The ethics of the prompt: writing a query for an LLM is an epistemic and normative act, not merely a technical one

Two researchers argue in favour of ‘injecting’ one’s knowledge into the prompt rather than treating the model as an authority, and propose five principles for responsible prompting.

The rapid engineering as a technical know-how to obtain useful and well-structured answers. Federico Germani and Giovanni Spitale argue that it is much more than that: the formulation of a request determines the information requested, the assumptions introduced and the points of view put forward. Writing a prompt would therefore be both epistemic and normative.

Generate or inject knowledge

The authors oppose two ways of doing things. In the "generation of knowledge", the user implicitly treats the model as an authority that knows. In the "knowledge injection", it is the user who brings proof, context and expertise, which the model organizes and prolongs. The second seems to them to be more responsible, for no prompt substitute for the subject's control or the critical examination of the answers — even if any request inevitably conveys values and intentions.

They draw five principles: epistemic humility, explicit consideration of uncertainty, attention to sources, refusal of misleading framings, and clarity on the limitations of the model in choosing the tasks to be entrusted to it. The responsibility for the generated content, they conclude, is shared between users, developers, institutions and companies — and some uses cannot become ethical simply because of a better speed.