Artificial moral characters: constitutional AI and the challenge of alignment

Jörg Noller · 2026 · AI and Ethics 6:196 interlocutor medium priority coded

Main argument

Thesis: Constitutional AI produces 'Artificial Moral Characters' (AMCs) - engineered, stable, norm-guided dispositions that structurally resemble Aristotelian hexeis (mesotes-alignment between recklessness and over-caution) - but these do NOT constitute virtue, which requires right reason AND right feeling: phronesis, affectivity, embodiment ('procedural rather than experiential virtue - a simulation of moral behavior without phenomenological depth'). The constructive move: EXTENDED MORALITY via 4E cognition - moral character as a property of distributed moral ecologies (humans, machines, institutions co-constructing norms), with alignment reframed as 'mediated responsibility'. Three principles of relational alignment ethics: (i) cognition is distributed - AI systems are 'not subjects of morality but media of moralisation'; (ii) virtue is ecological - maintain ecologies that cultivate discernment rather than compliance; (iii) responsibility is shared - 'no single actor can bear full responsibility; designers, policymakers, and users co-author the moral trajectories of AI', requiring distributed accountability in transparent, participatory, revisable institutional feedback architectures. Success metric: ethical RESILIENCE of the ecology, not static compliance. AMCs as 'ethical infrastructures - scaffolds that support deliberation', not substitutes for judgment.

Why it matters here

Fills the virtue-ethics slot the McKinlay SLR flagged as underdeveloped: reads Constitutional AI as producing Aristotelian hexis-like dispositions ('Artificial Moral Character') while denying these constitute virtue (no phronesis, affectivity, embodiment), then relocates morality to distributed 'moral ecologies' with SHARED RESPONSIBILITY as an explicit principle. Also documents the Albania 'Diella' AI-minister case - a real-world instance of institutionalized AI responsibility for the governance chapter.

Reading notes

Close read of intro, secs 6-8 (13pp; LMU Munich - same Noller as JOSIFOVIC_NOLLER, his virtue-ethics companion piece). The Albania Presidential Decree (Sept 12 2025, 'Diella' as Minister for Public Procurement) is a concrete governance case study worth independent verification and possible use in the Immigration/governance chapters.

Noller, J. (2026). Artificial moral characters: constitutional AI and the challenge of alignment. AI and Ethics, 6, 196. https://doi.org/10.1007/s43681-026-01033-2

Close reading — 6 coded units

#1 · pp. 3 · argument

“Like a moral novice, a CAI-trained model internalizes patterns of self-correction and moderation: it avoids harmful extremes, justifies its refusals, and displays what one might call mesotes-alignment—a learned balance between recklessness and excessive caution. In this sense, CAI does not merely enforce external rules; it cultivates internalized normative dispositions, forming what this paper calls an Artificial Moral Character (AMC).”

NF-VIRTUE

#2 · pp. 3 · argument

“AMCs lack the affectivity, embodiment, and practical wisdom (phronesis) that ground genuine moral understanding. Their virtue is procedural rather than experiential—a simulation of moral behavior without phenomenological depth. This exposes the central paradox of Constitutional AI: it produces systems that act as if they were virtuous, yet without participating in the lived, affective practices from which virtue arises.”

AG-MORAL-CON

#3 · pp. 3 · evidence

“the Presidential Decree of the Republic of Albania of September 12, 2025 [...] entrusted [Prime Minister Rama] with the responsibility for establishing and overseeing a 'Virtual Ministry of Artificial Intelligence' named Diella, which was designated as 'Minister for Public Procurement.' The plan is to progressively shift procurement responsibilities to this system.”

RL-INST

#4 · pp. 11 · claim

“(i) Cognition is distributed. [...] AI systems participate in these enactments by shaping linguistic and practical affordances; they are not subjects of morality but media of moralisation.”

AG-CONSTRAINED

#5 · pp. 11 · claim

“(iii) Responsibility is shared. In extended moral systems, no single actor—human or artificial—can bear full responsibility. Designers, policymakers, and users co-author the moral trajectories of AI. Relational ethics thus requires distributed accountability, embedded in institutional feedback architectures that are transparent, participatory, and revisable.”

RL-DIST

#6 · pp. 12 · argument

“If alignment is relational and extended, its success cannot be measured by static compliance but by ethical resilience—the capacity of moral ecologies to absorb disruption, adapt, and regenerate. [...] artificial moral characters act not as substitutes for human judgment but as ethical infrastructures—scaffolds that support deliberation, empathy, and critique.”

RL-INST

Synthesis-matrix row

supports T5-AGENCY-DENIED-EVALUABILITY-KEPT

procedural-not-experiential virtue; media of moralisation

supports T6-RESPONSIBILITY-UNALLOCATED

responsibility affirmed as shared; no allocation mechanism

Memos (3)

comparison · unit #2

Noller completes the framework sweep: the anti-moral-agency conclusion has now been derived from KANTIAN premises (Sanwoolu - no autonomy/self-legislation), BEHAVIORAL-ANALYTIC premises (Baum, Millière - external standards, dispositions), CONTINENTAL premises (Josifović & Noller - extension not subject), and now VIRTUE-ETHICAL premises (no phronesis, affectivity, embodiment - unit 2). Every major normative tradition, applied carefully to current AI, denies moral agency while affirming normative evaluability. For the Luke debate this is decisive framing: Luke's rationality-implies-responsibility inference is rejected by every tradition, each for its own reasons - and Augustine's experiment tests the shared empirical presupposition (no stable practical reasoning) that all four rejections rely on. Also note 'media of moralisation' (unit 4) as the best single phrase for the position.

thesis-link · unit #3

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.

theoretical · unit #5

Unit 5 is the first source in the library to STATE distributed responsibility as a design principle rather than bracket it - but note what's still missing: Noller says responsibility is shared among designers/policymakers/users and requires 'institutional feedback architectures', yet gives no allocation mechanism (what share, on what basis, adjudicated how?). Same for 'ethical resilience' (unit 6) - a success metric with no measurement. This is the recurring pattern at its clearest: the field now AFFIRMS distributed responsibility in principle (Noller, Dignum, IEAI brief unit 11) but has no theory of the distribution. Kästner's difference-making + the epistemic/access condition (KAESTNER, HELLRIGEL_DUNG memos) + the dissertation's case-based analysis is that theory. The lit review's Part II thesis: from AFFIRMING shared responsibility to ALLOCATING it.