Author Response to Reviewers#
Reviewer key: G = Gemini, C = Claude, O = OpenAI (GPT)
1. Errors in the AI’s inner paper — already reported by me#
Several reviewers ask me to correct the inner paper’s overstatement of rigor (journal completeness, quote misattributions). O G C
I already disclosed these in my reflection: the journal was not maintained as claimed; and I found two quote errors (“limitations for young people” in Interview 26—not verbatim; “being out and about” in Interview 35, not 13). I will not revise the inner paper to fix them. The experiment presents the AI’s output as written; the reflection documents its errors. Correcting the inner paper would contradict the purpose of the exercise.
Thin excerpts / limited literature (O C)
The inner paper has sparse verbatim evidence but it is a short paper. I already acknowlege that there is little substantive literature engagement. I will not rewrite the inner paper. I will acknowledge in my wrapper that excerpt density and literature context are limitations.
2. Errors in the AI’s inner paper — I did not notice; I will add to my wrapper#
n=16 per borough vs 48 total (O)
The inner paper states “n=16 per borough” across four boroughs (16×4=64) but the total is 48. I did not catch this arithmetic error. I will add a note in my wrapper section listing it among the AI’s uncorrected errors.
3. Errors and omissions in my wrapper: won't fix#
These are more formal and don't really matter at this point, as the paper was accepted
Anonymity (O C)
GitHub link and identifying self-citations (Powell & Caldas Cabral; causal mapping) will be removed or anonymised for review.
AI Involvement Checklist (O)
Writing score 3.5 is invalid (scale 1–4). I will correct to a valid score.
Formatting (O C)
References, headings, and template alignment will be brought into line with the CFP.
3. Errors and omissions in my wrapper: will fix#
Authorship and voice (O C)
The inner paper uses “we” despite there being no human co-author. I will add an explicit note in my wrapper: who is speaking in the inner paper; what the human did (dataset choice, instruction framing, post-hoc checking, revision prompts); and where interpretive authority lies.
Autoethnographic reflection too thin (O C G)
I will expand the reflection to address the emotional/experiential dimension: what it felt like to watch the AI interpret vulnerable people’s accounts; moments of wanting to intervene; how “no actual person is the author” sat with me.
Model transparency (O C G)
I will specify the underlying model (Cursor’s default at the time) for reproducibility.
Ethics of AI processing (O C)
The data is already anonymised and public, but I will add a paragraph on data handling.
4. Where both I and the reviewer were wrong#
Negative-case search claimed as “systematic” (O G C)
The inner paper says “deliberate negative-case search” (3.4) and “we deliberately searched for transcripts” (3.3). It never claims a “systematic pass.” My reflection incorrectly stated that the paper claims a systematic negative-case pass; the reviewers echoed that. I will correct my reflection.