Good feedback is key to helping students understand their progress and improve their work. But with increasing marking workloads, it can often be difficult to provide students the depth of feedback they need, in a style that supports and encourages them.
This is why educators Fran van den Berg and Minh Huynh from the University of Sydney, and Shu Hui Koh from Murdoch University, turned to Cogniti agents to support them and their marking teams. By designing agents that could expand upon brief feedback notes, these educators were able to save markers’ time and improve the quality of feedback that students received.
Steering AI towards expanding feedback comments
The system message was carefully crafted to have the Cogniti agents act as expert educators who wrote constructive and supportive feedback. The agents were given the assignment brief, some examples of well-written feedback, and rules to follow (such as to keep feedback expansions short, to be constructive, to be forward-looking, and provide suggestions).
The agents were also given resources such as marking rubrics, lab manuals, a marking guide, and other documents. This meant that the agents were able to draw on similar information that markers would draw on.
Human feedback, augmented
It was philosophically important to Fran, Minh, and Shu that human markers determined the grade and the feedback comments that students would receive. Their markers have the disciplinary expertise to make the right academic judgements about the quality of student work, and knew how to support students to improve.
The role of the Cogniti agents was to help the markers word their written feedback comments in a more fluent, consistent, and student-friendly way. Markers would mark the student reports, and use Cogniti to expand on brief feedback comments. They would then take the expanded comments, modify as needed, and provide this to students through the LMS.
Loved by markers
Even though it took a bit of time to get used to writing brief comments that could be expanded by the Cogniti agents, markers made much use of these. Many commented on the usefulness of the expanded feedback for students, and how much time it saved. One even commented that it helped to convert their frustration into nicer comments for students.
Shu is already building her next Cogniti agent for the next major assignment this semester, and Fran’s colleagues are looking to create similar agents for their subjects too. These feedback agents are examples of when generative AI can act as ‘augmented intelligence’, as long as it is steered and resourced in the right way – which is possible with Cogniti.