The phrase comes immediately: fluid, well built, without hesitation. It could present a cultural project, product or research group. Changing a few words seems to fit each one. Its convenience makes a question evident: how much of what we wanted to say did it really get into the text?
Creative automatism begins when an available form takes the place of observation. It can happen with a repertoire of slogans learned over the years or with a proposal generated in seconds. The issue is the relationship we build with that proposal.
An individual help, a variety problem
A study by Anil Doshi and Oliver Hauser, published in Science Advances In 2024, he examined the use of ideas generated by a linguistic model to write short stories. Some participants could receive one or more cues; another group wrote without that assistance. The stories were then evaluated separately.
I results of the original work show benefits in the creativity evaluations of assisted texts, accompanied by a greater similarity between those stories. The task was circumscribed: stories of eight sentences, precise experimental conditions, without a free and continuous interaction with the model.
This limit is essential. The work does not decide the fate of any art form nor does it stipulate that an author should give up the instrument. However, it brings to light a tension: an aid that improves an individual result can help to narrow the variety of the whole.
For a culture interested in many voices, the question becomes concrete. What choices make it possible to use a starting point without being anchored to it?
Plausibility still requires verification
A sliding text may make the moment when a proposal contains a given incorrect or non-existent source little visible. The NIST profile dedicated to generative artificial intelligence deals, among the risks, with confabulation and homogenization, together with the integrity of information. The NIST document AI 600-1 proposes an evaluation and management framework.
The useful distinction is between a sentence that works linguistically and a statement that we can support. A summary may assign a decision never made; a quotation may sound perfect without corresponding to the text; a reference may exist but not say what we attribute to it.
The editorial responsibility then passes from simple and demanding operations: returning to the original, controlling the passage, distinguishing the data from one's own reading. The machine can participate in the work; the public takes on the task of recognizing what it is offering the reader.
Recovering what resists the formula
A creative practice can begin before any generation. Annotating an observed detail, an uncomfortable question, an unsolved image allows a specific starting point to be retained. The instrument thus encounters a material that already has a history.
During the review we can look for phrases that would fit anywhere. What would replace a formula such as authentic innovation, deep connections or better future? A scene, a verifiable commitment, a contradiction would perhaps make the text less interchangeable.
Comparing multiple solutions also helps, as long as the variety is examined in the content. Changing adjectives and rhythm can leave the same idea intact. A more substantial difference concerns the point of view, the people present, the question to which one tries to answer.
Creativity retains a link to the risk of choosing. We can welcome a cue, reject it or transform it because we have seen something that deserves another form. It is in this passage that the text begins to have a need of its own, and one voice can become recognizable within many others.
