Designing an AI course at HKU at 24

I did not expect to be designing an AI course at HKU at 24. (I started at 23.)

That sentence still feels strange.

Partly because HKU is HKU (#1 in Asia in the QS)

Partly because I am still building inside the field I am supposed to help explain.

And partly because AI is one of the worst subjects to teach if the goal is to sound certain.

Everything moves.

Tools change.

Models change.

The examples get outdated.

The hype gets louder.

The fear gets louder too.

So the hardest part is not finding enough material.

The hard part is deciding what deserves to become a course.

There is already too much information about AI.

Too many tool lists.

Too many predictions.

Too many headlines.

Too many “AI will change everything” sentences that are technically true and practically useless.

Students do not need another person telling them AI is important.

They know.

Institutions know.

Parents know.

Employers know.

The harder question is:

That question is much less exciting than a trend report.

But it is more useful.

One of the clearest decisions from our course planning was that students need outputs.

Not just discussions.

Not just reactions.

Not just “what do you think about AI?”

Actual artifacts.

A design brief.

A stakeholder map.

A risk register.

Mitigation ideas.

A final showcase.

A fair-style pitch format where students can show what they built and how they thought.

That matters because AI education can easily become either tool training or opinion theater.

Tool training gets outdated.

Opinion theater stays too abstract.

The middle is harder.

Students need to make something.

Then explain the tradeoffs.

Then defend the assumptions.

Then revise.

That is where learning starts to become visible.

Another discussion was about adding a social theory lens to the assignments.

I liked that.

Not because theory makes the course sound smarter.

Because without a lens, AI becomes too smooth.

A model gives you an answer.

A tool gives you output.

A demo gives you magic.

But the world around the output is still messy.

Who benefits?

Who is excluded?

Who takes the risk?

What gets automated?

What gets hidden?

What does the tool make easier to ignore?

That is why the theory matters.

Not as decoration.

As friction.

A good lens should slow students down just enough to notice what the tool is doing to the situation.

The meeting also reminded me that the course is not just the syllabus.

It is Moodle.

Peer review.

Department feedback policy.

Mentimeter UID tracking.

Reading lists.

Blank weeks.

MP3 readings.

Slide style.

How much text goes on a slide.

How much structure is enough.

How much freedom is too much.

This is the part people usually do not see.

But it is where teaching becomes real.

A course is not only an idea.

It is an operating system for attention.

If the system is too loose, students get lost.

If it is too rigid, the subject dies.

If it is too tool-heavy, it becomes obsolete.

If it is too theoretical, it floats above the work.

That balance is the actual design problem.

This is why teaching helps me build.

Teaching forces clarity.

If I cannot explain something calmly, I probably do not understand it yet.

If I cannot turn an idea into an assignment, maybe the idea is still too vague.

If I cannot show students what a good output looks like, maybe I am still hiding behind big language.

That is uncomfortable.

But it is useful.

Building AI products pulls me toward speed.

Teaching AI pulls me toward structure.

I need both.

The builder sees what is changing.

The teacher has to decide what stays useful after the tool changes.

I do not want to pretend I have AI education figured out.

I do not.

I am still learning how much technical depth is enough.

I am still learning how to teach AI without making it feel like either magic or threat.

I am still learning how to bridge builders and non-specialists.

I am still learning how much theory helps before it starts making the work feel distant.

And honestly, that is why I want to write about it now.

Not after the course becomes a clean case study.

Now.

While the decisions are still messy.

While the structure is still being shaped.

While I am still close enough to the work to feel the confusion.

The Not Busy version of this is not to make AI education smaller.

It is to make it calmer.

Less noise.

More structure.

Fewer tool tricks.

More judgment.

Fewer abstract claims.

More student outputs.

Less “AI will change everything.”

More:

That is the kind of course I want to help build.

Not because I am far above the field.

Because I am inside it.

Building, teaching, making mistakes, changing my mind.

And trying to turn the noise into something students can actually use.

— Chris, still trying