Analytic memos (44)

The author list (Zhi-Xuan, Franklin, Gabriel, Kasirzadeh, Lehman, Levine + Meaning Alignment Institute) makes this the successor-programme manifesto - which raises the stakes for the dissertation's timing: the proposal should cite it as evidence the field is converging on the dissertation's diagnosis (thick values, institutional co-alignment) while the dissertation's distinctive contributions (Rossian adjudication, responsibility allocation, non-Western engagement, naturalistic corpus) remain outside TMV's five application areas. Also note their fifth area, 'democratic regulatory institutions,' overlaps the governance chapter - read that section closely before the proposal. Update the skeleton's conclusion, which currently lists this paper as awaiting integration.

The African-philosophy thread now has its institutional anchor. Structural observation for the dissertation: Mofuoa's four Basotho institutions map ONE-TO-ONE onto the functions the Western alignment literature reinvented from scratch - Pitso = alignment assemblies / STELA focus groups / G&K deliberation; Lekhotla = adjudication and contestability (J&N's normative interface, S&K's recourse); Lekhotla la Baeletsi = STELA's expert correctives and S&B's epistocratic layer; Baholisi = ongoing oversight/mentoring (calibration, Article 14 human oversight). The decolonial point is thus sharpened from representation to PRIORITY: the deliberative-adjudicative-advisory architecture the field converged on has centuries-old working implementations in African statecraft that the discourse never consulted. Pair with Metz's relational moral theory (values layer) and Mofuoa (institutions layer) for a two-level African contribution; connects to the corpus's Cross-Cultural category empirically. Also a candidate external contact/examiner-suggestion network node (NWU, South Africa).

Validation with an edge: the field's newest handbook OPENS by arguing the dissertation's framing premise (alignment needs metaethics). Use it to establish that the metaethical turn is now recognized - then differentiate: Unver diagnoses the deficit from a law/regulation vantage and gestures at metaethics generally; the dissertation supplies a SPECIFIC worked metaethic (Rossian pluralism + convergentism + expressivism) with empirical machinery. Also mine its reference base (Siapka's feminist metaethics of AI, AIES 2022) for the metaethics chapter's related-work.

The FMTI decline (58->40) is the single most useful number in the report for the dissertation: every explanation-based responsibility mechanism in the coded library (Kästner's MI regime, S&K's justification criterion, Huang's knowledge-condition interface, J&N's contestability) PRESUPPOSES disclosure - and the measured trend is the opposite. This grounds the governance chapter's case that voluntary transparency has failed and responsibility-allocation rules must be mandatory (EU AI Act art. 86 right-to-explanation; H&D's liability clarification). Cite alongside the agentic-capability numbers to show both blades of the scissors: capability up, epistemic access down.

The handbook's own architecture VALIDATES the dissertation's three claimed gaps by dedicating chapters to each: metaethics of alignment (Ch1), responsibility-without-agency and the responsibility gap (Chs 17,18,22), and non-Western frameworks (Chs 53,55,56 - including BASOTHO governance, the most direct African-philosophy-and-AI-governance source yet found; queue Ch53 for full coding and possible citation network contact). Also Ch39 (AI ethics in HRM) is directly the AI Interviewer domain and Chs 25-32 the AI Scribe domain. NEXT ACTIONS: dedicated session coding Chs 1, 17, 18, 22, 53 as separate sources - these five chapters may reshape the lit review's Part II and the African-philosophy subsection.

STRUCTURAL MODEL ALERT: this is a philosophy PhD dissertation (Bonn + PUCRS cotutelle, begun 2021, defended, then published in a major Springer ethics series with a Markus Gabriel foreword) on normative ethics + value alignment with EMPIRICAL components and practical artifacts (dashboards, repos - exactly Augustine's dashboards). It is simultaneously: (a) proof to Howard/the department that this dissertation-shape is examinable and publishable; (b) the closest competitor to differentiate from - Kluge Corrêa's consensus-mapping ('where consensus exists') is adjacent to convergentism but appears not to have the Rossian object-level theory, the responsibility-attribution focus, or the stakeholder-stratified corpus; (c) a checklist source - before the proposal presentation, deep-read Chs 4-7 to confirm the differentiation claims. HIGH PRIORITY follow-up read.

Units 4-5 give the case chapters their sharpest evaluative contrast: the SAME accuracy-first logic that vindicates AI in diagnosis (AI Scribe's home domain) DELEGITIMATES it in sentencing-like contexts where reciprocity/accountability are constitutive (AI Interviewer for hiring; Chibook for immigration status - both closer to sentencing than to diagnosis in what they do to a person's standing). Tasioulas's question - what is lost when the proximate decision-maker 'cannot be held accountable... in the way that a human judge can' - IS the dissertation's responsibility question asked from the dignity side. And his explanation-the-defendant-can-grasp point anticipates the intelligible-reasons requirement (S&K's justification criterion, J&N's contestability).

The survey's 'Ethicality' pillar and 'Human Values Verification' subsection are strikingly thin relative to the other 100 pages - formal machine ethics + game theory, with community norms invoked but never theorized (unit 2). This is the measured version of the McKinlay finding at the technical field's own summit: the E in RICE is a placeholder. Lit-review use: cite Ji as the definitive map of what technical alignment CAN do, then locate the dissertation precisely in the placeholder - the verification of value alignment (unit 2's 'adherence to community's social and moral norms') requires exactly what the xphi corpus + convergentist metaethics provide: an empirically grounded, normatively defensible account of which community norms, whose, and why.

Methodological accessory for the corpus chapters: when interpreting folk-comment distributions (e.g. availability-driven spikes after incidents, anchoring on sci-fi frames), Roy-Stang & Davies provide the bias taxonomy for a 'discourse hygiene' subsection - which folk patterns reflect stable normative positions vs known perception biases. Also supports the Brophy-style FILTRATION story: bias-aware filtering of folk judgments into considered judgments is precisely the CMJ step. One-cite use.

Direct methodological neighbor of Augustine's experiment: HVAE assigns value-personas and measures behavioral conformity - his multi-LLM debate assigns normative-theory-personas and evaluates argumentative conduct. HVAE's neutrality-drift finding predicts what his setup should also observe (assigned positions eroding toward hedged neutrality) and gives him a published autometric concept ('value rationality') to adapt. Also completes an ironic pairing with Rozen (via LI_2026): personas CREATE value coherence where none exists natively, yet pronounced personas can't be SUSTAINED - both directions undermine attributing stable values to the model itself.

This review is the case chapters' shared technical backdrop: all three dissertation cases are ADVISORY systems in its sense (AI Scribe = clinical decision support, Chibook = immigration advice, AI Interviewer = hiring support). Three direct uses: (a) unit 1's finding that the normative/principle-driven stream encodes 'ethical, LEGAL, or PROFESSIONAL standards' maps each case to its standard-source (medical professional ethics / immigration law / employment equity - cf. ZHIXUAN role-norms memo); (b) unit 4 (advisory AI perceived as moral authority, inviting overreliance) is the mechanism behind the scapegoat/consent-laundering pattern in advisory contexts - the clinician defers to the scribe, the officer to Chibook - connecting S&K's epistemic-deference analysis to the deployment class; (c) the underdeveloped fairness/cognition streams (unit 1) plus the missing responsibility dimension (this review, like McKinlay's, has NO responsibility category at all) let the case chapters claim measured gaps. Note for lit review structure: review's three 'logics' (individualist/institutional/collective) = another independent arrival at the constituency dimension (Baum) and the stakeholder axis (corpus design).

Unit 2 (dilemma responses vary across societies and time) is usually read as undermining folk moral data; the dissertation can invert it: cultural/temporal variation is only a defect if you want a UNIVERSAL benchmark - for the dissertation's purposes (which values do differently-situated stakeholders want AI to embody; where do they converge despite variation) the variation IS the phenomenon under study, and the corpus's platform/category/stakeholder stratification is designed to measure it rather than average it away (cf. the disagreement-as-noise critique, SCHUSTER_KILOV unit 13). Same inversion works against Awad's Moral Machine (which LaCroix implicitly targets): its cross-cultural clusters were its most philosophically interesting finding, mishandled as benchmark input.

Unit 1 is the ML venue's version of the dissertation's stakeholder premise: group-structured value plurality is now a FORMAL modelling requirement at ECAI, not just a humanities claim. Two uses: (a) cite alongside STELA and Lloyd's latent classes as the third methodological cousin of the xphi corpus design (deep clustering : their travel data :: stakeholder/category stratification : folk corpus) - and note the corpus's advantage: their data is revealed travel preferences (thin), the corpus is reason-annotated normative discourse (thick); (b) their 'socially shared value groundings' vs 'diverse value systems' distinction (shared MEANING of values, diverse WEIGHTINGS) maps precisely onto the folk corpus finding-structure - e.g. shared invocation of 'accountability' with divergent weightings across stakeholder groups - and onto the overlapping-consensus structure (shared groundings) the convergentist argument needs. A useful formal vocabulary to borrow.

The Albania Diella case (unit 3) is a gift for the governance strand: a state formally designating an AI system as 'Minister for Public Procurement' - i.e., institutionally ASSIGNING office-responsibility to an artifact while a human (the PM) 'oversees'. This is consent-laundering/scapegoating at constitutional scale: the decree structure creates exactly the responsibility-attribution puzzle the dissertation analyzes (if Diella's procurement decision wrongs a bidder, who answers - the 'minister' with no moral agency, or the overseer who cannot review each decision?). Verify the decree's current status, then consider it as the governance chapter's opening case alongside the EU AI Act material. Note it postdates and outruns the entire responsibility literature coded so far.

Unit 5 is the only place in the coded set where an alignment DESIGN explicitly targets a responsibility condition: explanations exist so users satisfy the knowledge condition and can bear responsibility for choices. This inverts KAESTNER's scapegoat analysis constructively (there: opacity defeats responsibility; here: explanation infrastructure manufactures it) - but it also creates a new consent-laundering risk the authors don't see: if the interface's explanations are unfaithful (MILLIERE's thought-injection; the wise-judge problem), the 'knowledge condition' is merely simulated, and responsibility is transferred to users on false pretenses. The dissertation's faithfulness-testing methodology is the missing audit for responsibility-aware designs like DVA. Direct Work-chapter relevance: an AI Interviewer built on DVA would make the HIRING MANAGER 'responsible' via explanations - genuine or laundered?

METHODOLOGICAL SELF-CHECK, important: unit 4's sycophancy findings apply to the dissertation's own instruments. The folk corpus's LLM-coded annotations (Llama 3.3 70B coding reasoning/values/stances) and the multi-LLM debate experiment both use RLHF-trained models that mirror framing on contested moral content - so prompt wording could systematically bias codings toward the prompt's implied view. The methodology chapter must document the defenses already in place (fixed rubric-based prompts rather than opinion-eliciting ones; human-truth validation samples; IRR against human coders - cf. the κ=0.698 study) and should add a sycophancy-robustness check (recode a sample with adversarially reversed prompt framings; report coding stability). Raising and answering this objection preemptively converts a vulnerability into a methods contribution.

The 'normative interface' concept (unit 4) + law-as-paradigm (unit 7) is a ready-made frame for the Immigration chapter: immigration systems ARE Josifović-Noller normative interfaces (statutory norms, administrative contestability, role-based responsibility) - and Chibook-style immigration AI is a test of whether the interface survives automation: can an affected applicant still demand reasons, contest outputs, and locate a responsible human? Where GABRIEL_KEELING unit 14 shows immigrants fall outside the justificatory demos, J&N's framework says what they are owed structurally (contestation, reason-giving, attributable responsibility). The two together give the chapter both its problem and its normative standard.

Unit 5 restates the dissertation's core analytical task in the field's own vocabulary: identify relevant reasons across normative domains, measure their strength, aggregate into overall deontic verdicts - this IS Rossian weighing (NF-ROSS-PF), extended beyond morality to law and social norms. Two implications: (a) the folk corpus's coding scheme (reasoning types, value dimensions, policy stances) is an empirical instrument for step (i) - identifying which reasons folk actually treat as relevant per domain; (b) the three case chapters each sit at a different domain-interaction point (Health: moral+professional norms; Immigration: moral+legal; Work: moral+economic+legal), so the dissertation can claim to study exactly the cross-domain reason-interaction S&B identify as the frontier.

Units 3 + 8 hand the xphi corpus two precise technical roles that answer 'who is this for' at the implementation level: (a) unit 3 - credences 'set by taking a survey of relevant stakeholders': the folk corpus IS the large-scale stakeholder survey, and its coded distributions over reasoning types (deontological/consequentialist/etc. framings in folk_ai.db) are empirical credence-setting data for exactly this parameter; (b) unit 8 - 'inferring a common scale from [human judgment] data': the corpus's naturalistic moral judgments are the data this named-but-unexecuted research program requires. A 2021 ICML paper specifies the empirical inputs the dissertation's corpus provides - the strongest possible 'the field needs this data' citation.

The prototype machinery (unit 4) is directly implementable on the xphi corpus and could yield a publishable empirical paper: the folk corpus's coded judgments (which reasoning type / value / responsibility locus the crowd applies to which incident types) ARE applicability-task data in Peterson's sense - the ex-post center-of-gravity method could locate FOLK moral prototypes for AI-harm categories, and the AV(M) overlap index could then quantify how well an LLM's carving of AI-ethics moral space matches the folk's. This would be the first conceptual-spaces alignment measurement using naturalistic (rather than survey) data, and it directly extends Augustine's LLM moral-reasoning experiment with a formal metric. Note the measure needs similarity judgments, which the multi-LLM setup can elicit.

Unit 10 (Ma et al. 2025: 81.22% of 20,000 cases yield ethically inconsistent suggestions) is the strongest single empirical datum in the library for the anti-moral-agency thread: LLMs fail COHERENCE, the minimal condition for having a stable evaluative standpoint at all. This slots directly into Augustine's LLM moral-reasoning experiment as both baseline and method (a reflective-disequilibrium test is implementable in his multi-LLM setup - probe initial judgment, present critique, measure revision consistency). Acquisition target: Ma et al. 2025 (reflective disequilibrium; venue TBD) and Lu et al. 2025. Also pairs with MILLIERE (disposition-following) and Khamassi weak alignment: three independent empirical results now converge on LLMs lacking the deliberative stability moral agency requires - the experiment's conclusion has a literature.

Units 8 + 11 hand the xphi corpus its precise slot in the field's research agenda: the review names 'ways to effectively track stakeholder values in dynamic situations' as a neglected need and concludes the field 'needs to move beyond theory and simulations to include more empirical research including humans.' A continuously-collected, multi-year, multi-platform corpus of folk normative discourse IS a value-calibration instrument in their sense - it tracks stakeholder values dynamically at scale. Present the corpus not as sociology-of-morals but as the missing calibration layer the field's own review demands. This is the cleanest single answer to Howard's 'who is this for' question: the JAIR review says the field needs exactly this.

This SLR turns three of the dissertation's positioning claims from assertions into MEASURED findings citable to the field's own systematic review: (a) unit 2 - normative alignment research is rare and stops at 'higher conceptual levels rather than actionable processes' (so a dissertation that operationalizes a normative framework into coding schemes and case analyses fills a documented gap); (b) units 5-6 - the ethical canon in 172 papers is three Western theories, with consequentialism structurally advantaged by ML architecture (independent SLR confirmation of GABRIEL_2020 unit 3) and deontology unimplemented (which makes Sanwoolu's FUL proposal and the dissertation's Rossian operationalization both novel); (c) unit 10 - the review CONFESSES non-Western neglect, joining Gabriel's parochialism admission and Schuster-Kilov's Ubuntu footnote as the third self-indictment. The African-philosophy subsection of the lit review can now open with the field's own systematic review admitting the gap.

Unit 8's epistemic-injustice reading (marginalized participants demand impartiality/factuality because they are routinely wronged as KNOWERS - Fricker) offers the dissertation a ready frame for the folk corpus's own demographic patterns and for the African-positioning thread: if epistemic injustice shapes what communities want from AI, then the Global-South absence Gabriel concedes (GABRIEL_2020 unit 20) predicts systematically unrepresented normative priorities - which the dissertation's cross-cultural corpus categories can begin to surface. Also connects to LI_2026 unit 10 (participatory methods favor the well-resourced).

STELA is the published methodological anchor for the xphi corpus design, and the comparison cuts both ways. SIMILARITIES: elicit normative perspectives from differently-situated groups; code discourse into normative units with independent coders; foreground reasoning over bare ratings. DIFFERENCES that favor the dissertation: STELA has n=44, 8 focus groups, researcher-curated samples, US-only - the folk corpus has n≈366k coded comments across platforms, naturally-occurring (not curated) discourse, and a government-vs-public stakeholder axis STELA lacks entirely. DIFFERENCES that favor STELA: genuine deliberation (participants revise after hearing others - pre/post Likert), demographic identification of speakers, informed consent. The methodology chapter should present the xphi corpus as the SCALE complement to STELA-style depth methods: cite STELA to legitimate the enterprise, then show what naturalistic scale adds (and concede what curation adds that scale can't).

Units 5-6 give the Health chapter a concrete evaluative frame: AI Scribe systems are becoming personalised (adapted to one clinician's workflow) and agentic (drafting documentation autonomously) - by Kirk's criteria (interdependence, irreplaceability, continuity) clinician-scribe dyads will become RELATIONSHIPS, with attendant trust-drift and dependency (deskilling echo: FISCHLI unit 11). The folk corpus's AI-companionship and trust discourse (relatedness themes in the LinkedIn value coding) provides empirical folk articulations of exactly the three dilemmas this paper derives theoretically - a clean theory-meets-data pairing.

Unit 11 (latent-class stakeholder modeling from small judgment samples) is a methodological blueprint the xphi corpus already half-implements: the folk corpus's platform/category stratification and the stakeholder split (government vs public) ARE latent-class structures over normative positions. Lloyd cites Awad's Moral Machine clustering as feasibility evidence; the dissertation can cite Lloyd back: the corpus provides exactly the empirical stakeholder-preference distributions any of his three formal methods (parliament, MSEC, bargaining) would need as input. This positions the empirical work as UPSTREAM of the entire formal-alignment literature rather than as mere description - a strong answer to Howard's so-what.

Network/positioning note: Oluwaseun Damilola Sanwoolu (Yoruba name, U Kansas) is a natural node for Augustine's African-philosophy positioning and scholarly network - a philosopher of African origin publishing on AI ethics + Kant. Combined with the Ubuntu concessions in GABRIEL_2020 (fn14) and SCHUSTER_KILOV (fn1), and Metz's African moral theory (cited by Gabriel), there is a coherent 'African philosophy and AI alignment' thread forming across the corpus worth its own lit-review subsection and possibly worth reaching out to as part of building the dissertation's scholarly community.

This paper is the published mirror of Augustine's own position and gives him both an ally and a sparring partner. THE ALLY: Sanwoolu independently reaches Augustine's anti-Luke conclusion - AI is not a moral agent (units 5,7,8) - and, crucially, provides the exact rebuttal to Luke's inference (from AI-has-consequences/rationality to AI-is-responsible): units 7-8, the banking-software / sniffer-dog / rock cases show having morally-significant consequences does NOT entail moral agency. Augustine can cite Sanwoolu as independent published support for the experiment's conclusion, which directly answers Howard's 'unnecessary territory' worry - a peer-reviewed 2025 paper builds its whole argument on establishing exactly this point. THE DIFFERENCE: Sanwoolu argues from armchair Kant exegesis; Augustine argues from an empirical LLM experiment. Frame the experiment as the empirical complement to Sanwoolu's conceptual argument.

Unit 3 is a gift and a warning. Lundgren explicitly SEPARATES the control problem from the responsibility-gap question and BRACKETS the latter ('value alignment concerns... not the responsibility distributions for such outcomes... we have good reason to set those concerns aside... even if they are highly relevant'). This is a third top-venue author (after Gabriel & Keeling unit 12, Kästner et al fn2) who treats responsibility as adjacent-but-separate and defers it. The pattern is now overwhelming: the alignment literature systematically brackets responsibility attribution. The dissertation's contribution is to STOP bracketing it - to make responsibility-attributability the organizing question rather than the deferred one. Cite units 3 (Lundgren), 12 (G&K), and Kästner fn2 together as evidence the gap is real and self-acknowledged.

Agentic escalation, unit 12: Millière argues language agents INHERIT and AMPLIFY the shallow-alignment vulnerability (persistent memory, autonomous planning, indirect prompt injection via accessed web pages). This is concrete evidence for the dissertation's central agentic-AI claim: a known failure of pre-agentic systems becomes MORE dangerous and less detectable in agentic ones, and the responsibility question sharpens (who is responsible when an agent is jailbroken via a webpage it autonomously chose to read? - connects to KAESTNER type-(i)/(iii) difference-makers and the no-human-in-command residue). Strong material for the agentic-AI framing section.

Unit 4 (Mock Debate) and the whole attack-template family are structurally identical to Augustine's LLM moral-reasoning experiment (multi-LLM debate judging, per the course-video-production and phd-proposal memories): both have models argue assigned normative positions. Millière uses the setup to demonstrate a SAFETY failure (models produce harmful content under debate framing); Augustine uses a similar setup to test moral agency. The experiment can be reframed to speak directly to Millière: does having models deliberate over prima facie conflicts (rather than debate fixed positions) change the outcome? This connects the experiment to a live Phil Studies debate and to the Res Practica responsibility venue. Also: Millière's Thought Experiment template literally uses a philosophy professor persona - a pointed irony worth noting.

Unit 14 is the Immigration chapter's framing citation: prospective immigrants are the field's own named example of the all-affected justificatory gap - people profoundly affected by (AI-mediated) state decisions who receive no justification. Chibook sits exactly here: an AI system operating on people who are outside the demos that legitimates it. The chapter can ask: what do fair principles for immigration AI look like when the primary affected party cannot participate in the alignment assembly? The xphi stakeholder corpus (government side vs affected-public side) is a first empirical pass at surfacing exactly the claims G&K say must be consulted.

Role-specific normative standards (units 10-11, 15) map directly onto the three case studies: AI Scribe = the clinical-documentation role (norms from medical professional ethics - cf. LI_2026 unit 4's nursing/medicine example), Chibook = the immigration-advisory role (norms from administrative justice), AI Interviewer = the hiring role (norms from employment equity - cf. SCHUSTER_KILOV unit 12). The dissertation can test the role-norms framework empirically: does the stakeholder corpus show people actually reasoning in role-normative terms rather than preference terms? If yes, that is empirical support for Zhi-Xuan's framework from data they never had; if no, a documented gap between the philosophical proposal and folk practice. Either result is a contribution.

Footnote 1 (unit 3) is the second explicit concession in this batch (after Gabriel's geographic-parochialism admission, GABRIEL_2020 unit 20) that the field's dominant frameworks cannot speak for Ubuntu/communitarian perspectives - and here it is named: 'Ubuntu philosophy assumes the conceptual and moral priority of the community over the individual.' The dissertation's African-philosophy positioning is thus invited by the literature itself, twice. A lit-review section on 'the acknowledged Ubuntu gap' can be built purely from the field's own concessions.

Unit 12 is the AI Interviewer case, nearly verbatim: generative-AI job-application screening, easy-case feedback failing to generalize to hard cases (less experienced female candidate vs male-dominated workplace), trainers unable to detect their own biases. Work chapter should use S&K's analysis as the philosophical frame and then show what the xphi stakeholder data adds. Same for TU-HEALTH: their organ-allocation referendum contrast (transplant waiting lists) is a ready-made Health-chapter thought experiment.

Unit 4 (healthcare professional norms) is an almost verbatim design brief for the Health chapter's stakeholder methodology: professional-role values (nursing vs medicine) are structured collective norms that individual-preference aggregation destroys. The xphi stakeholder sampling (government/policymakers vs users/developers/deployers) can claim exactly this advantage over preference aggregation - it preserves role-structured normative positions. Also note both professions share Beauchamp & Childress principlism - a real-world overlapping-consensus-with-divergent-emphasis case.

Khamassi's strong/weak alignment distinction (unit 9) is a ready-made analytic frame for Augustine's LLM moral-reasoning experiment: his Claude/GPT moral-agent role-play experiment tests exactly the 'strong alignment' criteria (reasoning about intentions, representing causal effects of actions). The reported result - LLMs pass dignity pattern-matching but fail intentionality/causation tests - runs in the same direction as his conclusion against LLM moral agency, and gives it a peer-reviewed empirical anchor (also unit 6: LLM 'values' are persona-contingent, Rozen 2025). Acquisition targets: Khamassi et al. 2024 (Sci Reports 14, DOI 10.1038/s41598-024-70031-3), Khan/Casper/Hadfield-Menell FAccT 2025, Rozen et al. ICLR 2025, Shen et al. 2024, Edelman et al. 2025.

Third source in this batch to state Howard's descriptive-vs-normative worry, and this time it is the FIELD saying it about ITSELF: 'the positioning of preferences as normative when they are descriptive... philosophically naive.' Sequence across sources: Gabriel 2020 states the is/ought bar (GABRIEL_2020 unit 7); the IEAI brief shows the field now concedes current practice fails that bar; the dissertation supplies what neither provides - a worked metaethical account (Rossian pluralism + expressivism) of how descriptive stakeholder data CAN carry normative weight without the naive slide. The lit-review's motivation section practically writes itself from these three units.

The scapegoat-in-the-loop passage (unit 8, citing Ranisch 2024, American Journal of Bioethics - medical AI) is direct Health-chapter material for the AI Scribe case: a clinician blindly confirming AI-generated clinical notes is the paradigm scapegoat. And the structure of the worry - formal human sign-off without genuine understanding or control functioning as a responsibility-transfer ritual - is Augustine's comps 'consent laundering' concept operating on the deployer side rather than the data-subject side. Worth developing as a named parallel: consent laundering (data subjects) and scapegoating (human overseers) are two faces of the same illegitimate-responsibility-transfer pattern.

Responsibility silence, all three approaches: liberal approach (unit 10) lets the agent execute self-harming stated preferences (reckless gambling, unit 11) - who bears responsibility for the resulting harm, the user qua principal, or the developer who chose the liberal design? Boosting (unit 12) has the agent shaping preferences - responsibility for instilled preferences? Meta-autonomy shifts the design choice to the user's meta-preference - a consent-based responsibility transfer that looks structurally like the 'consent laundering' concept from Augustine's comps essays. The paper never raises responsibility at all. This is the exact junction where the dissertation's responsibility-locus analysis (RL-* codes) adds something the Gabriel program lacks. Directly feeds the Work chapter (AI Interviewer = an agent acting on delegated authority in employment decisions).

Units 20 + Gabriel's own Ubuntu footnote (p423 fn14, citing Metz 2007 and Mhlambi 2020) give the dissertation a principled, source-internal warrant for its African-philosophy positioning: Gabriel concedes the consensus he needs is geographically parochial. The dissertation's cross-cultural corpus work (incl. Cross-Cultural category in folk_ai.db) is an answer to a gap Gabriel himself flags.

Unit 19 (Jobin placeholder-consensus + Mittelstadt 'not action-guiding') is an empirically testable claim sitting exactly on the governance strand: the AIA corpus can test whether policy documents' invocations of fairness/transparency/responsibility show substantive convergence or placeholder convergence. This would be a novel empirical contribution to a debate Gabriel leaves at the level of citation.

Unit 7 is the single most important passage for the methodology chapter: Gabriel states the exact is/ought objection Howard raises about the xphi corpus ('AI cannot be made ethical just by learning from people's existing choices... cannot be solved by inference from large bodies of human-generated data by itself'). The dissertation's answer must be explicit: the corpus is not the normative ground; it feeds a reflective-equilibrium/convergentist argument (and, contra Gabriel's unit 25 moral-error point, convergence across frameworks + stakeholder groups is the error-check). Cite this passage, then answer it.