The AI Commons #4
Welcome back to the AI Commons. This fortnightly newsletter is funded by the Institute for Teaching and Learning and run by Mark Carrigan and Eva Parr. We're exploring how colleagues are integrating AI into teaching and research, with the aim of supporting learning across the university so we can approach this change in thoughtful, creative and collegial ways. If you'd be interested in contributing or have suggestions for future topics please contact [email protected]
📝 In this issue, Liz Birchinall, David James, Natalie Jayson, Karen Kilkenny and Andrew Rhodes (MIE) reflect on what happened when they stopped asking whether trainee teachers should use AI and started researching how they actually do:
Our UoM primary PGCE programme trains postgraduate students to become primary school teachers. When ChatGPT emerged in 2022, our academic team considered AI's potential impact on trainee development. We teach trainees the importance of creating their own lesson plans, resources and assessments tailored to the pupils, yet AI can now generate these materials very quickly.
We asked, "Should we allow our trainee teachers to use AI?" and realised that we could neither prevent nor reliably detect it. Our concerns were that trainees might shortcut their thinking by accepting AI outputs unchallenged, use resources that did not meet their pupils' needs, or fail to fact-check materials and end up teaching something inaccurately. We had two choices: ignore AI and hope it would go away, or embrace the challenge and investigate what it can and cannot do. We chose the latter, and our AI journey began.
In September 2024 we launched a three-year project exploring how trainee teachers used Teachmate, a generative AI teaching assistant. Trainees could use Teachmate during teaching practices to create resources and lesson plans. We chose it because it uses a closed model protecting trainee data and privacy, is used by over 228,000 teachers worldwide, aligns with the National Curriculum, and has won several educational awards. Four themes emerged from our first-year findings.
Context awareness: Trainees demonstrated strong awareness of their classes by adapting AI outputs to meet pupils' needs. For example, Teachmate can instantly generate differentiated texts at different reading levels so all children can access the same learning. Used in this way, AI supported inclusive learning.
Workload management: AI handled much of the heavy lifting of preparation and administration, freeing trainees to focus on creative, pupil-facing work. Trainees produced resources they would not normally have time to make. For example, one trainee wanted to teach new vocabulary to her pupils, so she used AI to create a song to help them learn. Another used AI to create a playscript for finger puppets to engage her class. A third wanted to teach body parts in Spanish and asked AI to create a picture of the Alhambra with Dalí-esque hands, eyes, legs, arms to engage and aid learning.
Fact-checking and subject knowledge. Effective AI use depended on trainees' own subject knowledge to judge accuracy of the resources. Although Teachmate's outputs were curriculum-aligned, generally reliable and age-appropriate, trainees learned to adapt resources rather than accept them unchallenged.
Prompt engineering: Trainees developed skills in writing clear and specific educational prompts. They learned that the quality of AI output depended on the quality of their thinking and instructions.
A strand running through all of this was professional judgement. We had worried that beginners might lack the experience to evaluate AI critically. In practice, using AI strengthened reflection and discernment. This pilot offers one model for researching AI integration in teacher education, and we hope it provides useful evidence for colleagues exploring similar questions.
💭 Should we stop talking about ‘AI’? And if so what should we replace it with?
The term "artificial intelligence" is remarkably unhelpful when we're trying to think clearly about what's happening in our universities. It groups together radically different technologies (social media algorithms, predictive systems, and large language models) as if they were interchangeable. It draws us into overblown rhetoric about civilisation-defining technology just over the horizon. And it leaves it fundamentally unclear what exactly we're discussing: the current generation of chatbots? Tools like NotebookLM that build on them? The "artificial general intelligence" that AI labs hint will arrive in a few years?
The technologist Jaron Lanier observes that "AI is a story we computer scientists made up to help us get funding once upon a time". It's now used to promote commercial products such that we contribute to viral marketing whenever we use the term uncritically. Perhaps more importantly, the shininess of "artificial intelligence" makes it harder to think critically about the limitations of what we're actually using. These are simply software products, with novel and impressive capabilities, but software nonetheless. Once we recognise that, the problems become more specific and more tractable. Instead of asking how we respond to "AI", we can ask how we deal with the rapid spread of consumer-facing chatbots among staff and students without any agreement about appropriate use.
👋 Are you using Copilot Chat?
We've had access to Copilot Chat for all students and staff since last year, yet there's been little discussion about what use we should make of this in teaching and learning. The fact that all students can access a capable chatbot through Microsoft Teams makes it far simpler to integrate interaction with chatbots into classroom activities. Are you already doing this? Are you planning to? Have you tried it but encountered difficulties? If so, we'd love to hear from you. Please get in touch: [email protected]
📚 Recommended reading:
We hope you enjoyed this fourth issue of the AI Commons. If you found it valuable, would you consider forwarding this newsletter to your colleagues? Comments, suggestions and questions are always welcome.