Teaching Judgement in an age of Confident Machines

AI can produce an answer in seconds; deciding whether that answer deserves to leave the building still requires somebody with judgement.

At the RegRisk Control Forum in London, a slide headed “Bridging the ‘Trust Gap’” brought together a succession of developments in AI policy, regulation and governance. It was quite a reading list; the sort that makes you briefly nostalgic for a regulatory update with only three attachments. That pace of change makes judgement an essential part of preparing people for work. For further education, it raises a practical question: are we teaching learners how to recognise when AI has produced something beautifully presented and completely unsuitable? Colleges face this question from both sides; they are preparing learners to use AI in the workplace while deciding where it belongs in their own teaching and administration. The habits they teach and the practices they adopt will help determine what trustworthy use looks like.

Trust needs more than a demonstration

There is a particular charm to a good technology demonstration. Everything works, the example is obligingly straightforward and nobody has yet uploaded the wrong document. The difficult questions tend to arrive afterwards. Who checks the output? What are they checking it against? Can they reject it, and what happens when the person affected believes it is wrong? Those questions need answers before trust becomes a reasonable expectation, and learners have questions of their own.

Jisc’s Student perceptions of AI 2025 report describes learners calling for clearer guidance, fairer access and a greater say in how AI is integrated into education; it also highlights concerns about losing skills through overreliance. Using the technology, it seems, does not settle every question about its place in learning. For colleges, clarity needs to reach the assignment brief. A learner should understand what assistance is permitted, what must be acknowledged and what they are expected to demonstrate themselves. Staff need enough shared understanding to explain those boundaries consistently; a policy cannot do that work entirely on its own, however impressive its contents page.

The answer is where the lesson begins

Clear rules give learners a starting point; teaching them to apply judgement is the next step. Prompt writing is useful, but I would like to see more attention paid to the moment after the answer appears. Imagine an apprentice asking AI to draft advice for a customer. The response is fluent, reassuring and wrong on a crucial point; a combination that would be rather less endearing in a real customer interaction. Ask the learner to find the error, check a reliable source and explain the correction. Then ask what might have happened if the original advice had been sent. That exercise connects AI use with subject knowledge and professional responsibility; it gives the learner a reason to look beyond the polished wording.

The same approach can work with an unsupported recommendation or an answer that quietly assumes information it has never been given. Learners can practise identifying the gap, seeking clarification and deciding whether they have enough evidence to act. Checking takes knowledge. We should be careful about promising that a tool can save learners the effort of acquiring the very understanding they will need to supervise it. Staff deserve protected time to practise this too, using realistic examples and discussing difficult cases. Attending a demonstration and being ready to challenge an output are different stages of learning; the training budget should recognise both.

Give the human something to do

Training gives people the knowledge to challenge an output; the institution must also give them the opportunity to act on it. “There will be a human in the loop” is reassuring until you ask what that human is actually allowed to do. If they lack the time to check, the knowledge to challenge or the authority to override, their presence may be doing more for the presentation than for the decision. DfE’s guidance for England says intended uses of generative AI should be specified, with clear benefits that outweigh the risks; it also emphasises professional judgement and the relationship between teachers and pupils.

That calls for decisions about particular tasks. Helping a lecturer develop ideas for an exercise raises different questions from using a system to inform a decision about an individual learner; approving the product should not quietly approve every possible use. Consider a hypothetical tool that flags learners who may disengage. A college should ask what informs the flag, how errors will be identified and who will speak to the learner. Could patterns linked to disability or caring responsibilities be misunderstood? What would staff do if the recommendation conflicted with what they knew about that person?

These are useful questions for suppliers too. I would want to know what happens when the product makes a mistake, how changes are communicated and what support staff receive when they challenge an output. The demonstration can have its applause; those answers should influence the purchase.

Let learners question the system

A supplier’s account of the product is only part of that evaluation; the learners affected need a say too. As an autistic professional with ADHD, I am interested in how technology can make information and work more accessible, and whether the people expected to benefit have been invited to explain what actually helps. A learner who finds an AI-supported explanation useful should be heard; so should someone who finds it confusing or unsuitable. Accessibility needs room for individual experience, including experiences that complicate the success story.

Leaders can start with a defined use, agree what improvement would look like and involve the people affected in reviewing it. Count the time spent checking and correcting alongside the time saved; efficiency is much easier to announce before anybody has done the arithmetic. Taken together, these steps give the trust gap on that conference slide a practical answer: teach people how to check an output, give staff the authority to intervene and make sure learners can challenge decisions that affect them.

When a learner asks, “How do we know this is right?”, we should recognise a valuable professional habit forming. They may be the person who catches the mistake before it reaches the customer, the board or the regulator; personally, I would rather like them on the team.

By Mitzi Danielson-Kaslik MICA CG, Risk, Governance & Compliance