Assistance systems: using AI tools without skipping the learning

A modern cockpit is full of assistance systems that hold a heading, watch the fuel, and flag a conflict before the crew sees it. None of them fly the aircraft, and a large part of training exists to keep that distinction intact under pressure. Language models sit in the same place beside your studying: genuinely useful for a short list of jobs, and quietly destructive when they take over the part that was supposed to be the learning.

What these tools are actually good at

The clearest use is translation, not between languages but between registers. A paragraph that has defeated you three times can be restated in plainer words, with the jargon unpacked, and often that is enough to get you moving again. The same applies literally across languages: if you study in a second language, having a term explained in the language you think in is not cheating, it is removing an obstacle that has nothing to do with the subject.

The second use is question generation. Give the tool a chapter, ask for questions with definite answers, then close everything and answer them. You are still doing the retrieval yourself, which is the part that counts. The tool has only taken over the clerical work of turning headings into questions.

The third is the patient examiner. Explaining something out loud is one of the fastest ways to find your own gaps, and it usually fails for a mundane reason: nobody is available at eleven at night. A tool that will listen to a rambling explanation, ask what you meant by one term, and not get bored is a reasonable stand in. Ask it explicitly for follow up questions rather than for a verdict on how you did.

The fourth is sorting. Twenty pages of messy notes become an outline in seconds, and you will see at once whether the structure matches the course.

The work you skipped was the work

This is where it turns. If the tool writes the summary, you did not learn the chapter. Summarising is not an administrative step before studying, it is studying: deciding what matters, finding the words, discovering that you cannot explain the third section. Hand that over and you get a clean document and an empty head.

The same trap sits under flashcards generated in bulk, outlines you did not argue your way to, and solutions you read rather than attempted. Each looks like output. None of them left anything behind. The test is simple and worth applying honestly: once the tool has finished, could you produce that same thing yourself, on a blank page, without it? If not, nothing was learned, something was only delivered.

This is not a moral point. It is arithmetic about where the hours went.

Fluent and wrong look identical

The characteristic failure of these systems is not that they say “I do not know”. It is that they produce a confident, well organised answer that happens to be wrong, and nothing on the surface of the text tells you which kind you just received. Invented sources are the sharpest version: a plausible author, a plausible journal, a paper that does not exist.

It does not weigh the same everywhere. A rephrased explanation you will then check against the lecture is low risk. A number, a date, a legal provision, a formula, a definition you will repeat in an exam: those have to come from the course material.

Which gives the working rule. The standard is the course, not the tool, because the course is what you are examined on. When the two disagree, the tool is wrong by definition for your purposes, even if it is right in some general sense. Marking happens against the syllabus.

Rules first, not afterwards

Institutional rules vary widely, by faculty, by module, and by type of assessment within the same degree. Some coursework permits tool use with a declaration. Some forbids it outright. Some distinguishes between tidying your language and generating your argument. Some has not been updated and says nothing, which is its own risk.

Read the assessment regulations for this specific piece of work before you start, and if the wording is ambiguous, ask the person who will mark it. Beforehand that is a short, boring conversation. Afterwards it is neither, because the question is no longer what you used but what you declared.

Produce first, then let it criticise

The one pattern worth internalising fits in a sentence: produce your own version first, then ask for criticism, never the other way round.

Write the paragraph, then ask what a marker would object to. Solve the problem, then ask where the reasoning is thin. Explain the concept from memory, then ask what you left out. In that order the tool is checking work that already exists and the thinking stays with you. In the reverse order you are editing something you did not build, and editing is a much weaker activity than construction, which is why it feels so much easier.

The order also protects your judgement: if you have already formed an answer, you can notice when the output is wrong. If you have not, you have no basis on which to disagree.

It does not know what your examiner cares about

Prioritising is the job these tools do worst, and it is one of the most valuable decisions in a course. What gets emphasised, which topics reliably appear, whether the exam rewards derivations or applications, what the person marking treats as essential and what as background: none of that is in the text. It is in the lectures, in past papers, in the way a question was answered in a seminar, in the things your tutor repeats.

Ask a model what to focus on and you get a general answer about the subject in the abstract, delivered in the same confident register as everything else. Use the course for that, and the tool for the parts where the answer does not depend on who is teaching you.

The hour that cannot be delegated

The honest limit has nothing to do with accuracy or rules. No tool replaces the hour you spend stuck on the material yourself, and that hour is the learning. Every sensible use above is a way into that hour faster, or a way of making it less wasteful. None is a way around it, and a study routine built to avoid it will feel efficient for a term and leave very little behind.

The question at the end of a working hour is not how much got produced but who produced it. An hour of writing, recalling and getting stuck leaves residue. An hour of reading finished text leaves a similar feeling and nothing else. Assistance is worth having, up to the point where it starts flying the aircraft.