The value alignment problem in advisory AI: a systematic literature review

Loukas Triantafyllopoulos; Evgenia Paxinou; Diamanto Tzanoulinou; Vassilios S. Verykios; Dimitris Kalles · 2026 · AI and Ethics 6:147 background medium priority coded

Main argument

Thesis (review findings): across 83 studies of alignment in advisory/decision-support/recommendation contexts, four approaches dominate - (1) personalized preference-based tuning, (2) normative/principle-driven frameworks encoding ethical-legal-professional standards, (3) fairness and cultural adaptation, (4) cognitive-bias mitigation - with preference-based and normative strategies dominant and fairness/cognition underdeveloped. Key findings: alignment is 'a dynamic, context-sensitive process shaped by evolving user values, cultural conditions, and domain-specific norms', not a static mapping - challenging fixed-value-set accounts; three coexisting LOGICS of alignment (individualist/preference, institutional/normative, collective/fairness) that theory should integrate pluralistically rather than privileging one; documented risks include hidden biases, overreliance, adversarial exploitation, echo-chamber steering of user judgment, and systems 'being perceived as moral authorities in ways that invite overreliance'. Publication explosion from 2022 (conceptual -> technical experimentation). Future needs: pluralistic alignment frameworks, standardized evaluation, longitudinal/cross-cultural studies, interdisciplinary governance.

Why it matters here

The PRISMA review of alignment specifically for ADVISORY/decision-support AI - the system class of all three dissertation cases (AI Scribe = clinical decision support, Chibook = immigration advice, AI Interviewer = hiring decision support). Its four-approach map, its finding that fairness/cognition approaches are underdeveloped, and its documented risks (overreliance, perceived moral authority) supply the case chapters' shared technical background.

Reading notes

Targeted read of abstract, discussion (24pp; PRISMA methodology - 83 studies 2011-2025, Hellenic Open University). Complements McKinlay (general SLR to 2023) with advisory-specific coverage through 2025.

Triantafyllopoulos, L., et al. (2026). The value alignment problem in advisory AI: a systematic literature review. AI and Ethics, 6, 147. https://doi.org/10.1007/s43681-026-01015-4

Close reading — 4 coded units

#1 · pp. 15 · evidence

“four main approaches can be distinguished. One stream focuses on personalized, preference-based methods [...] A second emphasizes normative or principle-driven strategies that encode ethical, legal, or professional standards directly into advisory processes. A third highlights fairness and cultural adaptation [...] research has concentrated most heavily on preference-based and normative strategies, while fairness- and cognition-oriented perspectives remain less developed.”

TU-LITREV

#2 · pp. 15 · argument

“alignment should not be understood as a static mapping between human preferences and machine outputs but as a dynamic, context-dependent process. This view challenges earlier accounts that treated alignment as the encoding of fixed value sets and instead emphasizes negotiation shaped by interaction, memory, and adaptation.”

VC-INTRA-VALUE

#3 · pp. 15 · argument

“Preference-based approaches adopt an individualist model focused on reflecting users' subjective values. Normative approaches prioritize shared principles and institutional standards. Fairness- and culture-oriented strategies extend this by foregrounding collective identities and historical inequalities. Rather than privileging a single paradigm, alignment theory benefits from a pluralistic perspective.”

NF-PLURAL-OTHER

#4 · pp. 15 · evidence

“Even when tuned to explicit preferences, systems can still reproduce decision-making vulnerabilities by subtly steering users' judgments and reinforcing 'echo-chamber' dynamics, or by being perceived as moral authorities in ways that invite overreliance.”

AG-UNDERSTANDING

Synthesis-matrix row

supports T4-ROSSIAN-DEMAND

three logics need pluralistic integration (SLR-level)

Memos (1)

thesis-link · unit #1

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).