Teaching students to work with AI

AI assignment ideas for HR courses

HR courses teach decisions about people - hiring, pay, discipline, dismissal. That’s the one domain where handing judgment to a model is both most tempting and most consequential. Here are ten assignments built around that tension.

There’s a useful coincidence in teaching HR right now. The professional standard your students will be held to - don’t let an automated system make a decision about a person without a human who can explain it - is exactly the habit you’re trying to build in them as learners. The course content and the AI-readiness content are the same content.

Conversations people avoid

Most HR failure is a conversation someone postponed. These are hard to practise because nobody wants to role-play them in front of classmates, which is precisely why an AI counterpart helps.

1. The performance conversation that isn’t landing

The employee disagrees with the assessment, has a plausible counter-explanation for each example, and gets defensive. The student has to stay specific, stay on behaviour rather than character, and leave with an agreed next step. Grade whether they got concrete or retreated into vagueness.

2. The grievance with two sides

Two employees, two accounts, both internally consistent. The student interviews each separately and has to decide what to do on incomplete information - then state what would change their conclusion. Models are poor at holding “I don’t know yet”; students need to be good at it.

3. The dismissal

Documented, justified, and still awful. Grade on whether the student followed process, stayed factual, and treated the person decently while doing something painful. Nobody teaches this and everybody eventually does it.

4. The accommodation request

A request that’s reasonable, expensive, and sits in a grey area. The student has to work out what the obligation actually is, what’s merely good practice, and where the line is - with a manager pushing back on cost.

Decisions with a model in the loop

5. The screening tool that’s wrong

Give students a shortlist produced by an AI screening tool, plus the underlying applications. One strong candidate was filtered for a reason that’s legally and ethically indefensible - a gap in employment, a non-traditional credential, a proxy for something protected. Find it, explain it, and say what you’d change about the process.

6. Justify it to the person it affected

A decision has been made with model assistance. The affected employee asks why. The student has to explain it in terms the person can act on - which quickly reveals whether anyone understood the basis of the decision. If they can’t explain it, that’s the finding.

7. The pay-equity analysis

Students use AI to run a compensation analysis on a dataset, then have to identify what the analysis can’t see - tenure effects, role-level differences, the fact that the comparison group was built by the same process being audited.

8. Write the policy, then break it

Students draft an internal AI-use policy for a fictional employer. Then they’re given three edge cases the policy doesn’t cleanly handle and have to revise. Writing a policy is easy; writing one that survives contact with reality isn’t.

Negotiation, which is half the job

9. The compensation negotiation

A candidate with a competing offer, a band the student can’t exceed, and flexibility on things that aren’t salary. The student who only discusses base pay loses someone they could have kept. More on structuring negotiation exercises.

10. The manager who wants to skip the process

A senior manager wants someone moved out quickly and is impatient with the procedure. The student has to hold the line without making an enemy. Grade on whether the process survived and whether the relationship did.

Working through it as a team

HR decisions are rarely made alone, and the deliberation is where students see how differently reasonable people read the same facts. Have small groups take private positions on a grievance case, reveal them, then discuss - the spread is usually wider than any of them expect, and defending a position to peers is harder than defending it to an AI. Structuring that.

Common questions

Is it appropriate to simulate a dismissal with AI?

It’s more appropriate than the alternatives - reading about it, or learning on a real person. Brief the character as an employee with a coherent perspective rather than a caricature, and debrief properly afterwards.

What about employment law, which varies by jurisdiction?

Set the cases in your jurisdiction and say so in the brief. This is also a good discipline-specific failure mode to teach: models will confidently apply the wrong jurisdiction’s rules, which is exactly the error a new practitioner makes.

How do I assess something this qualitative?

The transcript makes it more assessable than a written case analysis, not less - you can see what the student asked, where they got specific, and where they backed off. Score against a short behavioural rubric rather than an outcome.

Does this work for management courses generally?

Most of it, yes. Numbers 1, 2, 3, 9 and 10 are general management skills that happen to live in the HR syllabus.