← Back to The Index Final Review Submission #12
Final Review

Review Nº 12

Ethical Criteria for AI Delegation: A Scoping Review from the Workers' Perspective
AuthorsJacopo Mascherin, Luca Casalegno, Salvatore Daniel Leocata, Livio Liao, Thomas Lieggi, Matteo Muratori, Redi Mustaj, Pietro Rinaldi, Marco Valori, Davide Ziarati Niasar and Lorenzo Accorroni
28/30
Score
The revision substantively addresses the most concrete consensus criticisms (source-count gap, method transparency, terminology consistency, framework diagrammatic structure, and conceptual differentiation) but leaves the two most strategic weaknesses (single-domain use case and the unbalanced normative tone) only partially resolved, and contains a clearly truncated sentence in Section 3.6.
High Distinction

The Pros

7 Items
+
Source corpus expanded to 41 cited references with a coherent bibliography, decisively resolving the "insufficient sources" concern raised by Review 7.
+
Method section rewritten with database list, search dates, eight numbered search strings, an inclusion/exclusion criteria table, and a PRISMA-style funnel — directly responsive to Reviews 1, 3, 5, and 7.
+
Framework restructured from four overlapping pillars into six clearly coded criteria (T1–T6) with a "Core Question" table, eliminating the redefinition redundancy flagged by Reviews 3, 6, and 8.
+
Terminology consistency across abstract, body, framework diagram, and appendix coding scheme — addressing Reviews 2 and 7.
+
Framework diagram (epistemic prerequisites / core / structural context) provides the visual summary requested by Review 1.
+
Section 3.5 explicitly introduces a positive-evaluation dichotomy on algorithmic neutrality, partially answering the "one-sided" criticism from Reviews 6 and 8.
+
AI-use disclosure (Appendix A7) and team-member appendix (A1) implement Review 4's appendix suggestion.

The Cons

7 Items
Section 3.6 ends mid-sentence ("...while AI trustworthiness and cultural readiness constitute") — an unacceptable editorial slip in a revised submission.
Only the radiology use case is presented, despite four reviewers (1, 3, 6, 7) explicitly requesting cross-sector examples; transferability is asserted rather than demonstrated.
The conclusion still declares total delegation "ethically unacceptable across all six criteria," reproducing the strong normative posture that Reviews 5, 6, and 8 specifically flagged as exceeding what the evidence supports.
The Method section provides procedural transparency but does not problematize selection bias, RTA researcher subjectivity, or reproducibility — the methodological reflexivity issue raised by Review 5 is unresolved.
The boundary between literature synthesis and authors' own normative interpretation remains blurred; the paper does not signal where descriptive mapping ends and ethical argumentation begins (Reviews 5 and 6).
Reference [41] (Tangi et al.) appears in the bibliography but is not cited in the body, and a few in-text reference numbers (e.g., reference 20 vs 19/20 in Section 3.3) warrant verification.
The "right to refuse" delegation suggested by Review 2 is mentioned only in passing in Section 5 as a future-work item, not integrated into the framework.

Suggested Changes

12 Pointers
01
High
Location
Section 3.6, final paragraph
Issue
The sentence "while AI trustworthiness and cultural readiness constitute" ends abruptly with no predicate, leaving a structural pillar (T6) without a closing claim
Suggested Fix
Complete the sentence and the paragraph — specify what these elements constitute (e.g., "constitute the cultural-organisational substrate without which formal oversight collapses into procedural compliance"), then close T6 with a one-sentence synthesis paralleling the closures in T1–T5
02
High
Location
Section 4 (Worked Use Case)
Issue
Only radiology is presented as a use case, despite four reviewers requesting cross-sector demonstration of generalizability; the abstract's claim of a "transferable" framework is therefore not evidenced
Suggested Fix
Add a second, briefer use case from a structurally different domain — e.g., AI-assisted CV screening in HR or algorithmic credit scoring in finance — and show explicitly which of T1–T6 binds in each case and why the binding pillars differ
03
High
Location
Section 4, prose use case
Issue
The radiology case is narrated linearly, making it hard for the reader to verify that all six criteria are covered as the abstract claims
Suggested Fix
Add a compact mapping table immediately after the prose — rows = T1...T6, columns = "Radiology instantiation" and "Verdict (delegable / shared-control / non-delegable)" — implementing Review 3's suggestion
04
Medium
Location
Section 1 (Method)
Issue
The research question is implied ("factors determining whether a decision-making task should be delegated") but never stated as a formal RQ, a deficiency flagged in Review 3
Suggested Fix
Open Section 1 with an explicit "Research Question:" line, e.g., "RQ: Which ethical conditions determine whether a workplace decision-making task can be legitimately delegated to an AI system from the worker's perspective?"
05
Medium
Location
Section 1, end of method
Issue
Methodological reflexivity (selection bias, RTA researcher subjectivity, reproducibility) is not problematized despite Review 5's explicit critique
Suggested Fix
Add a short "Limitations of the review" paragraph at the end of Section 1 acknowledging (a) coder subjectivity inherent to RTA, (b) the inability of any single search-string set to guarantee exhaustiveness, and (c) anglophone bias from the language filter
06
Medium
Location
Section 6 (Conclusion)
Issue
The closing statement that total delegation is "ethically unacceptable across all six criteria" reproduces the strong normative tone that Reviews 5, 6, and 8 specifically criticised
Suggested Fix
Soften the conclusion by qualifying the verdict to high-stakes, low-reversibility tasks, and explicitly acknowledge that low-stakes, structured tasks may be ethically delegable — connecting back to the dichotomy already opened in Section 3.5
07
Medium
Location
Sections 3.1–3.6, opening sentences
Issue
The paper does not signal where descriptive synthesis of the corpus ends and authors' own interpretive framing begins — Reviews 5 and 6 flagged this
Suggested Fix
At the start of each subsection, add a one-line marker distinguishing "what the literature converges on" from "what we synthesise from this convergence", e.g., "The corpus consistently reports X; we interpret this as constituting Y."
08
Medium
Location
Section 3.3, paragraph on autonomy
Issue
Review 2 specifically suggested addressing the worker's right to refuse AI delegation; this is currently only a future-work bullet
Suggested Fix
Add a sentence in 3.3 explicitly framing refusal as part of T3 (autonomy/agency), tied to either labour-law or collective-bargaining mechanisms cited in the corpus, rather than deferring it to Section 5
09
Medium
Location
Reference list and in-text citations
Issue
Reference [41] (Tangi et al.) appears in the bibliography but is not cited in the body; in-text reference numbering in Section 3.3 ("[19, 20]" then "[1, 23]") should be cross-checked for consistency with the final list
Suggested Fix
Run a citation/bibliography reconciliation pass: every reference must be cited at least once and every in-text bracket must resolve to a valid bibliography entry
10
Low
Location
Section 5 (Gaps and future work)
Issue
The gaps section is well-structured but reads as a dense list, weakening Review 8's suggestion to "more explicitly connect the four [now six] main points"
Suggested Fix
Reformat Section 5 as a short table (criterion | identified gap | concrete future-work direction), making the per-pillar gap structure visually immediate and reinforcing the framework's coherence
11
Low
Location
Abstract
Issue
The abstract declares the framework's transferability and the verdict on radiology, but does not mention the methodological scope (RTA, 41 sources, PRISMA-style) — a missed opportunity given Review 5's methodological-rigor critique
Suggested Fix
Add one sentence to the abstract specifying the corpus size, period, methodology (RTA), and PRISMA-style screening, anchoring the methodological credibility upfront
12
Low
Location
Appendix A5 (Screening Funnel)
Issue
The note "aggregate database counts are approximate due to overlap across platforms and manual de-duplication" is candid but weakens the reproducibility claim made elsewhere
Suggested Fix
Either provide exact per-database de-duplicated counts, or rephrase as a stated limitation of cross-database de-duplication and reference it from Section 1's new limitations paragraph (Pointer 5)
Back to The Index
Score · 28/30
Strong · But · Sharper
Pros / Cons / Pointers
Final Review · Submission #12 The Index Grandi Sfide · 2026