This blog post describes the two interventions as technology probes (see the blog post Research Question and Research Methods), and demonstrates how they work.
5.1. GenAI Checklist
Inspired by the “keeping track” and “showing processes” aspects highlighted by multiple relevant documents (see the blog post Relevant Documents and Contextual Challenges), I created a GenAI Checklist for the Math&Stats unit. The aims is to provide detailed guideline on how to use AI coding tools in an academic context, and how “keeping track of processes” might look like in coding and programming with AI.
Specifically, it uses the following questions as template to log the interaction with Copilot:
- What did you ask the AI to do?
- What did the AI attempt? What was added to the code/notebook by the AI?
- What did you decide to keep, change, discard? What did you learn from this?
The checklist is rolled out as a mandatory submission element if any AI coding tools is used in the assessments.
5.2. “Rubber Duck” Chat Mode
For context: Visual Studio Code is the code editor used by students at CCI, it integrates an AI coding assistant GitHub Copilot (by Microsoft, the educational version is free for all students at UAL). A typical workspace is shown in the screenshot below, the left panel is for writing and running codes, the right panel is Copilot’s chat area. Copilot have access to the entire workspace, files, and it’s able to directly modify/add/delete the code in the left panel.
Customised chat mode is a feature in Copilot that allows programmers to tailor the behaviour of Copilot to fit specialised roles (Visual Studio Code, 2026). In practice, programmers create a set of “pre-task instructions” to define how the AI should operate. Quote from Visual Studio Code’s documentation:
“For instance, a planning mode could instruct the AI to collect project context and generate a detailed implementation plan, while a code review mode might focus on identifying security vulnerabilities and suggesting improvements.”
I create a “Rubber Duck” mode that instructs Copilot to support learning, break down the task, communicate and negotiate plans with the user, implement the solution step-by-step, and always prompt the user to review what was done and make changes. A full configuration file can be found here (requires UAL login).
The image below shows a side-by-side comparison of the default chat mode and the Rubber Duck chat mode. Rubber Duck offers a more detailed account of how to approach the task, how the approach maps to knowledge, what will be done, and it negotiates with the user about how to proceed.
These two interventions are rolled out in Math&Stats unit’s code repository.
Next step
The next blog post 6. Project Findings and Reflections presents my findings, reflections, and some directions for future works.
Reference list at: https://jaspersz.myblog.arts.ac.uk/2025/11/20/arp-references/