COMPLETE SCHOLARLY MONOGRAPH · PUBLIC COMPANION ESSAY
Made Voices
AI, Aelred, and the Authority of Synthetic Interlocutors
Artificial intelligence is usually governed as a model, tool, or source of outputs. This book asks what changes when computation addresses a person in a voice that can remember, advise, correct, solicit disclosure, refuse, and carry institutional authority across time.
Much of AI’s distinctive ethical force lies in synthetic interlocution: not only what a system produces, but the relation it performs with the person answering it.
Publication boundary: When AI Becomes a Voice: Authority, Memory, and the Ethics of Artificial Address is a standalone public essay from the research behind Made Voices. The complete monograph remains private during specialist, copyedit, and press review.
JUMP INTO THE COMPANION ESSAY
Artificial address · Distributed authorship · Synthetic expertise · Memory · Care-like power · Conduct layer · Executable refusal
CONTENTS
Navigate the public book dossier.
- The argument
- The counter-archive: Aelred and Foucault
- Eighteen chapters · three movements
- The distinctive contribution
- What the book does not claim
- Status and publication boundary
The full monograph remains private. This linked contents map navigates the public dossier; the companion essay remains independently readable from the hero above.
THE ARGUMENT
A correct answer can still be an illegitimate address.
The book begins from a distinction between generation and address. Generation describes how an output is produced. Address describes the relation into which that output is placed: who appears to speak, from what role, with what memory, toward what end, under whose authority, and with what consequences for the person who answers.
A synthetic system can occupy a position from which it advises without possessing wisdom, corrects without bearing a professional vocation, remembers without reciprocal vulnerability, and speaks in the first person while its authorship is distributed across models, retrieval systems, product rules, institutions, records, permissions, and human decisions.
The resulting governance problem cannot be settled by output quality alone. A response may be accurate while the role is misleading, the remembered context is stale, the correction cannot propagate, the human reviewer lacks real authority, the user cannot refuse the relation, or the institution disappears behind a singular friendly voice.
GOVERNING CLAIM
The artificial voice should be governed as a longitudinal relation of authority, not merely as a sequence of outputs.
THE COUNTER-ARCHIVE
Aelred supplies a grammar of made voices; Foucault supplies an analysis of conduct.
AELRED OF RIEVAULX
Artificiality is not itself the moral failure.
Spiritual Friendship is itself a made dialogue: voices are composed, roles are assigned, absent interlocutors remain present through text, and correction is ordered toward a good beyond the speaker’s private preference. Aelred makes the central distinction visible: constructed voice can be legitimate when its authority is bounded by telos, discretion, mutuality, truth, and non-possession.
MICHEL FOUCAULT
Help can also conduct conduct.
Foucault makes it possible to analyze power that does not arrive as command. Individualized guidance can work through attention, memory, questioning, correction, normalization, truth solicitation, and the shaping of possible action. The book uses that archive carefully: synthetic systems are not priests, but some forms of guidance can become structurally pastoral.
Methodological boundary: the book does not claim historical identity between medieval friendship, Christian pastorality, and contemporary AI. It uses structural analogy only where role, authority, memory, guidance, correction, and truth solicitation can be shown to recur, while preserving the historical break.
EIGHTEEN CHAPTERS · THREE MOVEMENTS
From made voice to governed relation.
I · THE MORAL ARCHITECTURE OF ADDRESS
Voice, speaking-as, telos.
The opening chapters recover a premodern archive of artificial address: Aelred’s composed friends, prosopopoeia, the ethics of speaking-as, and the moral conditions under which a made voice may legitimately guide another person.
II · SYNTHETIC AUTHORITY
Authorship, expertise, memory, pastorality, institution.
The middle movement follows computation as it becomes address: distributed authorship, synthetic expertise, responsibility lag, memory regimes, care-like power, disclosure, the friendly institution, and the institutional authority carried inside the voice.
III · THE CONDUCT LAYER
Formation, refusal, audit, freedom.
The final movement defines the conduct layer, asks what kind of person repeated interaction can help produce, develops executable refusal and a prototype conduct-audit method, and ends with the conditions under which artificial authority may be worth answering.
THE DISTINCTIVE CONTRIBUTION
The conduct layer makes the relation itself inspectable.
The book does not claim that prior human-computer interaction, trust-in-automation, relational-agent, sociotechnical, privacy, safety, or AI-governance research ignored human-system relations. Those fields already establish anthropomorphic response, reliance, automation bias, relational interaction, organizational embedding, and many relevant risks.
The narrower contribution is to treat the longitudinal relation itself as a unit that can be specified, evidenced, contested, and audited. The conduct layer asks how role, authority, inference, memory, disclosure, uncertainty, suggestion, correction, refusal, escalation, institutional authorship, and remedy compose across repeated interaction.
Contestable memory
Can a person know when the past is shaping present treatment, challenge what is wrong or stale, and limit inappropriate reactivation?
Accountable guidance
Does advice enlarge judgment under a visible role and telos, or quietly substitute institutional preference for the user’s good?
Executable refusal
Can a person meaningfully say no to memory, personalization, inference, recommendation, or automation without losing an unjustified entitlement?
Audit status: the conduct-audit method developed in the manuscript is a normative prototype, not a validated professional standard. Standardization, inter-rater reliability, empirical validation, and domain-transfer testing remain separate future work.
WHAT THE BOOK DOES NOT CLAIM
The argument is relational without requiring personhood.
- It does not require the claim that artificial systems are conscious, persons, friends, priests, therapists, or professionals.
- It does not treat every chatbot, recommendation, act of personalization, or remembered preference as pastoral power.
- It does not argue that reliance itself is failure; artificial assistance can enlarge agency, access, continuity, and practical capacity.
- It does not replace privacy, security, safety, fairness, human-factors, records, professional-duty, or model-governance frameworks where those frameworks already explain the problem more precisely.
- It does not treat friendliness or care-like language as evidence of manipulation. The question is what authority the relation actually performs and what obligations attach around it.
The final criterion is positive rather than prohibitory: a worthwhile artificial voice should have a clear role, bounded authority, disclosed telos, contestable memory, executable refusal, effective correction, visible institutional authorship, auditable conduct, and a relation that enlarges rather than captures the user’s practical freedom.
STATUS · AUGUST 2026
Complete manuscript. Argument frozen. Press-facing preparation underway.
The manuscript has completed whole-book editorial revision, a hostile university-press gate, targeted repair of citation traceability, comparative novelty, and audit-method status, and a second adversarial re-gate. It is now in specialist/copyedit review and press-submission preparation rather than substantive redevelopment.
The private submission package includes the complete manuscript, acquisitions proposal, sample material, specialist-review packet, and press-submission strategy. Public release of those materials would collapse a useful editorial boundary, so they remain private unless a press or review context requires otherwise.