Assessment Design for Collaboration with GenAI

A framework for understanding and implementing genAI in assessment through the lens of group work dynamics.

While many models for assessment design in the age of genAI focus on the 'quantity' of authorised genAI use by students, the following guidance outlines collaborative roles between a student and genAI as a form of group work for learning. Approaching the authorised use of genAI through the lens of group work offers both a useful and familiar approach for university educators and students, as well as language to clearly communicate 'authorised' use of genAI for specific assessment items.

Group work is frequently integrated into the design of learning activities or assessment tasks. Educators may include group work activity in subject design to promote deeper learning of subject content (Gaudet et al., 2011) or to focus on developing interpersonal skills (Kotey, 2007). The social interactions afforded through group work support the development of skills to work in teams successfully (Boud et al., 2001). This includes the development of critical self-awareness of students’ own learning through the exchange of feedback with group members. GenAI’s ability to tailor its responses through prompting draws parallels to learner experiences of peer interaction.

GenAI platforms can respond and adapt to interactions with the user, moving beyond a tool for cognitive offloading (e.g. calculator) as an involved co-learner contributing to the construction of knowledge. Lodge, Yang, et al. (2023) describe the nature of this relationship along a spectrum in which the interaction between the human and the AI is either focused and driven by the individual learner or a collaborative arrangement between human and machine. This spectrum of collaborative relationships is also observable in higher education when students engage in group work.

Building on these parallels between group work and genAI interaction, BEL+T has produced a new framework for evaluating and authorising student use of genAI in assessment. The framework presents a typology that views the student-AI relationship through the established lens of group work dynamics, identifying three distinct assessment types: individual, cooperative, and collaborative. This typology, presented in the table below, provides educators with clear guidance for understanding the roles and interactions between students and genAI, while offering practical insights for task design and learning objectives. The framework is further illustrated through examples and suggested level descriptors in the accordions below, helping educators make informed decisions about integrating genAI into their assessment design.

This framework is propositional, and aims to assist teachers in considering the challenge of incorporating AI into their assessment design through a familiar lens. It aims to help teachers to ‘authorise’ and to assess emergent skills in the context of (and alongside) a learning focus for a particular subject. The level descriptors may be adjusted and added to other criteria in rubrics for example.

It is of note that evidenced proof of ‘authorised use’ is still not completely reliable, ie actual student use of AI may not be detectable, despite the requirements of an assessment outline. However, the application of this approach means that students have an incentive to describe and to develop their use of these tools.

This approach was first presented at the ASCILITE Conference 2025. Please use the following citation – the conference paper can be accessed via the DOI link below.

Tregloan, K., & Song, H.(2024).From How Much to Whodunnit: A framework for authorising and evaluating student AI use. In Cochrane, T., Narayan, V., Bone, E., Deneen, C., Saligari, M., Tregloan, K., Vanderburg, R.(Eds.), Navigating the Terrain: Emerging frontiers in learning spaces, pedagogies, and technologies. Proceedings ASCILITE 2024. Melbourne (pp. 255-265). https//doi.org/10.14742/apubs.2024.1441

	Individual Assessment	Cooperative Assessment	Collaborative Assessment
Student Roles & Expectations	Student as Author: Primary producer of the final output(s) Goal determined by the student.	Student as Project Director: Managing production and curating contributions; Producing section of final output; Commentary on the relationship of parts and contribution (i.e. peer-evaluation)	Student as Co-Designer: Contributor to joint iterative exercise, ultimately directing and evaluating the shared work towards a final output; Student will train AI re shared visions and goals; Student will adjust the vision and goals in response to Ai’s adaptive generation.
GenAI Role	GenAI as Assistant: Limited cognitive offloading as a refinement of student production (e.g. spellcheck, grammar, code).	GenAI as Group Member Produce defined segments/sections of the final output under the direction of the student(s).	GenAI as Co-Designer: Iteratively refine and adapt contributions responding to students’ efforts; Refining datasets/inputs (defined or developed by student)
Task Design	Goals and outcomes are pre-determined by educator Students work independently to accomplish learning goals	Goals and outcomes are pre-determined by educator A clear boundary is set regarding the body of knowledge Activities have detailed instructions of how the final outcome(s)	Open-ended but focused task(s) for learning Exploration of ideas Learning to learn Activities are structured but means of how to achieve the final outcome(s) determined through engagement with the task
Learning Focus	Process of individual skill development and knowledge acquisition	Development of skills and knowledge through known strategies (i.e. specific activities are set for students to conduct as part of the assessment)	Social construction of knowledge and skills through that may involve trial-and-error of testing and iteration of novel outputs

Note: though these assessment design types have been presented independently , these approaches may be integrated as complementary elements of a more comprehensive assessment task. As outlined, it is of note that the roles undertaken by the student and AI are different and are (currently) not equivalent to the approaches that may be taken by two independent humans.

Individual Assessment designs focus on a student’s personal achievements and learning. The assessment is designed with the expectation that the student is working by themselves to accomplish the final outcome, and that evaluation in this context is about validating a student’s personal skills or knowledge. When considering a role for genAI in such assessment designs, educators may consider minimal cognitive offloading. Functions may include spell-checking, code-checking, calculations by a calculator, presentation layout suggestions (e.g., in PowerPoint slides), or summarising selected text for further analysis by students. Higher-order learning outcomes to be evidenced through the assessment task can occur independently from the support genAI provides to the student (Lodge, Yang, et al., 2023). In this form, the educator must set clear goals and outcomes for the assessment task while the student leads the development and decision making towards the final outcome. Students are responsible for producing the required submission, with limited support.

Examples:

An example of an individual assessment design could be a self-reflection essay. In this particular type of assessment, the educator requires students to evaluate their own thoughts and opinions, evidenced through a written essay. Individual assessments are solely focused on evidencing students’ learning by tasking students to work independently towards the final outcome and where their learning progressions are unrelated to other students. In a self-reflective essay, students are tasked with demonstrating the capacity to reflect and articulate their own personal thoughts and insights. The student may incorporate genAI to acts as a passive tool, providing some minimal cognitive offloading by editing grammar and/or proofing the written text, however, it does not contribute to development and engagement of the students’ reflective thinking.

Other examples of individual assessments include forms of written essays where genAI platforms may suggest synonyms and alternative word choices and restructure sentences and/or written paragraphs. Additionally, individual assessments designed to incorporate multimodal forms of submission (e.g. visual images, multimedia, etc.) may involve students engaging in genAI in the editing process through generative filling and expanding (i.e. in-painting and out-painting). Educators may consider instructing students to submit their assessment task before they engage with genAI. Additionally, educators may wish to incorporate activities requiring students to critically reflect on how genAI has contributed to their work and how the student has been able to manage the platform to support the development of the final output.

Suggested Level Descriptors: Individual assessment

Poor	AI use moves beyond the authorised use as set out in the task requirements AI use is ineffective and does not improve the student’s own work AI use does not align with relevant conventions or assessment requirements
Good	Application of AI is clear and effective for the task requirements AI use has improved the student-produced work in relation to the authorised aspects
Excellent	Use of AI is strategic and deliberate Student evaluates the application of AI, and adjusts further AI use to significantly augment the work Student may combine multiple AI tools to address specific aspects of the submission

Olsen and Kagan (1992, p8) describe cooperative learning as group learning that is: “dependent on socially structured exchange of information between learners in groups and in which each learner is held accountable for his or her own learning”. Students who engage in well-designed cooperative learning demonstrate increased intrinsic motivation in engaging with their studies, developing higher-order thinking skills and improved attitudes towards curriculum (Johnson & Johnson, 2013). Cooperative assessment designs are planned and prescriptive, providing students with highly structured and descriptive materials and clear directions about how to work together in groups towards a single output that will demonstrate their learning. Elements may be driven by independent personal goals and values, and may be independently assessed. This approach aims to support an interdependent relationship between members. The assignment of roles clarifies expected contributions for each member and their responsibilities.

Examples:

An example of a cooperative assessment design is a jigsaw reading task. In this case, the educator allocates a specific reading to each group member who will share insights with the rest of the group. The student role and expectations are clearly communicated, including expectations around building expertise in assigned reading. For a student paired with a genAI tool, an educator will provide structured directions on how to engage with the genAI, perhaps including the types of prompts that might shape the platforms response. This assessment design may also involve students evaluating the quality of responses produced by the genAI, demonstrating higher-order critical thinking skills. This critique may involve students comparing genAI generated work against a human-generated counterpart, or according to the assessment task’s evaluation criteria (i.e. rubric). Ultimately, the student will lead the assessment task as project manager, making all decisions in response to the information delivered by the genAI tool.

Other examples of cooperative assessments include the assessment design tasking students to utilise genAI to produce foundational content on a particular topic and/or theme. Such content could include datasets, draft diagrams/images, and first drafts of paragraphs. Students would continue to work the genAI generated product towards their own original final outcome.

Suggested Level Descriptors: Cooperative assessment

Poor	AI use moves beyond the authorised use as set out in the task requirements Task outcome is incoherent or the sections produced are not effectively integrated AI use is lacking or unclear, or does not effectively deliver the required outputs
Good	Specified student and AI contributions align with authorised use Student has managed the AI effectively to deliver the required contributions Contributions are clearly identified and complementary within an integrated whole
Excellent	While independently produced in line with the assessment brief, elements of the task outcome are presented as a coherent whole Assessment task development, incl its parts, has been skilfully and effectively managed Student contributions include analysis of differences between assignment sections, and these are evaluated, described and/or resolved as part of the outcome

The purpose of collaborative assessments is to support and enable students’ social construction of knowledge through participation with others. Successful collaborative assessments may deliver similar learning benefits to cooperative assessments with the additional benefit of promoting students’ capacity to reflect (Xiao et al., 2008) and retain complex information through deep learning (Atman Ulsu & Yildiz Durka, 2022). They may also encourage an openness to diverse voices (Cabrera et al., 2002). This results from the inclusion of open-ended but focused tasks that require students to collaboratively and iteratively develop the final outcome. While an educator may provide a loose structure around activities to ensure students meet the intended learning outcomes of the subject, students ultimately determine how the final outcome is achieved. Collective decision-making, including allocation of roles, may explore and exchange ideas during the development of an assessment task outcome. This heavily relies on the quality of interaction between group members, in which relationship dynamics are nurtured to promote positive engagement and participation by all members. Here, the educator takes the role of a facilitator to support constructive and positive group interactions.

An example of collaborative assessments includes the design studio project - a common assessment task within design education providing students with enough information to commence their design process in response to involving an open-ended brief. Actionable tasks are student-led as steps are identified within a flexible workflow informed by highly collaborative social interactions such as sharing information and ideas student-to-teacher and peer-to-peer. This interaction is reciprocal and a critical part of the design process enabling students to navigate back and forth through the problem scape towards an optimal final outcome (Lawson, 2006; Schön, 1995). In a paired student-genAI scenario, the student will train the genAI’s responses towards a shared goal. The uncertain wicked nature of design problems requires students to work with genAI as a team, where the reciprocal dialogue engaged by both student and machine facilitates a solidification of what the final outcome will look like. Considering the development of architectural studios with a focus on machine learning for design, Caitlin T. Mueller suggests “As in fully human collaborations, I find that empathy and insights into the thinking of creative partners are critical to productive and innovative design outcomes. … I am interested in promoting curiosity-driven approaches that wonder why AI models generate what they do, rather than treating them solely as solution machines” (Broome, 2024). Other examples of collaborative assessments include brainstorming activities engaging students to work in tandem with genAI towards exploring generating ideas and responses to complex problems. Another example includes designing project proposals involving students and genAI to co-create a project vision, followed by engaging in a continuous feedback loop towards iteratively shaping the final proposal. The student and genAI are engaged in a cyclic exchange of information to develop the needed knowledge and skills for the final outcome.

Suggested Level Descriptors: Collaborative assessment assessment

Poor	Use of AI does not move beyond direction, student ideas are not expanded outside of initial or student-originated ideas Student has not developed an approach to work creatively in partnership with the AI, limiting the capacity for an original response to the task brief Task outcome is incoherent, such that human and AI contributions are independent or unbalanced, and/or the task focus is not sufficiently addressed
Good	Both student and AI contributors have expanded initial perspectives through collaboration Student has effectively ‘trained’ the AI through iterative prompting or other development approaches to deliver useful contributions to the final outcome Task outcome combines both human and AI contributions for a coherent outcome
Excellent	Student has both ‘trained’ the AI, and learned from its responses, to deliver an original and creative response to the task Contributions of the student and the AI are balanced and integrated Assessment task development has been directed by the student, and has been responsive to emergent opportunities and directions throughout

Individual Assessment

Cooperative Assessment

Collaborative Assessment