Responsible GenAI Use Statement: template
[!IMPORTANT]
Check the edition page
Your edition page gives the deadline and the email address for the submission. Do not include confidential or personal research data.
Length: 1,200 to 1,500 words, without the bibliography and the peer note. Work: alone, about your own research context. Peer discussion: discuss your draft with at least one peer who knows your field. Assessment: a formal check, pass or fail, with the seven criteria at the end. Submission: by email to the course administration (GRACE), or through the Use Statement issue form on GitHub. A GitHub issue is public.
Replace each guiding question with your own text. Keep the seven headings.
Title of your statement:
Research field:
Word count:
1. Intended use or justified non-use
- Which task do you want to do with generative AI, or which task do you decide not to do with it?
- Why do you consider generative AI for this task? What is the alternative?
2. System and data provenance
- Which system, version, and provider? How do you access it?
- What is known about its training data, its documentation, and its license? What is unknown?
- Which of your own data would the system receive?
3. Lifecycle risks and people affected
- Name at least two risks. Give a source or evidence from the course for each.
- At which lifecycle stage does each risk arise: design, data, training, deployment, evaluation, or governance?
- Who is affected: you, colleagues, research participants, data workers, the public?
4. Mitigations and less harmful alternatives
- What can you do to reduce each risk?
- Which less harmful alternative exists: another tool, a smaller model, a manual method, no use at all?
- Which mitigation can you carry out yourself, and which needs someone else?
5. Evaluation and limitations
- How do you check the outputs? On what sample, and by which criteria?
- What can the system not do for your task? What remains uncertain?
6. Disclosure, citation, and governance
- How do you disclose and cite the use in your publication or thesis?
- Which rules apply: your university, your faculty, your funder, your journal, the law?
- Which institutional owner must agree, or should know?
7. Defensible final decision
- Constraints: Which conditions must hold before you consider any use? Think of consent, data protection, and the rules that apply.
- Comparison: How do benefits, burdens, and uncertainty compare with the alternative, for the same task and the same quality?
- Accountability: Who decides, who can object, and when do you review the decision?
- What do you decide, and why? Which result would change the decision?
Follow-up action
- One specific action: what you will do, by when, and who must agree.
Bibliography
- List every source that you cite.
Peer discussion
Discuss your draft with at least one peer who knows your field. Then revise it. Do not write the name of the peer.
- Field and role of the peer:
- Date of the discussion:
- Main feedback:
- What you changed:
Publication
The instructors publish a statement on the course site only when its author agrees. Write one answer: no, yes, with my name, or yes, anonymously. Without an answer, the statement stays unpublished.
- My answer:
Questions for your peer
These questions can guide the discussion. They are not graded. Ask your peer whether the statement:
- Adequately scoped use: describes one concrete use or non-use, with the system and the task.
- Relevant evidence: names at least two risks, each with a source or evidence from the course that applies to this system and task.
- Feasible mitigation: proposes at least one mitigation or alternative that the author can carry out, and that addresses the cause of a named risk.
- Transparent practice: covers disclosure, citation, documentation, and the applicable rules.
- Reasoned conclusion: reaches a decision that follows from the analysis, names a check whose result would change it, and states its limits.
Formal check (pass or fail)
The instructors are not experts in your field. They check the form only. The statement passes when it meets all seven criteria:
- Length: 1,200 to 1,500 words, without the bibliography and the peer note.
- Structure: the seven headings, each with your own text.
- Sources: a bibliography that lists every source that you cite.
- Follow-up action: one action, with what, by when, and who must agree.
- Peer note: the field and the role of at least one domain-expert peer, the date, the main feedback, and what you changed.
- Publication: an answer in the section “Publication”.
- Data protection: no confidential, personal, or unpublished research data.
Original template, CC BY 4.0 · Stochastic Parrots and Beyond, Laura Wagner and Moritz Mähr.