2025 - Lucy Macnaught: Frustrated by ChatGPT? Using LCT translation devices to design AI agents

Описание: As educators we face the reality of pervasive AI use, such as students using GPTs as writing assistants. To have better control of GPT responses, we can create AI ‘agents’ (customized chatbots) with specific performance instructions. The good news is that we don’t need to be tech geniuses fluent in specialized programming languages to create them. However, for giving feedback on draft writing, AI agents work best when they have a specific role with defined parameters. Deciding these parameters and creating instructions for how we want an agent to perform is where LCT is very useful.

In this Roundtable, I’ll share the backstory of the reflective writing agent that I’m currently developing. This involves my experience with creating translation devices for analysing epistemic-semantic gravity within the dimension of Semantics. I’ll touch on why creating them was useful for better understanding blog critiques in teacher education and for expositions in ancient history. I’ll then discuss my current challenge, namely, working out what constitutes an ‘advanced’ reflection by nurses already working in their field. You’ll see my in-progress translation device and how it informs the agent that I’m building through Cogniti, an award-winning platform from the University of Sydney. To wrap up, I’ll highlight how using text examples, from a translation device for a specific data set, are exactly the kind of ‘fodder’ that AI agents need to function well. Choosing one concept within one LCT dimension also ensures agents have a clear purpose.