The difficult moment does not usually arrive when someone opens an AI tool. It arrives later, when a colleague asks who chose that image, why a sentence sounds so authoritative, or how a familiar voice became part of a public piece. Saying that a tool assisted is not yet an answer. People need to be able to trace the path of the work: its purpose, its experiments, its discarded options, and the people whose choices shaped it. Authorship is not a badge attached to a completed file. It is the willingness to stand behind the judgments that made the file possible. Artificial intelligence can offer speed, variation, and a starting point. It cannot take on that human relationship to consequences.
Consider a community group creating an exhibition about a neighbourhood workshop. One member brings family photographs, another researches the history, several young volunteers write captions, and a designer tests an image generator for layout ideas. The project can become richer because it has many contributors. It becomes vulnerable when those contributions vanish inside a result that appears to come from nowhere. A photograph loaned for a private conversation is not automatically a public design asset. A generated layout does not replace historical checking. A coordinator should not silently absorb the work of every participant. A useful plan makes the different roles visible without insisting that creativity must belong to one solitary person.
Making the context clear#
The first step is to frame the assignment as a decision rather than a request made to a machine. Before producing prompts or drafts, write down what the work is meant to accomplish and what it must not do. “Help visitors understand how the workshop changed without identifying people who did not agree to be pictured” gives the group a standard for judging options. “Make an impressive exhibition panel” does not. The first statement creates a real design problem; the second lets surface polish set the agenda. The same short brief can name boundaries: do not invent records, do not present unverified memories as fact, and do not reproduce the recognizable manner of a particular living artist. Boundaries are not enemies of imagination. They make room for choices that the group can actually defend.
The first step is to frame the assignment as a decision rather than a request made to a machine.
Every decision also needs a clear point of approval. This does not mean that the approver owns every contribution. It means that someone has responsibility for checking whether the agreed conditions have been met before publication. In the exhibition, the coordinator may confirm the final sequence, the researcher may verify dates, and each lender may confirm the agreed context for a photograph. One plain sentence is often enough: “Nothing is printed until captions and permissions have been checked.” That sentence prevents a last-minute export from becoming an irreversible act of publication. It also gives participants a defined place to bring concerns before speed turns them into disputes.
What changes in practice#
The second step is to distinguish decisions from materials. A decision might be choosing a color range, questioning a claim, rejecting a title, or suggesting a better order for a story. A material might be a photograph, recording, private message, contact list, screen capture, or sensitive data point. Decisions can be credited and discussed. Materials may require permission, anonymisation, or a strict limit on use. The distinction is easy to lose when a convenient tool accepts everything in the same input box. Pasting an entire meeting transcript to obtain a summary can turn a contribution to a room into data shared outside that room. Uploading a reference photograph can move a relationship of trust into a setting the photographed person never chose.
Making the distinction creates better alternatives. If a family photograph conveys an atmosphere, the team can describe the scene in its own words, use a silhouette, or select a non-identifying detail. If a voice matters, the person can record a new passage after receiving a clear explanation of its intended use. If a document contains delicate information, the team can extract only what is essential and verify it without sending the source file elsewhere. These choices do not reduce the ambition of a project. They require attention, and they often produce work that is less literal but more respectful. Technical convenience is not permission. Consent is a condition of responsible making.
Deciding with information#
The third step is to keep a small decision record. This is not a demand for exhaustive diaries or a mechanism for monitoring everyone. A single page can capture what the team wanted to do, what materials it considered, which tool assisted, what a person changed, and what still needs checking. For example: “Four heading structures were tested; the second was rejected because the timetable became unreadable; the drawing was made from scratch; the event date still needs confirmation.” Such notes give the project a shared memory. When a question arises, people do not have to argue from a hazy recollection or from the claim that something has always been done this way. They can return to the particular decision that produced the result.
A good record protects understanding without becoming surveillance. It should not contain passwords, private conversations, or a tally of every click. It should not be used to assign blame whenever a mistake appears. Its purpose is to show why an option was retained or discarded. A contributor can say that an assistant produced ten possible titles, that they selected one, rewrote it, and checked the claim it made. This describes assistance without pretending that the final wording arrived fully formed. Someone can also record that an image was rejected because it looked too close to a particular reference. A refusal is creative work too: it shows that judgment did not end when the first polished option appeared.
Reviewing and adjusting#
The fourth step is to review the public footprint of the finished work, not just its appearance. A page can look complete while still containing a misleading attribution, an unsupported claim, or a resemblance the group failed to notice. Review should ask what an audience will understand when it encounters the result. Will a visitor think a quotation is exact when it has been reconstructed? Will an image be mistaken for a photograph of a real person? Does the tone promise an outcome the project cannot provide? These are questions of context, not merely spelling. When a reasonable doubt appears, the most creative response is not always further polish. It may be removing an element, changing an example, requesting permission, or explaining a limit.
This is where teams need a practical distinction between inspiration, imitation, and appropriation. Inspiration may begin with a shared need: making dense information readable, creating a calm mood, or inviting participation. Imitation attempts to reproduce traits that make a particular creator or work recognisable. Appropriation treats another person's contribution or material as though it had no owner, history, or conditions. Generative tools can make the leap between these categories feel invisible because they return a fresh-looking result quickly. Human responsibility is to ask the question that speed conceals: does the assignment depend on the audience recognising a specific person? If it does, the team can reframe the request around general qualities and solve it with its own resources rather than requesting an indirect copy.
The fifth step is to agree on how assistance and collaboration will be explained. Transparency does not require a public diary of every experiment, and it does not require disclosing information that was protected for good reason. It does require avoiding a false impression that one person did all the work, or that no tool played a meaningful role when it affects how the work should be understood. Different audiences need different levels of detail. A private project note may describe the process closely. A public credit can be simpler: “Support tools were used to explore composition. Research, selection, final wording, and permission review were completed by the project team.” The explanation should be proportionate and readable, not a performance of technology.
Credit is also a conversation rather than an automatic calculation. One participant may want their research acknowledged. Another may not want their name attached to a version that moved away from their original intention. Those positions are not administrative inconveniences. Before publication, it helps to ask directly: “Is there any contribution that should be credited, corrected, or withheld?” The question creates room for a concern without forcing someone to prove they have more status than other participants. If disagreement remains unresolved, the careful choice is to pause the contested contribution rather than hide tension behind vague collective language. Respect is not collecting the greatest number of names; it is not tying a person to work against their wishes.
The exhibition group may face a concrete version of this issue. A family provides an old cassette recorded at the workshop. The group wants to improve its audio so visitors can hear it. Before processing the recording, they discover that a second identifiable speaker never expected the tape to be used in an open exhibition. Instead of abandoning the project, the team selects a section where only the consenting lender speaks and describes the source plainly. Other memories are introduced through new narration recorded by present-day volunteers who understand the purpose. The outcome takes more work, but it is more honest about what the exhibition can claim. Technology becomes part of a documented choice, not a device for erasing relationships.
Some difficulties begin with language rather than files. A volunteer might receive an elegant sentence from an assistant for the exhibition entrance. Reading it aloud, they notice that it implies historical certainty the group has not established. They discard it and write a quieter sentence based on what the research actually supports. This does not weaken the text. It avoids a subtle loss of authorship: placing one's name under words one cannot defend. People may use assistance to find possibilities for language. They must still retain the ability to answer for what the chosen language promises, excludes, or causes an audience to believe.
The plan becomes more useful through repetition. At the end of a project, fifteen minutes may be enough to ask what was easy to explain, which material caused uncertainty, what permission should have been requested earlier, and which rule would make the next project safer. The group then accumulates more than files; it develops shared judgment. It can prepare consent wording, workable ways to describe assistance, and a vocabulary for separating sources, references, and finished claims. This small infrastructure does not drain spontaneity from creative work. It prevents spontaneity from placing its cost on someone else's privacy, reputation, or uncredited labour.
Artificial intelligence can belong in a mature creative practice when it is not used as an excuse to erase the route to a result. Defining the assignment, separating decisions from materials, keeping a minimal record, reviewing the public footprint, and explaining collaboration make shared authorship visible. The next action can be modest: before beginning the next piece, write four lines covering the purpose, the boundaries, the permissions required, and the person who will approve publication. That note will not make the work for anyone. It will make it possible to answer honestly when someone asks where the work came from, and to include the people who truly made it possible.
