Be in Control When Using AI Tools: Reflections from the CAS Online Community

If you were unable to join us for the Be in Control When Using AI Tools online community meeting, don't worry! You can catch up on all the content and a recording of the session below.

Key Takeaways

• AI literacy is about more than learning to use tools – it's about helping learners remain in control of how and why they use them.
• Developing students' agency means supporting them to make informed, intentional and reflective decisions when working with AI.
• AI systems introduce new challenges, including bias, over-reliance, persuasive outputs and opaque decision-making.
• Teachers have an important role in modelling thoughtful AI use rather than simply encouraging or prohibiting it.
• Designing learning experiences that promote questioning, verification and reflection can help pupils become more critical users of AI.

Be in Control When Using AI Tools

AI continues to dominate conversations in education, but this online community session took a slightly different approach. Rather than focusing on the latest tools or debating whether AI should be used in schools, the discussion centred on a much more fundamental question: how do we help young people stay in control when they use AI?

Led by Vidminas Vizgirda (Vid), a postdoctoral researcher working on the CHAILD (Children's Agency in the Age of AI) project at the University of Oxford and UCL, the session explored the concept of agency—our ability to make informed, deliberate decisions and act upon them—and why it matters when using generative AI.

The session began by asking participants which AI tools they and their learners currently use. Unsurprisingly, familiar names such as ChatGPT, Copilot, Gemini, Claude, NotebookLM and Canva featured prominently. This provided a useful reminder that, for many people, "AI" has become almost synonymous with generative AI, even though the field is much broader. The discussion briefly widened this perspective by highlighting the complex ecosystem behind modern AI systems, from training data and hardware to commercial deployment.

The conversation then turned to how widely AI is already being used by young people. Drawing on recent research, Vid shared evidence suggesting that AI use among children and teenagers has increased rapidly, with schoolwork consistently emerging as one of the most common reasons pupils engage with AI tools. Depending on the study, between half and almost all young people have now used AI in some form, demonstrating that this is no longer a future issue but a present-day classroom reality.

Participants were invited to share their own school policies on pupil AI use, revealing a wide range of approaches. Some schools actively encourage responsible use, others restrict access by age or only allow teachers to model AI use, while several attendees described policies that are still evolving or yet to be developed. This discussion highlighted just how varied the educational landscape remains, particularly when balancing safeguarding, age restrictions and practical classroom realities.

The heart of the session focused on agency. Rather than defining agency simply as independence, the CHAILD project describes it as a person's capacity to make, revise and carry out intentional decisions. In simple terms, Vid suggested thinking of agency as being in the driver's seat. AI should support our decisions—not quietly replace them.

This led into a lively discussion about the ways AI can reduce a learner's agency. Participants identified a range of challenges, including:

• Accepting AI-generated answers without checking them;
• Over-reliance on AI for thinking or creativity;
• Misunderstanding how AI systems work;
• Trusting outputs because they sound authoritative;
• Allowing convenience to outweigh careful evaluation; and
• Practical barriers such as age restrictions or unequal access to premium AI tools.

Vid grouped these into several broader themes. External pressures—such as deadlines, assessment pressures or peer expectations—may encourage learners to hand over decision-making to AI. Opaque systems can make it difficult to understand where outputs come from or why they were generated. Human habits can gradually shift towards accepting "good enough" responses without verification. AI bias can subtly influence ideas or writing style, while persuasive responses may encourage users to place more confidence in outputs than they deserve.

One particularly thought-provoking discussion explored the idea of bias. Rather than describing AI as simply "biased", Vid explained that all AI systems are necessarily biased in the statistical sense—they are designed to favour likely outputs over random ones. The more important distinction is between useful statistical bias and harmful societal biases, such as reinforcing stereotypes or undesirable associations. This nuanced explanation encouraged participants to move beyond simplistic narratives about AI bias and think more carefully about what kinds of bias matter in educational contexts.

Importantly, the session did not present AI as something to avoid. Instead, it explored practical ways of supporting agency.

Four key areas emerged:

• Developing accurate mental models of how AI systems work;
• Designing AI tools that encourage user choice and reflection;
• Creating supportive classroom environments where adults scaffold rather than take over; and
• Nurturing learner motivation so pupils feel ownership of their decisions.

Several examples demonstrated what this might look like in practice. One research prototype presents responses from two different AI models side-by-side, encouraging users to compare outputs rather than assuming there is a single correct answer. Another example, RealChat AI, guides users before, during and after prompting by encouraging reflection on prompts, outputs and possible limitations. These examples illustrated how thoughtful interface design can help learners remain active participants instead of passive consumers.

The session concluded with an engaging Q&A covering topics such as trust, safeguarding, age-appropriate AI use and AI literacy resources. One recurring theme was that we should not necessarily aim to maximise children's agency at every age, but instead provide opportunities that are developmentally appropriate. Even where younger children cannot use AI directly, they can still learn about how AI systems work, discuss their implications and develop critical understanding before becoming independent users.

Rather than offering simple answers, the session encouraged us to think more deeply about our role as educators. If AI is becoming an everyday part of pupils' lives, perhaps one of our most valuable responsibilities is helping them remain thoughtful, questioning and ultimately in control of how they choose to use it.

Next Steps

As you reflect on the session, you might ask yourself:

• How much opportunity do my pupils have to question AI outputs rather than simply accept them?
• Am I teaching pupils how AI systems work as well as how to use them?
• How can I model responsible AI use without encouraging over-reliance?
• Does my classroom encourage pupils to remain "in the driver's seat" when using AI?
• How well does our school's AI policy support thoughtful, age-appropriate use?

Some classroom activities to explore include:

• Compare responses from two different AI tools and discuss why they differ.
• Ask pupils to identify factual claims in an AI-generated response and independently verify them.
• Have learners improve an AI-generated answer by editing it for accuracy, clarity or personal style.
• Explore how different prompts influence AI outputs and discuss why prompt design matters.
• Debate situations where using AI is helpful and where completing a task independently may be more beneficial.

Further Resources

The following resources were highlighted during the session or are closely connected with the discussion:

BCS' AI Confidence - Free AI CPD

CHAILD (Children's Agency in the Age of AI) project

Stay up to date with the CHALID project

Barefoot's AI Zone

RealChat AI

Event Slides