From digital pets to robot pals: how AI is changing Hong Kong classrooms

Supported by a government blueprint for digital education and a HK$500,000 (US$63,733) grant for each publicly funded school, schools are developing their own AI platforms or working with industry to adapt the technology to their students’ needs.

AI has helped teachers plan lessons and overcome barriers to learning. But teachers said assessing students’ AI-assisted work required more training, while principals called for closer collaboration between schools and industry to support pupils with special educational needs.

The Education Bureau released its Blueprint for Digital Education Development in June, requiring teachers to complete at least 30 hours of digital training every three years and directing schools to incorporate AI strategies into their annual development plans.

The bureau has also provided a one-off HK$500,000 grant to every publicly funded primary and secondary school through its AI for Empowering Learning and Teaching Funding Programme, launched in 2025. The money can be used to buy or lease AI software, hardware and platforms, and to fund activities that develop students’ AI literacy.

SKH Kei Oi Primary School has adopted both off-the-shelf and custom-built AI tools, initially focusing on science and English for Primary Four to Six pupils.

Ready-made applications include image generators that turn pupils’ English compositions into comics, as well as platforms that assess their speaking and writing skills.

Principal Szeto Kar-ying said finding suitable products in a crowded market was not easy, but the school’s guiding principle was to put teaching needs ahead of the latest technology.

When no suitable product is available, the school works with industry to develop its own solution.

With support from a programme run by the bureau and Cyberport, it partnered with a technology company to create an AI-driven science reading platform that guides pupils through texts and concepts.

“Funding goes first to projects that fit curriculum needs and can be sustained and scaled up. We also encourage our teachers to experiment with AI in everyday teaching and classroom activities,” Szeto said.

Teacher Cynthia Doo Chi-yin’s budgeting game, in which her Primary Two humanities class manages a digital pet’s monthly spending, is one example of teachers developing their own AI teaching tools.

“I make the spending choices together with them in class. They are excited to explore it first-hand and discover on their own that irrational spending has consequences, affecting the character’s mood,” she said.

While AI has opened up new possibilities for lesson design, it has also made assessing students’ abilities more challenging.

“If we only assess the final product, students can easily get AI to do the work for them. Assessment should value the process over the result and cover how students collaborate with AI, not just test their knowledge,” said Chan Yuk, the school’s IT coordinator.

He said this approach added to teachers’ workloads, and schools needed to provide training and time rather than leaving staff to work it out themselves.

Dragan Gasevic, chair professor in the Faculty of Education and the School of Computing and Data Science at the University of Hong Kong, said assessments now needed to shift their focus from the final product to the process behind it.

“When a sprinter races, the coach does not just record the finishing time. They replay the whole race step by step to find the moments where performance could improve,” he said.

“We need tools that let teachers and students replay these steps too, and identify the points of weakness when students interact with AI.”

Such tools should be co-designed with children and teachers, he said, with students given control over their own data and schools fostering a coaching culture.

Gasevic said schools needed to distinguish AI designed for education from general-purpose AI.

“Real learning gains require specialised adaptive systems, not simply asking ChatGPT or Gemini to run a Socratic dialogue. These systems, built on more than 50 years of research in AI in education, need components that adapt to the individual needs of each student,” he said.

For students with special educational needs, AI can help overcome learning barriers by responding to their emotional needs and offering alternatives to typing, such as pictures and voice input.

Hong Chi Winifred Mary Cheung Morninghope School, which caters to children aged six to 18 with intellectual disabilities, has introduced various robots as learning companions.

Principal Chu Pui-man said the school had invested in smart devices and AI platforms to immerse students in technology and help them become confident using it in daily life beyond school.

“We bought two humanoid robots that students can learn to programme, and they once led the whole school in morning exercise,” she said.

“In our classrooms we keep more than 10 companion robots, our AI learning buddies, which chat with students when they feel unhappy and comfort them.”

The school has also partnered with a technology company to create an AI avatar that serves as an assistant on students’ tablets. Because commercial products do not meet all its students’ needs, it maintains an in-house team to develop a school-based platform.

“Our students might not speak fast or clearly, so we need to adjust the platform to their pace, guide them step by step and not rush ahead, and allow them to enlarge the text. We also preset question choices on the platform for students who cannot type,” she said.

She added that AI also allowed students to ask questions through images or speech rather than using a conventional keyboard.

But customising and trialling AI products took additional time and money.

“We hope to see cross-school resource sharing and centralised procurement for special education, so schools do not each end up buying the same thing at a high price,” she said.

“If the tech industry could recommend well-tested products and explain why they suit our students, we could save time searching and customising on our own.”

The task of matching products to schools falls partly to Hong Kong Education City (EdCity), the government-funded education portal.

Launched in 2000 to promote the digitalisation of education, EdCity has evolved in recent years from primarily a repository of digital content into a platform connecting technology companies with schools as AI use grows.

Its executive director, Dr Tenny Lam Fung, said schools could draw on AI solutions from around the world. For mainstream schools, the challenge lay in choosing the right ones.

“Mainstream schools have many choices. The question is which one to pick, how to use it, and how to get clear feedback. We provide a platform, EdMarket, for schools and companies to share their successful cases and help schools choose,” he said.

For special schools, the challenge worked both ways: schools had fewer suitable products to choose from, while companies struggled to reach the right classrooms.

“Students with special needs have many different learning needs, and it is very hard for one solution to meet them all. So the challenge for special schools is how to find the right solution for each application context,” he said.

He added that EdCity had stepped up support for special schools, including through an AI in Education Forum co-organised with the Education University of Hong Kong, at which special education researchers present their work.

Lam described the digital education blueprint as an important milestone in setting the overall direction but said choosing solutions was a matter of implementation for schools and the Education Bureau.

“The blueprint sets out the framework, principles and policies. It will not go into detail on how to choose solutions. That is where execution comes in, and that is the role we play,” he said.

“In execution, things must keep evolving as technology updates. But some principles will remain unchanged: people first, technology for good and using technology responsibly.”