There’s a gap opening up in vocational education that’s becoming difficult to ignore. Generative AI can help students produce better work. It can improve the structure, polish the language and fill in the gaps. But a stronger-looking submission doesn’t necessarily mean stronger learning has taken place. In fact, the evidence seems to point in a very different direction.
That distinction matters everywhere, but it matters enormously in our Australian VET. Our assessment decisions are supposed to establish that a person can actually perform to the required standard. Not that they can produce an impressive document. Not that they know how to prompt an AI tool. And certainly not that a chatbot can assemble something resembling a competent response on their behalf.
The question hasn’t changed: did this person demonstrate the skills and knowledge required to meet the competency?
What has changed is how easily a convincing piece of work can now hide the answer.
Polished work is a reason to look closer
Fresh research published in Communications Psychology highlights what is sometimes called the learning-versus-performance gap. Generative AI can improve a student’s immediate output while allowing them to outsource the mental effort that develops lasting competence.
In other words, the work gets better while the learning may not. For RTOs, this should change how we interpret polished written submissions. A strong document can still be useful evidence, but it should increasingly be treated as a signal to investigate—not proof of competence on its own. Ask the student to explain their reasoning. Observe them performing the task. Give them a variation they haven’t rehearsed. Have them apply the skill in a realistic workplace situation. That isn’t what I would think is a radical departure from good assessment practice. It’s the same direction ASQA’s AI guidance has been pointing. That is – use multiple forms of evidence, establish authenticity and keep the competence decision grounded in what the learner can actually do.
If a learner can’t explain or reproduce the thinking behind a polished answer, the polish tells us very little.
Use AI simulations for practice, but not the final decision
AI-powered role-plays, avatars and simulations have plenty to offer. Here is the introduction video to just one example that we currently use:
They can give learners more opportunities to practise difficult conversations. They can create a relatively safe environment for making mistakes. They can also provide immediate prompts and feedback without requiring a trainer to be available for every attempt. That is valuable—especially when trainer time is limited.
Overseas universities are already experimenting with virtual reality and AI-generated “students” to help teachers rehearse classroom situations. Similar approaches could be useful across VET, from customer service and supervision to interviewing, conflict management and safety conversations. But rehearsal and assessment are not the same thing.
Recent Australian VDC commentary, drawing on ASQA case material, restated the boundary plainly: AI tools can support practice, but assessment still requires human judgement. That line is worth protecting.
Use simulations to help learners build confidence and get more repetitions. When it is time to make a competence decision, retain meaningful observation, professional discussion and realistic role-play involving an appropriately qualified person. A simple test is this: if learners can game the simulation without demonstrating the underlying skill, it is not reliable assessment evidence.
The AI features are arriving whether you asked for them or not
RTOs do not need to go shopping for an experimental AI platform to encounter this issue. The technology is already being added to the systems many providers use every day. Canvas is moving tools such as IgniteAI and Knowledge Chats into early access. Moodle has introduced a pluggable AI provider model that can connect with services including OpenAI. These features may be useful. They may help learners find information, assist teachers with routine tasks or make course content easier to navigate.
But switching them on is the easy part. Before enabling any AI feature, an RTO should be able to answer three fairly boring—and extremely important—questions:
- Where does the student data go?
- Can we see and record what the AI did?
- Who remains accountable for the assessment decision?
If the answers are unclear, the feature is not ready for unrestricted use. The chatbot may be new. The responsibilities around privacy, evidence, validation and accountable decision-making are not. Features are easy. Governance is the real work.
Employers are already buying practical AI skills
There is also a commercial opportunity here, and it is not limited to overseas technology companies.
Australian employers are already looking for practical “AI at work” capability. TAFE Queensland has AI-for-productivity workshops available. Victoria University has launched an eight-week Applied AI for Business program with Victorian Government backing. Nationally recognised AI qualifications at Certificate IV, Diploma and Advanced Diploma levels are appearing on the scope of multiple RTOs. International developments matter as well. OpenAI Academy is building workplace learning pathways, while Microsoft continues to put stronger AI models into Copilot—the environment many corporate clients already use.
The message from employers is fairly straightforward. They want staff who can use these tools productively and responsibly. They need people who can draft efficiently, check accuracy, protect sensitive information and make sound decisions about the output. Most do not need every employee to complete six months of AI theory before applying the technology to an ordinary workplace task. That creates space for short, practical offerings built around real work:
- using AI to draft and improve routine documents;
- checking AI output for errors and unsupported claims;
- protecting confidential or personal information;
- using AI within organisational policies;
- recognising when human judgement must take over; and
- documenting how important decisions were made.
Short, responsible-use courses will continue to sell because they solve an immediate workplace problem. If an RTO’s entire AI position is “don’t use ChatGPT”, it will miss both the learning challenge and the market demand.
Trainers need a house rule they can actually use
Trainers are currently caught between rapidly changing tools and assessment obligations that have not gone away. Recent Australian research involving newly qualified VET trainers found that GenAI-generated drafts could look plausible while still being unsafe to use without careful checking against the unit requirements and assessment conditions.
That finding will surprise very few experienced practitioners. AI is extremely good at producing something that looks like a training resource, an assessment question or a marking guide. It is less dependable at ensuring every performance criterion, knowledge requirement, condition and evidence requirement has been interpreted correctly. When internal guidance is unclear—and staff have uneven access to approved tools—trainers end up developing their own rules one assessment at a time. That is inefficient for the trainer and risky for the RTO.
A practical house rule can be much simpler. AI may assist with the draft. A qualified person still owns the decision. That principle can apply to learning resources, assessment materials, feedback and administrative tasks. It gives staff permission to use the technology while keeping professional accountability exactly where it belongs. Pair the rule with a short, practical AI literacy session. Show trainers what the tools do well, where they regularly fail, what information must never be entered and how to check a draft against the training product and assessment conditions.
Staff should not need to invent organisational policy every time they open an assessment.
The standards still apply
ASQA has not created a separate AI Standard. It is asking RTOs to apply the standards and assessment principles they already have to an environment in which GenAI exists. That may be less exciting than announcing an AI transformation strategy, but it is the work that matters. Over the next year, the strongest RTOs will not necessarily be the ones with the flashiest chatbots. They will be the providers that have:
- clear and workable rules for student and staff use of AI;
- assessment practices that can still establish authentic competence when GenAI is available; and
- short, commercially useful AI-at-work programs that employers can purchase without committing staff to a 12-month enrolment cycle.
The technology will keep changing. The obligation to make defensible decisions about competence will not. Use AI for practice. Use it to help with drafting. Use it to remove low-value work where that can be done safely. But keep people accountable for judgement, and keep assessment focused on what the learner can explain, apply and demonstrate. That is not resistance to AI. In my view, it’s simply good VET practice.