AI is changing the way schools teach and students learn. University of Sydney Professor Danny Liu—a molecular biologist turned educator—chats to science communicator Ceana Priest about his career path to AI, and how it led him to create Cogniti, a platform that helps teachers “clone” themselves to better support student learning.
Danny Liu’s journey into artificial intelligence began far from computers—in his parents’ suburban garden in Sydney. There, he was fascinated by the hidden workings of life beneath the soil. This curiosity led him into molecular biology, then teaching and eventually to reimagining how AI could support educators in the classroom. “I remember growing potatoes and just the kind of magic of being able to put a spud in the ground and then a few months later collect 50 spuds. I was fascinated by the way the world worked.” After attending James Ruse Agricultural High School, Danny trained as a molecular biologist, earning a Bachelor of Science (Advanced) (Hons) in 2006 and a PhD in 2010 at the University of Sydney, followed by a Graduate Certificate in Higher Education from Macquarie University in 2016. Today, he is a Professor of Educational Technologies (Education Focused) at the University of Sydney—describing himself as a faculty developer by day, a programmer by night and an educator at heart. It’s here he created Cogniti, a platform that empowers educators to build their own AI-driven teaching tools.
Seeing AI on the horizon
During his PhD studies in 2010, Danny was “bitten by the teaching bug” while tutoring undergraduate students. After completing his doctorate, he spent four years teaching first-year biology before transitioning into academic development, helping teachers refine their practice, first at Macquarie University and later at the University of Sydney.
It was in Sydney that Danny first became aware of GPT3 in mid-2022, about six months before its public release. “I was like: ‘This is curious. This is a machine which can generate human-sounding language on command.’ And, so at that point, I got a bit hooked. I realised, sitting around with a few colleagues, that this was going to shake up education…in the short term because it meant that students now had a machine that could write things for them.”
Historically, he says, higher education has relied on students producing artefacts, such as essays or projects, to demonstrate their learning. So, the arrival of AI represented a significant shift in how learning could be expressed and assessed.
“I realised that there was a lot of potential in [AI] if teachers had the ability to control it. Because at the time, there was this moral panic around academic integrity and what we do about the future of writing and those kinds of things. [There was] this kind of massive mistrust between students, teachers and the AI, and so at that point we were like: ‘Well, GPT4 is powerful. It’s actually quite controllable and quite steerable.’”
Cogniti
Danny decided to build a platform that teachers could lead, which included equitable access for students, built trust between teachers and students, and supported meaningful learning. This led to the creation of Cogniti, an online tool that lets teachers build custom chatbot “agents,” effectively cloning themselves and extending their presence to students 24/7. Currently in pilot, over 100 schools and universities worldwide are using it.
Danny likens the agents to stunt doubles in movies: they don’t replace teachers, but step in to handle tasks that would be impossible to handle alone. Just as a stunt double allows an actor to perform safely, AI agents can guide students, reinforce lessons or focus on specific parts of the curriculum—acting as a customisable stunt double to enhance classroom learning.
Cogniti runs on OpenAI’s large language models, is hosted on Microsoft Azure and uses a vector database to let teachers provide subject-specific materials—like sample reports, feedback or rubrics—so agents stay grounded in course content. Teachers can adapt agents shared by other educators, use templates or create their own.
Agents designed by teachers
Danny emphasises that teachers need control over how AI agents interact with students, since they best understand the learning context. For example, agents can be set up as Socratic tutors that guide students to unpack their thinking and explore ideas, rather than simply handing over answers.
“Maybe there are certain ways that the teacher wants the students to learn, or maybe certain parts of the syllabus they want them to focus on, or pedagogical approaches which the teacher finds more helpful. They understand the discipline, they understand the assessments that are coming up. They understand their students, so they’re the place to design that interaction.
“I think the reason why [Cogniti] fits well [into the classroom]…is that it is driven by the teacher. It’s not driven by a tech company. It’s not driven by a vendor. It’s not driven by the principal. It’s driven by the teachers and their understanding of what their students and their class need.”
Cogniti allows people to build both teacher-facing and student-facing agents. A teacher-facing example might be creating well-structured multiple-choice questions or assisting with the development of time-consuming rubrics. Or, for students, helping role-play responses to microaggressions and racism, or personalise exam preparation. Other classroom examples are available here.
For a Year 11 design technology teacher that Danny spoke to, she said her class of 30 students was working on a significant project, but she found it impossible to get around all 30 students in a one-hour class period. So, they created an agent to help speak with students at the start of class for five minutes. Danny says that agents like this could help unblock students before approaching their teacher with better questions and a better overall understanding of where they were.
“And so that’s the idea,” Danny says. “The stunt double. She couldn’t be physically there for all her students at the same time. But the AI can. And then after that, the conversation in class becomes much richer.”
Becoming aware of misconceptions
Anonymised student conversations with their agents are available for teachers to review, enabling teachers to identify areas where additional support is needed and gain insight into student misconceptions. “Students are much more likely to tell a robot what they think and ask those silly questions which they might otherwise be embarrassed to ask a teacher,” says Danny. “We’ve heard from academics and teachers that it helps them to appreciate what students get and don’t get, and then either adjust what the agent does or change what they do in class to meet the needs of students.”
The future of educational AI
Danny sees the future of AI in the classroom as a tool to help teachers maximise their time with students, allowing teachers to focus on what they do best while AI supports and extends their efforts. He acknowledges that many teachers worry AI might replace them, but stresses that its role is to support, not substitute. The true strength of education, he says, lies in the relationships teachers build with their students. Although those relationships can often become strained as teachers become stressed and overworked, Danny suggests that with AI now a powerful presence in classrooms, teachers can explore using it to strengthen these relationships.
“It’s about agency for the teacher to be able to control this AI and not feel like it’s just something that happens to them. If we can take AI from a standpoint that it is a cheating technology or that it is a technology that needs to be avoided because we don’t have any control over it, if we can change that narrative to AI is a technology that will partner with us, that we can steer and control to improve the care we can give to students and the relationship with them, then that’s the better view of it.”
Ngā kupu
Ākonga – student, pupil, learner
Akomanga – class, classroom
Kaiako – teacher, instructor
Koi te hinengaro – intelligent
Mōhio – to know, understand, realise, comprehend, recognise
Rorohiko – computer
Wharekura – house of learning, school
[Sourced from Te Aka Māori Dictionary]
Resources
Cogniti website
NZCER – AI in Primary Schools report
NZCER – Generative artificial intelligence in Aotearoa New Zealand primary schools—Teacher and student survey findings | 2025 | David Coblentz, Jessie Dong and Bronwyn Gibbs. read more