Building Buoyancy: AI Action Research with the ICGS

When I began my action research project with the ICGS, I wanted to understand whether a custom AI research coach could support my students’ academic buoyancy. What emerged was an unintentional and more nuanced exploration of the conditions that shape students’ choices as they learn.

You can read the full paper here.

Harrison.N_Building_Buoyancy.pdf

Download PDF • 255KB

When analysing the data, what I found most interesting was the conditions that shaped how students engaged with AI. Their willingness to question it, rely on it, persist with it or turn away from it was shaped by how they saw themselves as learners, how comfortable they felt seeking support, and how much trust they placed in me as their teacher and in the tool. At the same time, each interaction with the coach appeared capable of reinforcing or unsettling those beliefs.

I’ve come to see this as a trust loop, a concept I described in a previous post. It captures the reciprocal relationship between students’ beliefs about themselves, their learning environment and AI. These beliefs shape how students engage with AI, but their experiences with AI also feed back into those beliefs.

The loop involves three connected domains:

• Trust in themselves as learners: Do students believe they can manage uncertainty, persist through difficulty and make sound judgements?

• Trust in their teacher: Do they believe help is available, feedback is useful and uncertainty can be expressed safely?

• Trust in AI: Do they understand what AI can contribute, where it may be unreliable and when its output needs to be challenged or verified?

These domains influence whether students use AI to support their thinking or replace it. A student with confidence in their own capability and access to responsive teacher support may use AI selectively. They may ask for prompts, feedback or alternative perspectives, then apply their own judgement. A student who doubts their capability or feels unable to seek help may turn to AI for immediate resolution. This can reduce the cognitive struggle needed to develop understanding and reinforce the belief that they cannot complete the work independently.

The loop can therefore move in different directions. A well-designed AI interaction may help a student take a manageable next step and strengthen their self-efficacy. That increased confidence may lead to more critical and purposeful AI use. Conversely, repeated reliance on complete AI-generated answers may reduce opportunities to experience mastery, weaken ownership and increase future dependence.

Importantly, the trust loop is not about encouraging students to trust AI more. It is about calibrating trust across all three domains. Students need sufficient trust in themselves to question AI, sufficient trust in their teachers to seek human support, and sufficient understanding of AI to recognise when it is helping them learn and when it is allowing them to avoid the learning. Engagement with AI is both an expression of existing trust and a process through which trust continues to develop.

My paper explores this concept and considers what it might mean for the design of AI-supported learning that strengthens students’ academic buoyancy.