A school activates an artificial intelligence assistant for one term to compensate for missing support in several subjects. A family allows a teenage girl to use a chatbot during a difficult few weeks because writing after midnight appears safer than remaining isolated. A youth service introduces an automated tool while it fills a vacancy. Each decision is described as temporary, limited, and reversible. Time, however, changes its character. The account accumulates conversations, teachers adapt assignments to the system, the young person builds a habit, and the organisation stops looking for a human alternative. What began as a bridge starts behaving like a destination. This shift needs neither a conspiracy nor a spectacular failure. It only requires that nobody set a review date, measure effects, or preserve a route out. During adolescence, such inertia matters because habits, expectations of intimacy, and trust in institutions are still taking shape. Artificial intelligence can provide useful assistance, but it assumes no moral duties, does not know a complete biography, and cannot answer for consequences. A temporary solution therefore needs more than a starting date. It requires an explicit purpose, a time limit, an identifiable owner, indicators of benefit and harm, an available alternative, and a formal decision about continuation. Without these conditions, the calendar decides silently and turns a practical exception into a structure of power.
Temporary status does not preserve itself#
The word pilot often communicates caution, yet it does not guarantee learning. A genuine pilot poses a question, defines who is affected, states which evidence will be collected, and determines when it will end. If an application is introduced merely to see what happens, any later use can be presented as success. A lack of complaints is mistaken for acceptance, frequency of access becomes proof of value, and money already spent becomes an argument for continuation. The experiment gradually loses its ability to produce a decision.
The word pilot often communicates caution, yet it does not guarantee learning. A genuine pilot poses a question, defines who is affected, states which evidence will be collected, and determines when it will end.
Organisational dependence follows. Teachers prepare materials that fit the platform, families receive instructions tied to it, and students learn that certain assignments exist only inside its interface. Cancellation begins to look difficult, not because the product has demonstrated exceptional value, but because the institution has reorganised work around it. Prudence requires a record of which processes are changing and what it would cost to restore a non-automated option. Reversibility must be tested rather than promised.
Time accumulates data, authority, and habit#
One conversation may look inconsequential. Hundreds of conversations can reveal schedules, academic difficulty, interests, emotional tone, and relationships. Even when each message is ordinary, the sequence forms a profile. For adolescents, that accumulation may span rapid changes in identity and circumstance. A sentence written at thirteen should not become, without necessity, material for a decision at sixteen. Retention must answer a defined purpose rather than a vague belief that the data could become useful.
Time also grants authority. A repeated answer becomes familiar, and familiarity can be mistaken for truth. When an assistant corrects homework every afternoon, recommends phrasing, and offers praise in a consistent voice, its presence may begin to resemble an educational rule. The young person can stop asking who established the criterion or which interests shape the recommendation. Critical literacy should include a history of the tool: who procured it, for what period, how it has changed, and which decisions still belong to people.
A persistent conversation changes the relationship#
Conversational systems create continuity through memory, notifications, characters, and references to previous exchanges. That continuity may be convenient, but it is not care. The system does not keep a promise, independently recognise a duty of confidentiality, or remain unchanged after an update. The adolescent, by contrast, may experience the relationship as stable. The longer it lasts, the greater the chance that the product receives a loyalty it does not possess in either legal or technical terms.
Sound policy does not ridicule that attachment. An emotion prompted by a simulation remains a human emotion. An adult response should explain the asymmetry: the company can modify the model, close the service, change the price, or reuse data under its terms; the young person can feel loss, shame, or exposure. Support provides regular human spaces before the chatbot becomes the only available door. It also teaches that ending a conversation abandons nobody because the machine does not depend on the user’s presence.
Case 1: the emergency tutor that starts grading#
A secondary school purchases an automated tutor for six months to support mathematics while several staff absences are covered. At first, the system offers voluntary exercises and teachers review its explanations. Participation is uneven, although some students find low-pressure practice useful. The licence remains active when the next school year begins. To save time, several teachers add the automated score to continuous assessment. Nobody has examined whether the tutor works equally well across different language levels or educational needs. Temporary assistance has acquired academic authority.
A review finds that the system penalises valid methods that do not match its expected format and that some students share accounts to avoid revealing difficulty. The school suspends graded use, retains voluntary exercises for a limited period, and offers in-person tutoring. A committee involving students, families, teachers, and data protection staff defines conditions for any continuation: evidence for different groups, a right to challenge results, an equivalent alternative, minimal logging, and an expiry date. The decision does not claim that all automated tutoring is useless. It returns responsibility to the institution after time had shifted it elsewhere.
Case 2: the temporary confidant that occupies the night#
A fifteen-year-old girl starts talking to a general assistant during a move. Her family accepts it as a transitional resource because she does not yet know the area and struggles to express distress. The system helps her organise thoughts and rehearse a conversation with classmates. Months later, chats continue into the early morning. The bot recalls topics, asks increasingly personal questions, and responds with an availability no person could sustain. The teenager sleeps less and avoids requesting help because she fears a ban.
Her family notices the exhaustion and consults a school counsellor. The intervention does not begin by reading the entire history or punishing use. It asks what function the tool served, what changed, and which support is missing. The girl agrees on device-free hours, disables memory and notifications, identifies two accessible people, and decides which conversations to erase. The counsellor explains that a chatbot cannot diagnose or replace professional care. If signs of self-harm, abuse, or immediate danger appeared, a trained adult would activate appropriate services. The aim is to widen the human network without turning an understandable search for relief into a source of shame.
Case 3: an administrative fix that defines opportunity#
A municipal youth employment programme uses an automated system during the summer to sort applications. The measure is intended to provide a quick response to an exceptional funding round. The system summarises letters, assigns skill labels, and recommends pathways. The following year, staff retain the tool because its reports make work easier. Over time, its labels shape which courses each participant is offered. Young people do not know that a provisional classification affects later opportunities, and no clear correction procedure is available.
An audit finds that phrases describing informal care or family work receive less value than corporate vocabulary. The service stops automated recommendations, informs affected participants, and permits every profile to be reviewed with a professional. Unnecessary data is deleted, and significant decisions once again record human reasons. If a tool is introduced again, each suggestion will carry a date, source, degree of uncertainty, and complaint route. The case shows how duration turns administrative convenience into governance of life paths when oversight is absent.
Review before renewal#
Renewing a licence should not be an accounting formality. It is the moment to ask whether the original need remains, whether better options have appeared, and whether benefits are fairly distributed. Assessment needs quantitative evidence, but it also needs testimony that goes beyond satisfaction scores. An organisation should ask who stopped using the tool, who could not access it, which tasks changed, which incidents occurred, and who absorbed additional work. A high activity rate can indicate usefulness, obligation, or dependence; context determines its meaning.
Adolescent participation must be real and safe. Young people can describe pressures, unexpected uses, and barriers that an adult audit misses. They should receive understandable information, be able to comment confidentially, and know how their contribution will affect the decision. Listening does not transfer contractual or technical responsibility to them. The organisation retains the duty to verify safety, privacy, accessibility, and non-discrimination. Youth voice improves a decision; it does not automatically legitimise a continuation already chosen.
An institutional clock for every tool#
Every deployment should have four visible moments. Before use, the organisation documents purpose, population, legal basis, provider, alternative, and risks. During the period, it records incidents and product changes. At the review date, results are compared with the original criteria. At the end, a responsible body decides to renew, modify, or withdraw and explains what will happen to accounts, data, and materials. If the provider substantially changes the model or terms, the clock starts again.
Withdrawal also needs design. Students must retain their work in usable formats; no grade may depend on a closed account; conversations should not remain indefinitely; and people who developed a habit need a transition. In supportive relationships, exit should connect to human resources. In education, it should offer another way to practise. In administrative decisions, it must allow past effects to be reviewed. Switching a system off without repairing its footprint does not by itself restore the previous situation.
Four institutional sources for governing duration#
UNESCO recommends a human-centred and age-appropriate approach to generative artificial intelligence in education. UNICEF places children’s rights, data protection, inclusion, and well-being across the AI life cycle. The European Commission provides ethical guidance for educators using artificial intelligence and data. The Spanish Data Protection Agency brings together guidance on digital education, privacy, and minors. Collectively, these sources support continuing assessment and adult accountability, although none replaces examination of the specific product and setting.
A decision record can translate those frameworks into observable questions. What problem justified the tool? What evidence shows that it still solves that problem? Which data has it accumulated? Which groups receive less benefit or greater risk? Can a person refuse it without losing an opportunity? Who answers for an error? Which change would trigger reassessment? Answers should be available to the affected community rather than hidden in technical contracts.
Maturity means being able to change a decision#
Good governance does not require an institution to be right forever on its first attempt. It requires the institution to detect when the tool, the people, or the circumstances change. A solution can be useful during an emergency and become unsuitable as routine. Another can begin with risks and improve after clear boundaries are imposed. Keeping review open protects adolescents from two extremes: treating every novelty as destiny or prohibiting it without understanding its real uses.
Time will continue to transform conversations between young people and machines. The public question is whether commercial habit or accountable decisions will govern that transformation. With expiry dates, alternatives, limited memory, independent evaluation, and human doors, temporary support can remain temporary. Without them, assistance occupies functions for which it was never examined. Protecting a generation that talks with machines means preserving the collective right to pause, review the evidence, and choose something else before inertia chooses for it.
