts, they also present an additional challenge. In design disciplines, language isn't always thoroughly scrutinised due to the reliance on drawings and models for communication. Hence, to effectively engage with these emerging tools, graduates must become adept at critically examining language.

For a more comprehensive discussion, refer to this article on AI and landscape architecture by Landscape Australia.

Author:

• Wendy Walls - Lecturer in Landscape Architectural Design, University of Melbourne

The emergence of generative AI image generators is profoundly impacting the realm of architecture and design. These tools, deeply rooted in computational practices, offer new avenues for academics and creative practitioners to engage with their craft. The following video showcases the AI assisted Sketchbook project by Leire Asennsio Villoria and David Mah. They have pioneered a method of digital archaeology which involves reverse engineering and understanding the material intelligence of historical cultural artifacts, and embedding these into generative associative models. Through this process, new and novel design iterations are created that are deeply rooted in historical precedence. These outputs challenge traditional architectural paradigms and highlight the potential of AI generative tools in the design process.

However, a critical question remains: Can AI, built upon existing cultural artifacts, truly produce novelty, or does it risk anchoring culture in a repetitive cycle? The transformative potential of AI in design is evident, but its true capability to innovate and redefine remains a topic of exploration.

Understanding why students might decide to use AI for their coursework is critical to ensuring that we genuinely support learning, and that the institution meets its obligation to graduate employable and ethical citizens.

Students might elect to use genAI for their coursework for numerous reasons, and it is important to understand that not all of these reasons are mischievous or with an intent to cheat. Some students may lack confidence in producing work entirely themselves, whilst others may not feel motivated or supported to do so. Indeed, scholarship on why students participate in academic dishonesty more widely suggests that the possible reasons can extend beyond the desire to achieve certain results to include: feeling inadequately prepared for assessments; caring more about results than learning; confusion around what constitutes academically dishonest behaviour; feeling like the behaviour is commonplace amongst their peers; or feeling a lack of connection to their studies or institution more generally (Bryzgornia, 2022). Some scholars have even raised the notion of ‘ethical cheating’ in reference to students collaborating, sharing knowledge/information/ideas and using open-source platforms precisely to develop 21^st century skills, yet in ways that might traditionally have been considered cheating (see Brimble, 2016).

In relation to AI specifically, a survey of US higher education students conducted by Best Colleges in March 2023 showed that students hold diverse views towards the use of AI in university coursework, ranging from those who actively use it to those who believe it should be prohibited in educational settings (Richards, 2023). In the same survey, 40 percent of respondents said the use of AI defeats the purpose of education, and 63 percent said AI cannot replace human intelligence or creativity. Anecdotal reports suggest that students are also using genAI tools because they are fun, and also because they just want to explore what it can do. As discussed in the Assessment section of this guide, students deserve clarity and clear communication around what is considered proper versus improper use of AI in their studies and for each assessment task, what is encouraged and what may be required. This includes when and how students should disclose the use of AI tools, and any distinctions around expectations when it comes to AI use in text-based versus graphic-based formats.

Apart from clarifying university policies and expectations, it may be beneficial to discuss with students the use of AI by professionals and academics in the field, and the current set of ethical questions surrounding these practices. Siva Vaidhyanathan writes, this is a teachable moment for our students as well as ourselves. Not only are tools and technologies certain to develop over time, institutional and personal stances towards AI are context-dependent. Generally, if students feel uncomfortable or discouraged to discuss their views or habits with staff, this can contribute to a problematic gap between teacher assumptions/expectations and learner practices. ­­The more educators and students can feel like they are working together to promote learning and professional development the better. As Ouyang and Jiao (2021) argue, the advancement of AI technologies does not ensure good education outcomes; rather, the long-term goal of AI use in educational contexts is to contribute to a paradigm where learners are supported and empowered to take agency for their own learning.

GenAI has great potential in higher education, but it is crucial to approach these tools responsibly and consider the ethical considerations associated with them. This includes consideration of equity in assessment design, as paid and unpaid versions of genAI tools (such as ChatGPT) have access to different datasets.

A significant concern is the tendency of these models to perpetuate societal biases and discrimination (Dahmen et al., 2023). These models are trained on large amounts of data, and if that data is biased, the models will reflect these biases in their output (Atlas, 2023). In doing so, they reinforce existing societal issues and discriminations. To address this, it is essential for users to be educated about these biases, develop critical evaluation skills and gain technical expertise in mitigating biases when using these tools (Gimpel et al., 2023). This includes employing strategies such as proper prompt engineering to guide the genAI models towards generating content that is more inclusive, unbiased and aligned with ethical considerations. By proactively engaging in responsible practices, users can reduce bias and foster an equitable and ethically sound application

Paradigms of AI Usage by Learners in Higher Education: According to Ouyang and Jiao (2021) three paradigms can describe how AI is currently being utilised in education.

• AI-directed, where learner is considered as recipient (paradigm 1),
• AI-supported where learner is perceived as a collaborator (paradigm 2), and
• AI-empowered, where learner contributes as a leader (paradigm 3).

Paradigms 1 and 2 have been the focus of AI in higher education in the past two decades. There is a current call for Paradigm 3, an AI-empowered, Learner-as-Leader approach centred upon promoting human intelligence and integrated AI . This approach aims to resolve issues of bias in AI algorithms and datasets, lack of governance of AI decision-making, to promote learning and teaching experiences that are more socially just and inclusive.

Types of Datasets: For learning and teaching experiences to be more socially just and inclusive, it is important to understand how students use the AI platforms and the forms of information that are input and outputs (Dwivedi, 2023). In built environment education, students can engage with genAI platforms using several types of datasets including image, text, audio and/or code, depending on the subject.

Biases in Datasets: Depending on the type of genAI platform, datasets may not be curated or selected to identify an inclusive range of issues or perspectives. Such biases may be systemic, societal, cultural, racial, ethnic and/or methodologically, as well as intellectually fraught as large social datasets are fed into algorithms and unchecked algorithms can result in systemic discrimination that favours certain individuals or groups over others (Ferrara, 2023; Ray, 2023). Most datasets are Western-centric because of the dominance of these forms of information that are readily available for genAI platforms such as ChatGPT to utilise.

This range of potential biases is relevant to cultural, linguistic, ethnic and historic background to the content, or for a student, and should be recognised in support of socially just/inclusive learning (Ferrara, 2023). Datasets drive textual outputs such as essays, reports, summaries, reflective narratives, thesis, rendered images, development of audio outputs and drafts, as well as data analysis. Algorithms in AI/Machine Learning systems that seek to increase efficiencies can embed existing biases and propagate ongoing disparities. This compounding bias can hinder the achievement of social justice in classroom and decolonisation efforts by higher educational institutions.

Responding to Dataset Bias in your teaching: Datasets have a lifecycle of input, usage and interpretation. It is important that at each stage of the lifecycle, students are supported in how they relate to data and its interpretation for their learning and assessment (Dwivedi, 2023).

Some teaching strategies to consider include:

• Build students’ awareness of different types of biases that might be inherent to datasets;
• Encourage students to develop prompts that respond to biases by adding additional information such as ‘internationalise the prompt’ or ‘consider the Global South perspective’;
• Provide students with examples of how bias in datasets might impact their own worldviews about interpretation of readings/scholarship. This can include showing students that certain genAI outputs can impact respectful and ethical engagement from diverse scholars with varied cultural backgrounds, or may also lead to misinterpretation and distortion of information. Such distortion can be disrespectful and may project further bias/exclusion of diverse communities and places;
• Encourage students to check the authenticity of resources and not to rely solely on an output from a genAI platform as reliable information about various cultures, genders, races, ethnicities, histories or experiences of diverse communities to decolonise educational and professional practice efforts.

Important issues relating to genAI and teaching in ABP disciplines are related to creativity and authorship, and impact studios and other subjects involving innovation and creativity.  Each coordinator will need to explore and identify how students can best engage with these tools in relation to subject learning outcomes. Some questions and references are outlined below to assist this thinking.

When we ask students to ‘be creative’ in design-related disciplines or learning activities, we are asking them to contribute and iteratively refine their own beliefs, values and attitudes as they respond to a design challenge. Students learn to select from and/or transform ideas from precedents, research and their own experiences, as well as how to consciously reflect on and direct their approaches (Lawson & Dorst, 2009; Cross et al, 1994). We are asking them to participate in the ‘curious and beautiful relation between design problems and their solutions’(Lawson, 2007).  Students are rewarded for designs that contribute positively and innovatively in this context.

How is student innovation or creativity framed and identified in your subject through the ILOs and elsewhere? How is it assessed via brief and/or rubric?  

By contrast, genAI tools search, re-combine and deliver elements from data sets, producing a wide range of outputs including textual, numeric, code and graphic forms, in response to user prompts. Many doubt the capacity for AI to participate in ‘true creativity’ (Lawson, 2007; Kelly, 2019), claiming it lacks the motivation and independent judgement to create something truly new and useful. Human thinking is described as creative and flexible by nature, in contrast to the strengths of AI in relation to repetitive actions at vast scale, and managing complexity and multi-tasking. Some claim the Turing Test (in which an outcome that is indistinguishable from human production offers proof of intelligence) has been reached, while others claim creativity should be relocated to the perception of the beholder, rather than the contribution of a potential author (Natale & Henrickson, 2022).

Purported opportunities to ‘collaborate’ with genAI see typical models of design thinking transformed to propose linked human and computer contributions to identified phases. These outline perceived improvements to human thinking/expressing/building/testing/perceiving, using these tools to increase scope and decrease time and cost (Wu et al, 2021); or outline the ways in which different disciplines may creatively understand and engage with AI including as co-creators (Wingstrom et al., 2022).

A recent genAI-focussed panel discussion at the CSHE Teaching and Learning Conference heard panel members encouraging the design of learning experiences in which students could refine their creative practices and deepen their judgement by ‘sparring’ with the machine through reflective use of prompts and the creative recombination of outputs.

How might genAI strengths be distinguished from, and/or contribute to, student learning in the subject?

Elsewhere, human and AI production is being considered through the lenses of moral rights (Miernicki & Ng, 2020), intellectual property and copyright (Shtefan, 2021). A recent case before the US Copyright Office found the location of authorship to be via input (prompts) as opposed to output (AI-produced images), although the debate continues.

Simultaneously, some developers of various genAI platforms are in legal hot water, as artists claim these companies are infringing copyright by drawing on their published work without attribution, and the opportunities for recourse or even protection are overly limited (see a summary here). Concerns are raised for users, as the tools also collect requests and data from prompts, directly or via ‘plug-ins’.  As AI starts to draw on outputs it claims as ‘its own’ for future production, it all gets a lot more complicated.

How can students learn about and respond to the IP concerns of others? How can students protect their own IP in this subject?

Our challenges include supporting students to engage creatively with emerging tools, to build their own creative expertise and judgement, and to effectively demonstrate authorship and to protect their own IP and privacy in the process. The value and personalisation of creativity and its expression remain central to this learning, and a crucial aspect of the learning that students need space and support to practice and refine.

As the opportunities and challenges offered by genAI technologies continue to emerge, the University and the Faculty are working to clarify what is required of students and of teaching staff.  We’ll summarise key policy and guidance here.

Guidance for teaching staff

As outlined above, it is important to have open discussions with students—before they submit work—on the ethical use of genAI and on the University’s requirements around its use in assessment tasks.  Subject coordinators may also consider reviewing assessment task design in the context of genAI.

The BEL+T team are available, please reach out to us at abp-belt@unimelb.edu.au to discuss.

An AI writing detection tool has been integrated into Turnitin, and is being further refined through its use at the University.  Learning Environments has produced comprehensive information on the reliability of the tool, its functionality, and what to do if the tool returns a high percentage of text flagged as likely to have been AI-written. Note that:

As with the Turnitin similarity report, a high percentage of text flagged as likely to have been AI-written is not proof that academic misconduct has taken place but may be a sign that further investigation is warranted. Academic judgement should be used to determine whether to investigate the matter further, including by discussing with the student.

Investigations of such cases may look for additional evidence that the assessment material submitted was completed by the student, on the balance of probabilities. This might include looking at the metadata of the files submitted, comparing the assessment to other work completed by the same student, asking the student to provide drafts of the work, to describe how they completed it, or to explain the content of the assignment. Such evidence, taken together, may allow committees to come to a judgement about whether or not the work in question is likely to be the students’ own writing.

As with all potential student academic misconduct, subject coordinators are the main point of contact. Staff should not accuse students of academic misconduct, but should seek more information where appropriate, and gather evidence.  Steps for referring any suspected student academic misconduct for investigation are outlined via BEL+T’s Academic Integrity Guidance

If a student is found to have committed academic misconduct by representing work generated by artificial intelligence software as their own, they will be subject to the penalties outlined in the Schedule of Student Academic Misconduct Penalties.

For guidance on those investigations and on their possible outcomes, please reach out to the ABP Student Programs team at abp-misconduct@unimelb.edu.au. (Even if the outcome involves an educative approach—which may be suitable in some initial cases—please keep the ABP Student Programs team in the loop, so that the Faculty can maintain a comprehensive record of engagement with each student.)

For further information on identifying and responding to potential academic misconduct in ABP subjects, please see BEL+T’s Academic Integrity Guidance, and the links to Academic Misconduct on the ABP Faculty Intranet.

What’s required of students?

Per the University’s Statement on the Use of Artificial Intelligence Software in the Preparation of Material for Assessment (21/4/2023):

…all work submitted by an individual student must be their own. In the case of group work, the individual contribution of each student must be their own work.

If a student uses artificial intelligence software… …to generate material for assessment that they represent as their own ideas, research and/or analysis, they are NOT submitting their own work. Knowingly having a third party, including artificial intelligence technologies, write or produce any work (paid or unpaid) that a student submits as their own work for assessment is deliberate cheating and is academic misconduct.

If a student uses AI generated material in the preparation of their assessment submission, this must be appropriately acknowledged and cited…

Please encourage students to review Academic Integrity at the University of Melbourne for further information on why academic integrity matters and how to avoid academic misconduct. Students should also review the Student Academic Integrity Policy (MPF1310). The Advice for Students Regarding Turnitin and AI Writing Detection provides further information about the use of AI tools and the ways the University may detect and respond to inappropriate use of AI tools.  This document highlights the important role that subject coordinators play in clarifying the appropriate use, or not, of genAI tools for learning and assessment.

Further information

As the University and Faculty continue to engage with the opportunities and challenges offered by genAI, further policy and guidance will continue to be developed.

The Melbourne CSHE has compiled a set of useful resources on Assessment, AI and Academic Integrity.

The ABP Student Programs team is currently developing a guide for subject coordinators around genAI and academic integrity.  Further advice is also in development by Chancellery and will be available before the start of Semester 2, 2023.