AI in education: A system that learns with its students

‘As a public education system, we are strongest when we listen – to our data, and to our teachers and students. In embracing EdChat, we chose responsiveness over rigidity, learning over waiting, and trust over control.’ In his latest column for Teacher, Professor Martin Westwell – Chief Executive of the South Australian Department for Education – shares the thinking behind a generative AI chatbot that has been custom-built for teaching and learning, and its initial impact.

AI-based education technology is a new frontier at the moment with all manner of untested options entering the market. I feel sure things will settle down eventually but, for now, the fear of missing out caused by the combination of enormous opportunity and the enthusiasm of early adopters, is putting a lot of frantic energy into the system. It’s a situation ripe for teachers and school leaders to be misled into products that don’t provide the value they claim.

As educators and leaders, we are used to seeking out evidence, piloting initiatives, and scaling successful practices. But when it comes to emerging technologies like generative AI, the traditional path of waiting for research to crystallise into a notion of ‘good practice’ will leave us constantly playing catch-up. The pace of technological change, and the ways in which students engage with it, often outstrip our ability to keep up with well-defined clear-cut practices.

Rather than waiting, in South Australian public education we took a different approach. We focused on learning together with our students. It strikes me that we don’t do enough of this. While we’re quite good at learning from people who are identified as experts such as researchers and leaders in the field, we sometimes forget about the expertise of our own teachers and leaders, and even more frequently we neglect the insights that our students have from their daily experience of school. There’s an enormous amount we can learn from each group.

In 2023, when others were talking about banning AI in schools, we introduced EdChat, a generative AI tool, into 17 high schools across both metro and country South Australia starting with early adopters in the metro area and then expanding out to include more. We didn’t prescribe strict usage rules or define narrow objectives. Instead, we launched EdChat with guardrails to ensure safe and ethical use, and then gave students and teachers the freedom to explore, adapt, and respond. This was a deliberate step in treating our public education system as a learning organisation – one that learns from within, from the real experiences of those at the heart of education: students and educators.

Already the data tell us that generative AI is not just a tool; it can be part of a broader shift towards self-organising learning environments where students have agency, curiosity, and the means to pursue knowledge in ways that make sense to them.

Students leading the way

From the earliest stages of the proof of concept, students embraced EdChat with enthusiasm. Over 4,400 students have become active users since the initial rollout in July 2023. Importantly, 94% of their interactions were curriculum-related, with strong engagement across English, Science, and Mathematics. Even more revealing than the numbers was the way students used EdChat – not simply to retrieve information, but as a thinking partner.

Take, for example, a student who was grappling with the concept of narrative structure. By asking EdChat, ‘What is a narrative arc in easy terms?’, they received not only a clear explanation but also a way to make meaning. EdChat described it as a ‘rainbow shape’ representing the journey a story takes – beginning, rising action, climax, and resolution. This wasn’t just helpful; it was transformative for that student who was ready to learn at that moment.

In Science, another student explored variables affecting bacterial growth, prompting EdChat to help them brainstorm factors like temperature, pH, and nutrients. Rather than simply providing facts, EdChat engaged them in cause-and-effect reasoning, encouraging deeper exploration and learning.

One student got stuck with their work in class and had their hand up, waiting for the teacher to come round to them. Instead of pausing, she asked EdChat to help her move forward and just checked in with the teacher when they became available, saving time for student and teacher without compromising on quality. 

The use cases keep on coming: teachers and newly arrived students have used our generative AI to help overcome language barriers and there are many ways in which neurodiverse students are finding learning with AI easier than traditional modes. 

Importantly, these examples give us insight into how students naturally use EdChat in ways that are intelligent, creative, productive and independent.

A system that learns with its students

Our goal wasn’t to dictate how EdChat should be used, but to observe, listen, and learn; ready to amplify the positive and dampen the negative. This is the mindset of a learning organisation – where feedback loops, authentic listening, adaptation, and inquiry are central. 

Rather than assuming that best practice in AI-assisted learning could be determined in advance, we allowed good practice to emerge from within our system, shaped by students' experiences. This has proven to be not only practical but essential, given how rapidly AI technologies are evolving and how unpredictable their adoption can be.

We have learned that students want tools that are flexible, responsive, and always available. For instance, while the majority of EdChat’s use happened during school hours, usage continued into the evenings and on weekends. 

Quite quickly we could see how EdChat is more than a classroom tool and is becoming embedded in students’ study habits, accessible when they need it most, including outside traditional learning environments.

This kind of self-directed learning is exactly what we hope to encourage in students preparing for a future where adaptability is key and the next generation of AI tools can support learning.

The depth of student thinking

A key insight from the EdChat pilot was the depth of thinking students demonstrated through the quality of the questions they asked. In a focused study on student interactions about the Opium Wars, researchers applied the SOLO Taxonomy to evaluate the complexity of questions students asked. For those that don’t know SOLO (Structure of Observed Learning Outcomes), it is usually used to classify learning outcomes in terms of their complexity. Instead of looking at student responses we looked at the prompts they used to infer what level of response they were looking for from EdChat.

The results? Just over one-third of the interactions were classified as complex (multi-structural or above), and all 14 students in the study engaged in complex questioning at least once.

For instance, one student asked, ‘What did the treaty signed after the opium wars tell us about the effects of British imperialism?’ – a question that moves beyond surface facts to consider broader historical implications and in doing so invites more questions.

This challenges the assumption that AI makes students passive consumers of information. Instead, it shows that when given the right tool, students can and do use AI to help them think critically, connect ideas, and explore meaningfully. 

Equity and variation

One of our biggest concerns was whether EdChat would be equitable, and whether students from all backgrounds would have the same opportunities to use and benefit from it.

The data give us cause for optimism. Students from both metropolitan and country schools engaged at nearly identical rates: each around 40%. Similarly, male and female students participated at comparable levels, though female students showed slightly higher intensity of use, particularly after school hours.

Unsurprisingly, there was significant variation between schools and, of course we are likely to initially see high variation between classrooms too depending on the teachers’ level of comfort with the technology or with being comfortable learning with their students.

This tells us something important: school culture, leadership, and local decision-making matter. Where EdChat was supported, explained, and encouraged, uptake soared. Where it wasn’t, engagement lagged, not because of technical barriers, but because of context.

As a learning organisation, this insight is gold. It means we can focus on the conditions that enable effective usage and further learning: professional learning, leadership engagement, and school-level trust in students to explore. A learning system activates leaders, teachers and students to innovate together and learn from exploring the spaces between them.

Next steps

I have focused here on our early experience of students’ use of generative AI. We’ve learned a lot, and we’ll continue as we started: as a system that learns, using research evidence as it emerges alongside the evidence of practice in our classrooms.  

There are 30,000 teaching and support staff at South Australian public schools and preschools who now have access to EdChat and around 40% are active users. We remain focused on using our AI capability and capacity on the most impactful use cases that support learning and teaching while reducing teacher and leader workload. 

This year, we are carefully working towards the rollout of EdChat to all secondary students in public education. Our objective is to understand how to best support our schools in their student use in key areas such as appropriate use, student wellbeing and learning. This is not about mandating use but about enabling conditions where learning and innovation can thrive.

We are alert to a great many of the risks and I’m sure we’ll be surprised by others along the way. Technology always brings new opportunities and risks and it’s our role to properly deal with both.

Final thoughts

As a public education system, we are strongest when we listen – to our data, and to our teachers and students. In embracing EdChat, we chose responsiveness over rigidity, learning over waiting, and trust over control.

The result is a living, evolving understanding of how AI can serve education, not just in theory, but in practice, every day, in classrooms all across our state.

We’ll continue to learn – together.