Page 328 - CyberSecurityEssentialsEnglish

AI may not be used to fill missing values, or to generate participants or results, and then
present them as real data. Synthetic data has legitimate uses in testing, simulation, and
privacy, but it must be named clearly and separated from empirical evidence. Fabricating

data or results and recording or reporting them is defined as misconduct [11].

TABLE 7-21: A CHECKLIST FOR AI-GENERATED CODE

Stage The check question

Inputs Are the column names, types, and units correct?

Cleaning Is the handling of missing values and duplicates declared?

Method Does the test or model suit the design and the objective?

Verification Are the assumptions stated and diagnostic checks run?

Correctness test Was the output tested on a small example with a known result?

Reproducibility Are the versions, the seeds, and the files saved?

Does the interpretation distinguish magnitude, uncertainty, and
Reporting
significance?

TABLE 7-22: THE ROLE OF AI IN QUALITATIVE ANALYSIS

Possible help What should not be fully delegated

Formatting texts and dividing them Deciding the meaning of the phenomenon
into units. without the researcher reading it.

Proposing initial codes or disconfirming
Selecting only the extracts that fit the story.
cases.

Comparing the application of Interpreting tone and cultural context
codebook definitions. automatically.

Hiding the researcher’s influence and
Organizing reflexivity memos.
relationship to the field.

Running locally or on de-identified
Uploading sensitive data without permission.
data as policy allows.

Linking themes back to the original Replacing participants’ quotations with
passages. generated summaries.

327