Evidence hub · Innovation · Impact · Depth

Selected pathway: Depth

The project is strengthening educational quality, engagement and responsible use within its current context rather than treating expansion alone as success. Its development cycle is deliberately continuous: test each capability, teach responsible use, gather evidence, revise and repeat with the next cohort.

Depth roadmap showing role-specific practice and testing over the next 12 months, followed by longitudinal evidence and a reviewed teaching library over two to five years.
Figure 1. A staged Depth roadmap. Immediate work strengthens practice and evaluation; repeated courses then build longitudinal evidence and reusable teaching knowledge.

The next 12 months

Three follow-on grants support deeper role-specific AI literacy across the connected ecosystem:

  • Students will practise checking multimodal conversations in ChatNextWeb, tool-using outputs in OpenWebUI and agent steps in Claude Code and Codex.
  • Teachers will co-design activities and decide which capability supports each learning outcome.
  • Teaching assistants will prepare for bounded GradePilot review through demonstrations, rubrics and feedback practice, with explicit limits on AI authority.
  • Agent workflows and GradePilot will be tested before deeper classroom use.

The next 2–5 years

Repeated courses will provide longitudinal evidence on understanding, confidence and responsible use. The team will build a reviewed library of educational tasks, failure cases and teaching responses. CUHK educators, professors from other universities and secondary-school teachers will continue as a focused community of practice. Success means that each cohort and teaching team uses evolving AI more critically, transparently and educationally than the one before it.

Responsible, transparent and human-centred use

Human judgement remains central at every stage. Students are taught that text, images, calculations, code and agent actions can be wrong and must be verified. Teachers decide whether and how AI fits an activity and assess understanding rather than unexamined output. Agent-generated changes remain subject to human review. Teaching assistants may use GradePilot’s optional AI Fill action for a draft, but can edit or reject it and retain every final grading decision.

Authenticated GradePilot teaching-assistant grading screen showing rubric fields, feedback, source and PDF panes, AI Fill and Save controls.
Figure 2. Human authority is visible in the workflow. Evidence, rubric fields, optional draft assistance and the separate final save action appear together. No grading action was performed for this demonstration capture.

Transparency is supported through visible model choices, inspectable tool traces, logged use for evaluation and tutorials on disclosure, limitations and appropriate use. Data safeguards include informed consent, anonymisation for analysis, restricted access and limited retention, with additional review for sensitive contexts. Free access addresses economic inequity; role-specific AI literacy addresses capability inequity.

Continuous responsible-improvement loop connecting usage dashboards, surveys, representative tasks, safeguards and revisions.
Figure 3. Safeguards are part of evaluation. Educational effects, cost, consent, privacy and access controls are reviewed together and inform the next iteration.

Resources, expertise and sustained support

Five CUHK/UGC teaching-development grants provide up to HK$1,801,660 in programme support:

Programme support Period Amount
Free Generative AI for All (TDLEG 4171166) 2024 foundation HK$461,660
Embedded-notebook AI grant (CDGS 4171173) 2024 foundation HK$100,000
Free AI Tools for All 2025–28 follow-on HK$700,000
Vibe Coding for STEM 2026–27 follow-on HK$300,000
AI-Assisted Formative Feedback and Grading System co-Developed and Benchmarked by Postgraduate Students 2025–28; approval-in-principle Up to HK$240,000

The newest grant is approved in principle at 80% of the amount requested, subject to submission of a revised proposal addressing the funding recommendations and reviewers’ comments. The follow-on grants fund future development and are not presented as evidence of impact already achieved in 2024–25.

Prof. Yangqian Yan leads a Physics–Statistics team supported by co-supervisors, teachers, teaching assistants and student helpers. Sustained support comes from funded staff and student time, educator oversight, shared institutional credentials, CUHK AI guidance, consent and privacy controls, per-user cost metering, weekly budget caps, surveys and annual review of models and tutorials.

Diagram of role-specific AI literacy for students, teachers and teaching assistants, with verification and human judgement at the centre.
Figure 4. Depth is role-specific. The support structure strengthens student verification, teacher design and TA review rather than imposing one generic measure of AI use.

Evidence hub · Innovation · Impact · Depth