Lucía finishes a science assignment at the kitchen table. She asked an assistant why a plant grows toward the window and then used the reply to write her conclusion. Her brother Dani, two years younger, looks at the result and asks why he cannot do the same with his homework. Their father offers an improvised rule: “for studying, yes; for copying, no.” It sounds sensible, yet it leaves nearly every important question open. What counts as studying? Can they ask for an outline? What about checking a draft? What happens if the tool invents a source or retains what the children tell it? The family is not really debating an application; it is distributing trust, responsibility and opportunity without having agreed on the language for doing so.
Many household conversations about artificial intelligence begin and end with permission. People ask whether a child can use a tool, how long they can use it or whether a written assignment should be handed in without assistance. Those questions are necessary, but they are not enough. Family life does not work through one universal rule because ages, tasks, abilities and contexts change. Trouble begins when, to avoid a difficult conversation, every decision is left to urgency: an exam tomorrow, an insistent screen, a tired adult, a reply that looks useful. The tool then starts educating by default.
Before the rule, understand the situation#
The phrase “using AI for homework” contains very different activities. Asking for ten ideas to choose a topic is not the same as requesting a finished text and presenting it as your own. Neither is consulting an explanation of a concept, asking for a sentence to be simplified or entering a family’s private details so that it can draft a complaint. A useful conversation starts by separating these scenes. Once they are named, it becomes easier to discuss what someone is trying to learn, what information is being handed over and who answers for the result.
The phrase “using AI for homework” contains very different activities. Asking for ten ideas to choose a topic is not the same as requesting a finished text and presenting it as your own.
In Lucía’s case, the goal is not to memorize a definition of phototropism. It is to observe, form a hypothesis, compare an explanation with what she saw and communicate her reasoning. An assistant can help if it suggests questions for her experiment or points out that a claim needs checking. It can interfere if it supplies a conclusion so complete that the need to think disappears. The difference is not whether a screen participates, but which part of the process the student keeps. Sometimes a tool removes a barrier; at other times it removes the practice that gave the task its purpose.
The same distinction protects adults. A parent does not have to become a technical specialist to accompany these decisions. They can ask, “What did you ask it?”, “What will you check yourself?”, and “Which sentence do you really understand and which do you not?” Such questions turn use into an opportunity for shared learning. They also reveal when a reply is too persuasive to be reliable. If Lucía cannot explain in her own words what she will submit, the problem is not a moral failing; it is a sign that she needs another kind of help.
Consistency does not mean treating everyone identically#
Improvised rules often create conflict because children compare permissions. Dani may interpret the situation as his sister being allowed an advantage while he is required to work harder. Lucía may feel that any use of technology will be treated as suspicious even when it helps her organize a difficult task. A family needs coherent criteria, not necessarily identical rules. Age, demonstrated autonomy, the kind of activity and the risk in shared information can justify different decisions when the reasons are explained clearly.
For example, a family guideline can distinguish among exploring, producing and deciding. Exploring includes asking about concepts, finding vocabulary or requesting examples to understand a lesson. Producing covers writing, making images or preparing a presentation. Deciding concerns recommendations about health, money, relationships or personal conflicts. Each area calls for a different degree of accompaniment. In exploration, reviewing a sample may be enough. In production, keeping personal notes and declaring help where a school requests it is sensible. In sensitive decisions, the tool should not take the place of a conversation with a trusted person or professional.
This structure is not a list of prohibitions. It helps family members understand why one situation needs more care than another. A teenager can use an assistant to practice interviews but not to decide whether to stop a treatment. An adult can look up a recipe but should not paste in documents belonging to an older person they care for. Autonomy grows when the reason for a limit is understood and people participate in revisiting it, rather than receiving a “because I say so” that nobody can apply consistently later.
Privacy is the part that should not fall on children#
At home, details are naturally shared: a photograph of a bill, a teacher’s note, a conversation between siblings, a medical report. Generative AI can feel like a private interlocutor because it answers in a familiar tone. Yet entering information into a service means allowing it to circulate under conditions that are not always easy to understand. Expecting a child or teenager to calculate those risks alone would be unfair. Protecting privacy is an adult and collective responsibility.
One practical step is to agree on a short list of things not to type or upload: other people’s full names, addresses, passwords, health documents, identifiable images, private messages and school data that is not needed. The list needs alternatives beside it. If Dani wants help writing a message to a classmate, he can describe the situation rather than paste the conversation. If Lucía needs to review work, she can replace names and use an invented example. Alternatives matter because a rule with no route forward only teaches people to hide their use.
It is also worthwhile to review the service’s basic settings, minimum age and the possibility that conversations will be used to improve products. Families will not all make the same decision about which platform to accept, but every decision improves when it is made before an urgent situation. Set aside a calm moment, test the tool with an innocuous question, and notice what it keeps, requests and explains. That exercise makes privacy a practice rather than a lecture delivered after a mistake.
Make room for doubt and repair#
A family does not need to control every click. It needs it to be safe to say when something went wrong. A reply may present false information with confidence, a generated image may upset someone, or a child may enter data they should not have shared. If the automatic response is punishment and the withdrawal of all trust, the next incident will be hidden. If the response is to see what happened, limit harm and adjust the guideline, the mistake becomes an opportunity to regain control.
A simple household agreement can include an exit sentence: “If a reply makes you uncomfortable, asks for personal information or seems too good to be true, you do not have to solve it alone.” This applies to adults too. Someone using a tool to organize tasks or find information can admit that they do not know whether a source is real. Modeling that uncertainty has more educational value than pretending mastery. It teaches that critical thinking is not distrusting everything; it is knowing when to pause and look for an independent check.
Repair should be concrete. If a screenshot containing data was shared, it can be deleted where possible, a password can be changed where appropriate, and the family can discuss why it happened without turning the episode into humiliation. If an assignment relied too heavily on an automated reply, part of it can be rebuilt from personal notes and the teacher can be asked how to declare the assistance. If a recommendation shaped a family argument, what the tool contributed can be separated from what each person feels and needs. Repair does not mean erasing every trace; it means recovering the ability to decide after an error.
An agreement that changes with the family#
The best family guidelines are short, visible and revisable. They might fit on a sheet near the study table: we use tools to learn, not to hide someone else’s work; we check important claims; we do not share other people’s personal data; we ask for help before making sensitive decisions; we speak about errors without fear. Below them, a date to talk again in two months. The exact wording matters less than everyone knowing they can invoke the agreement and propose changes.
At that review, the family can examine real situations. Did the rule help Lucía understand her work better? Could Dani ask questions without getting trapped in an answer he did not understand? Did any limit prove so rigid that it encouraged shortcuts? Was any information difficult to protect? The answers allow the guideline to change with experience rather than fear or a company’s advertising. A tool changes, a school changes and children grow; a responsible decision cannot be fixed forever.
The kitchen conversation need not end with a perfect verdict. It can end with Lucía explaining her experiment in her own words, Dani knowing when to ask for support and the adults committing to review the agreement. That is already much more than permission. It is a way of building common criteria so that artificial intelligence does not dictate, from a silent screen, what learning, privacy or family decision-making should mean.
