Skip to main content

1.3 Responsible Use of AI

Take Home message

  • Responsibility is a design task, not a control task. Detection of AI does not work properly.
  • And AI detection is unfair when it fails. ThereDetectors iswrongly aflag riskhuman of mistakenly identifying answerswork as AI-generated,written, orand vicemiss versa.AI-written Severalwork. EuropeanEducation countriesauthorities havein alreadyFrance ruledand outthe theseNetherlands tools.advise against using them.

  • Confident invention and inherited bias follow from how these systems are built and judged. The consequence should be: teach them rather than waiting for a fix.  
  • The working principle is: think first, then AI. Making people commit to their own answer beats explaining the machine;machine and building limits into the software beatsremoves banningthe it.harm that unrestricted access does.

  • Prompts alone don't hold. A chatbot tasked with acting as a tutor tends to drift back to giving answers, which makes reliable limits a purchase question.
  • Amplification is the default; equalization must be built from access, design and scaffolding togethertogether. — and theThe scarcest but most important of the three is a trained teacher.
  • Inclusion is the clearest win, and where teachers themselves see the strongest case.

If you remember only one sentence from this page: The dams are breaking. Teach swimming.

The reading journal, second look

Back to our ninth-grade student and his suspiciously polished reading journal. The instinct is almost universal, and it is the instinct of a good teacher: find out whether he used AI. Build a better dam.

This chapter is about why that instinct, on its own, leads nowhere. We will come to what works instead.

Start with the uncomfortable part.  Teachers cannot reliably tell AI text from student text. In two studies by a German study,team, 28989 trainee teachers ofand different200 experienceexperienced teachers were given a mix of student writingessays and chatbot writingessays and asked to sort it.them. Neither group could.could Butdo so reliably but both groups were confident they could. were able to. Experience madehelped almostonly noa difference. Atlittle: scale,the itexperienced isteachers worse.still missed most of the AI texts. At one British university, researchers submitted chatbot-generated answers into theto real examinationunsupervised systemonline exams through the normal channels, without notifyingtelling the markers. 94 % were never questioned. Moreover, on average, they were graded above the work of actual students. So, what's the consequence? We reach for the machine! And here the finding stops being awkward and becomes an ethical problem!

The detector that fails the wrong students

Researchers tested seven of the most widely used AI detectors on TOEFL essays written by studentspeople learning English as a foreign language.language, collected from a Chinese online forum. Alongside, they put essays by native-speaking American eighth-graders. The native speakers' work was classified correctly almost every time. Of the foreign-language essays,  more than six in ten were wrongly flagged as machine-written.

The mechanism is not mysterious. TheseMany of these tools do not detect the work of Largelarge Languagelanguage models,models. theyThe detect tools measure predictability. Text that uses common words in expected orders scores as artificial. Anyone writingWriters in a second language, with a smaller vocabulary and safermore rigid sentence structures, producescan produce exactly that signal. Meanwhile, detectors trained for one specific test, and studies in other languages, show smaller differences. Up to now there are no detectors on the market that are reliable enough to accuse anyone of willful deception.

The tool's mistakes fall systematically on the students who are already carrying the most. Institutions have done the arithmetic. Vanderbilt University pointed out that even if such a tool were wrong only 1 % of the time, that would still mean roughly roughly 750 wrongly accused studentspapers a year wrongly flagged out of its 75,000 submissions. Consequently, they switched it off. TheA guidanceguide issuedwritten by researchers for Germany's federal education ministry is blunt: the technology is "not sufficiently reliable to provide legally secure proof." An independent European test of fourteen detectors reached the same verdict: "neither accurate nor reliable."

Much of Europe has stoppedmoved debatingon this.from France,this debate. France's education ministry and its data protection authority advise schools not to use detectors, the NetherlandsICT andagency SwitzerlandKennisnet advisein againstDenmark detectorscalls orthem highly unreliable. In higher education individual universities, such as the ZHAW in Zurich, forbid them.using Thethem Unitedfor Kingdom'sgrading. In England, the exam regulatorboards hasallow deliberatelydetection chosensoftware humanonly judgment over software. The Dutch argument goesas one stepelement furtherof anda judgement that rests on the teacher, who "knows their students best". A data-protection point is worth carrying home:home as well: pasting a student's work into a detector is itself an actonline oftool means processing their personal datadata.,  For this purpose, a legal basis and doinga itcontract withoutwith permissionthe breaksprovider datais protection law.required. → 1.5 Compliant Use of AI

It seems that the cheating panic is thinner than the noise around it. Researchers surveyed students at three secondaryUS high schools before ChatGPT existed and againafterward.again afterward. Cheating stayed broadly where it was. MostA follow-up survey of more than 4,000 students rejectedat havingsix other US high schools a chatbotyear simplylater producefound theirthe work, while thinking it fair to use one to get started or to have something explained.same.

The real question was never, "How  do I catch him"? It is: what does a homework task still measure, once its product can be generated on demand?

Two things that will not simply be fixed, so they must be taught

Confident invention is built in. These systems generate plausible continuationcontinuations of a text, whichand areplausible is not the same as truetrue. ones.Reviews Researchersof havethe nowresearch formallyagree shownthat whereno thistechnique comeshas from:yet removed the problem, although it can be reduced and partly detected. The reason is, that the standard tests used to judge Largelarge Languagelanguage modelmodels reward a confident guessingguess over admitting uncertainty, so as long as they are measured that way, a certain rate of fluent, well-formed falsehood persists.will Itpersist. is a property of how they are built and graded, not a bug awaiting a patch.

The slants in the training material are built in too. Moreover, they do not shrink as the technologymodels improves.improve. A European study of gender stereotyping across languages found that bigger models, even after the extra training intended to make them safer and fairer, sometimes stereotype stereotype more rather than less.  If a property cannot be engineered away, it becomes curriculum. And it can be taught: in one Finnish programme, the share of students who could identifyexplain where an image generator's bias incame an AI system's outputfrom rose from from 7 % to 44 %. The consequence for teaching must be training error analysis instead of error avoidance. In other words: the machine's failures become the lesson.

What works: design, not control

The evidence points to one simple principle that's easy to remember: think first, then AI.

  • Make the human commit first. In aan seriesexperiment ofwith experiments,about 200 adults, people who had to write downmake their own answerdecision before seeing the machine's suggestion caughtwent faralong more ofwith its mistakeswrong suggestions less often than people who were shown detaileda explanationsshort explanation of how the machinemachine's reasoned.reasoning. Explaining the AI did not helpsolve tothe solveproblem. The method that issue,helped butwas  forcing the judgement. Forcing judgement did.means Withforcing onepeople honestto caveat:think. Since thinking is exhausting the versions that worked best were the ones participants liked these methods least, and they helped most those who already enjoy effortful thinking. So itThat is awhat phenomenonyour alsoall knownknow as the Matthew Effect:effect: ToMore is give to those who have,already morehave. will

    be given. They benefit the most from this approach.
  • Build the limits in. The chatbot from Chapter 1.32 , that withheld answers and asked questions instead is the same idea in software: the students who used it lost nothing on the exam, while those with the unrestricted version lost badly.
  • Make the process visible instead of hunting the product. Drafts, working notes, a short conversation about the work.   These show you what a student can do without anyone having to prove what they did.
  • Denmark is piloting the principle: at participating schools, students may use AI while preparing for the oral English examination, and the written examination includes a handwritten, aid-free part.

Two limits are worth knowing before you rely on this. 

The first: the popular advice that teachers should simply instruct a chatbot to act as a Socratic tutor doesmay not survive contact with a long conversation. AtEarly leastevidence, atfrom thea moment,study thewith systemsimulated tendsstudents tothat has not yet been peer-reviewed (October 2026), suggests that under persistent pressure such tutors drift back into supplying answers. Reliable limits have to be built into the system, which makes them a purchasing decision rather than a prompting skill.

The second: making an AI explain its reasoning sounds like the obvious safeguard, but explanations do not reliably prevent over-trust. In one study they left people's acceptance of wrong advice unchanged, and explanations that are too technical or too simplified can increase misplaced confidence. Explanation is a condition, notonly a solution.condition for learning.

Attitude: two failure modes, not one

We worry that teachers are placing too much trust in AI. However, we should be equally concerned about the reverse.  Even experienced people can be caught out by over-trusting. In onea study,small teacherspilot reviewedin AI-generatedUS grades.schools, Theyfour judgedin 42%ten of the machine'teachers surveyed judged an AI grading tool's feedback tonot beuseful, vaguemostly vague, or wrong,even butincorrect. changedResearch onlyon 9%automation ofbias shows that people tend to go along with a machine's suggestion even when they have reason to doubt it. Spotting an error and correcting it are different processes.steps.

The opposite pole is concealment. Teachers hide their own use of AI from colleagues and students, forbidding them to do what they do themselves: 77% use AI privately, while 57% say students should never use it to generate ideas. Meanwhile, a study of academically able students in grades 6 to 8 found that between a third and two-fifths already said the AI knows more than their teachers.  A profession that hides its practices cannot model responsible practice, and transparency is not just a courtesy. It is the mechanism by which teaching by example works.  

Does AI level the field or tilt it?

In learning tasks, it does the opposite. One study found no average benefit at all, but a widening distance between students who already knew a lot and students who did not. Another looked at students writing in a foreign language: the weaker writers received more corrections from the AI and successfully acted on fewer of them, and several found the flood of feedback dispiriting. Feedback without the ability to sort it does not help; it buries.

The gap also exists before anyone touches a tool. In Germany, 80 % of students from the most advantaged homes see AI as an opportunity, against 55 % from the least advantaged. In Britain, the divide runs through the staffroom rather than the device cupboard: 45 % of teachers at private schools have had formal training in using AI, against 21 % at state schools. Within the state sector, wealthier schools again outpace poorer ones. A 2026 survey of more than 25,000 children across 2017 European countries finds the same social split in how much and how variedly they use AI. AI [PRÜFEN: SES-Befund am Bericht]. One commentator's line captures the whole problem better than any figure:  "the rich have access to both technology and people to help them use it, while the poor have access to technology only." Which is the strongest available argument that teachers matter to close this gap.

So the bottleneck is whatwhich learner has a teacher who has been trained. AI compensates when three things hold together: secure access, a tool designed for learning rather than answering, and a teacher scaffolding its use. Remove any one, and it amplifies instead. Since all three tend to be present in the same schools and absent in the same schools, amplification is the default and equalisation is something you have to build.