Page 340 - CyberSecurityEssentialsEnglish
P. 340
content owners’ policies. Using extracts for analysis may fall under different rules depending
on the country and the contract; where in doubt, consult the responsible body rather than
assuming that the technology grants permission.
In research with human participants, the use of AI must be consistent with what participants
consented to. If you promised to keep the data in a defined environment, it may not be
moved to an external service. And if you use a tool to extract attributes, sentiment, or
classifications, a new analysis may arise that the consent did not describe. Material changes
need ethical or institutional review.
Transparency does not always mean publishing every internal prompt, since sensitive data
or proprietary constraints may exist. But it means providing enough information to
understand the tool’s role, its limits, and the human verification. Technical detail can be
placed in an appendix or a secure repository, with a clear summary in the text.
Finally, do not use disclosure as a disclaimer. The sentence “AI was used and errors may
occur” does not transfer responsibility to the reader. A researcher declares the use after
verifying the output, not instead of verifying it.
TABLE 7-31: EXAMPLES OF DISCLOSURE BY TYPE OF USE
Type of use A suitable form of disclosure
A brief statement in the acknowledgements or as the
Light language editing
journal requires.
Help in formulating search A note in the method with the tool and how the final
queries strategy was approved.
Assistance in extracting A method statement with the verification sample and the
fields from sources error rate found.
Writing or reviewing The tool, the version, and the tests run, with the code
analysis code deposited where possible.
Initial coding of qualitative The tool, the codebook, the human review process, and
data the disagreement handling.
Generating a figure or a A statement that the visual is generated and what it does
visual and does not represent.
A clear statement that interpretation was reviewed and
Help in interpreting results
decided by the authors.
339

