Ryan Kim
Founder & Principal Investigator
Affective Sovereignty · Interpretive Authority · Human–AI Interaction · AI Evaluation & Governance
Ryan Research Institute
Ryan Research Institute is a Paris-based independent research institute studying interpretive authority in AI-mediated emotional life.
Across its research program, RRI asks a single question: when systems infer, classify, or act on emotion, who retains the authority to decide what that feeling means?
Affective Sovereignty is the institute’s core program for protecting first-person interpretive standing under technological and institutional mediation. The work connects emotion AI governance, human–AI interaction, narrative–affect geometry, philosophy of mind, and institutional design.
This site presents peer-reviewed research, open research resources, books, visual research materials, and selected public scholarship.
Subscribe to the research newsletter: profryankim.substack.com
Latest
24 Aug 2026 | Public Engagement
Ryan Kim was appointed Senior Specialized Reporter & Editorial Board Member at a Korean healthcare outlet, 보건의료신문, for its newly established AI·Mental Health / Interpretive Rights section. → Announcement (Korean)
24 Aug 2026 | Public Engagement | Interview
Interview | Who Gets to Interpret Our Minds in the Age of AI?
In a new interview with the same Korean healthcare outlet, Ryan Kim discusses why AI systems that infer emotion, mental states, vulnerability, and risk raise questions that cannot be reduced to predictive accuracy alone. The conversation addresses Affective Sovereignty, Interpretive Authority, meaningful human oversight, correction propagation, Mental AI, and the distinction between using AI as a valuable second opinion and allowing it to become the first definition of a person.
Affective Thermodynamic Relationship — Published
Communications AI & Computing, 2026
An empirical information-theoretic scaling relationship for interpretative collapse under normative conflict. The study separates model uncertainty from evaluative uncertainty and asks when AI evaluation becomes too quick to collapse genuine conflict into a single answer.
→ Article DOI · → Behind the Paper: The Curve Was Stable. The Judgment Was Not.
Narrative Complexity and Affective Energy — Published
Acta Psychologica, 270, 107689 (2026)
Across 351,734 relationship narratives, narrative complexity was almost unrelated to net valence and only weakly associated with cancellation-robust affective energy. The result separates net emotional direction from affective magnitude without treating either as a direct measure of a person’s underlying emotional state.
→ Article DOI · → Open Data · ANAD v1.4.0
AI Watermarking, Authorship & Interpretive Authority — Current Work
New research agenda on the relationship between technical provenance signals, intellectual contribution, and institutional evidentiary use.
→ English essay: Can Claude’s Watermark Prove AI Authorship? · → Korean edition
→ AI Times Korea contributed essay (Korean)
Status: Research agenda initiated August 2026 · Empirical protocol in development
Algorithmic Affective Blunting — Published
Discover Artificial Intelligence, 2026 · DOI: 10.1007/s44163-026-01573-w
Mental AI — Public Preprint
Post-Deployment Criterion Endogeneity in the Causal Loop of Human Mental Life · DOI: 10.20944/preprints202607.2118.v1
Policy Translation — AI Times Korea
Public policy work on interpretive rights, correction, and meaningful human review in high-impact AI, published 11 August 2026.
Featured Visual Evidence
Normative Collapse Follows a Rate Law
The Affective Thermodynamic Relationship · Communications AI & Computing (2026)
→ Read the article · → Behind the Paper: The Curve Was Stable. The Judgment Was Not.
Emotional AI Needs a Stress Test
Algorithmic Affective Blunting measures how emotional interpretation in a large language model degrades under rising semantic stress.
In a standardized single-model setting, mean Affective Degradation Index rose from 0.16 in Control to 2.92 in Extreme exposure.
The visual citation card turns the paper’s empirical collapse curve into a reusable research signal: emotional AI should be evaluated not only for apparent empathy, but for interpretative robustness under stress.
Status: Published in Discover Artificial Intelligence (2026).
DOI: https://doi.org/10.1007/s44163-026-01573-w
From Data to Geometry to Measurement
ANAD v1.4.0
351,734 narratives represented through an open derived-feature research resource
↓
PLOS ONE
Mapping the geometry of narrative–affect discrepancy
↓
Acta Psychologica
Distinguishing net valence from affective energy
Together, these studies ask not whether narrative and emotion simply “match,” but what different forms of mismatch reveal about human expression and the systems designed to interpret it.
Featured Publication
Narrative–Affect Discrepancy Published in PLOS ONE (2026)
Narrative–affect discrepancy as a regulated degree of freedom in 351,734 relationship narratives
PLOS ONE, 21(5), e0348715 (2026)
DOI: https://doi.org/10.1371/journal.pone.0348715
Media coverage:
This paper maps narrative complexity and linguistically inferred affective intensity across 351,734 English-language relationship narratives. It treats narrative–affect discrepancy not as residual noise, but as a regulated expressive degree of freedom.
The study identifies four empirically separable expressive regimes: coupled expression, strategic understatement, strategic overstatement, and collapse. As a comparative probe, it projects an RLHF-aligned language model into the same coordinate system and finds that the model occupies an approximately 1.70× smaller expressive region than humans.
The paper contributes a reproducible geometric basis for comparing expressive degrees of freedom across human populations and aligned AI systems.
Behind the Paper:
How Emotional Discrepancy Stopped Looking Like Error
Long-form essay:
The Missing Coordinate in the RLHF Diversity Debate
Featured in PsyPost
A new study mapped 350,000 relationship stories and found a communication style AI struggles to copy
PsyPost, May 24, 2026
PsyPost covered Ryan SangBaek Kim’s PLOS ONE study on narrative–affect discrepancy, highlighting how people often communicate distress through restraint, understatement, silence, or expressive collapse, and why aligned AI systems may struggle to reproduce these quieter forms of human emotional expression.
Read the PsyPost article · Read the PLOS ONE paper
Selected Recent Work
Recent published work spans AI evaluation and governance, narrative–affect measurement, human–AI interaction, and open research resources.
The affective thermodynamic relationship: an empirical information-theoretic scaling relationship for normative-conflict collapse in large language models
Communications AI & Computing (2026)
Narrative Complexity Is Weakly Coupled With Affective Energy, Not Net Valence, in Relationship Narratives
Acta Psychologica (2026)
Algorithmic affective blunting quantifies the collapse curve of interpretative failure in large language models
Discover Artificial Intelligence (2026)
Narrative–affect discrepancy as a regulated degree of freedom in 351,734 relationship narratives
PLOS ONE (2026)
Formal and Computational Foundations for Implementing Affective Sovereignty in Emotion AI Systems
Discover Artificial Intelligence (2026)
ANEST Narrative–Affect Dataset (ANAD v1): A Large-Scale Derived Feature Resource for Quantifying Narrative–Affective Discrepancy
Data in Brief (2026)
Additional manuscripts are in editorial assessment and peer review across psychology, philosophy, human–AI interaction, and AI governance; confidential venue information is not listed here.
Research
RRI advances a research program on emotion, selfhood, interpretive authority, and human-AI interaction.
Current lines of work include:
Affective Sovereignty
A normative and computational framework for protecting personal authority over emotional meaning in AI-mediated systems.
Affective Suppression Fatigue
A dynamical account of how chronic suppression can lead to numbing at low intensity and collapse at high intensity.
Algorithmic Affective Blunting and the Affective Thermodynamic Relationship
A research line on interpretive failure in large language models under affective and normative stress.
Narrative-Affect Geometry and Measurement
A research program examining how narrative structure, net emotional direction, and affective magnitude relate and diverge in human expression. ANAD provides the derived-feature infrastructure; the PLOS ONE study maps the geometry of narrative–affect discrepancy; and the Acta Psychologica study shows that apparent coupling between narrative complexity and affect depends critically on how affect and discrepancy are operationalised.
Resonant Amplification Framework
A mechanistic account of human-AI attachment, linguistic reinforcement, and intervention design.
Ecology of Inquiry
A metascientific line asking how research systems make certain questions thinkable, fundable, and governable.
Featured Book
The Interpreter's Seat
How the Power over Human Emotion Has Shifted
“To feel is biology. To interpret is power.”
This book traces the history of interpretive authority, from older systems of naming emotion to contemporary platforms that classify, stabilize, and act on feeling.
Written for readers already past introductory psychology, it brings together emotion, self-interpretation, dependence, narrative, and the politics of affective technology in a work of public intellectual nonfiction.
An independent POD first edition has been published and tested with paying readers. A revised and expanded commercial edition is now being prepared for publisher re-submission.
Recent
Narrative–Affect Discrepancy published
PLOS ONE (2026)
351,734 relationship narratives mapped into narrative–affect geometry; four expressive regimes and 1.70× aligned-model expressive-area contraction.
DOI · Behind the Paper · Essay
Affective Sovereignty published
Discover Artificial Intelligence (Springer Nature, 2026)
Formal framework for emotional AI governance introducing Sovereign-by-Design and DRIFT.
Resonant Amplification Framework published
Computers in Human Behavior Reports (Elsevier, 2026)
Mechanistic model of human–AI interaction and Cognitive Circuit Breakers.
ANAD v1 dataset released
Data in Brief (Elsevier, 2026)
351,734-text corpus for narrative–affect discrepancy research.
Media coverage and public writing
PsyPost (May 2026)
A new study mapped 350,000 relationship stories and found a communication style AI struggles to copy.
MIT Technology Review Korea (2026)
A series of essays on emotion AI, interpretive authority, and affective sovereignty.
AI Times (February 2026)
The Real Risk of Emotion AI Is Not Accuracy, but the Transfer of Interpretive Power.
Read PsyPost · Read MIT Technology Review Korea · Read AI Times
Music
RRI also archives original musical work, from orchestral textures to contemporary minimalism.
New Album: 살아내는 중입니다 (2025)
A full-length emotional narrative across 11 tracks, released on major platforms.
Contact
Email: ryan@ryanresearch.org
ORCID: https://orcid.org/0009-0006-2751-496X
EurAI https://eurai.org/
BCS – The Chartered Institute for IT (SGAI) https://www.bcs-sgai.org/
KAIA – Korean Artificial Intelligence Association https://aiassociation.kr/
KASBA – Korean Academic Society of Business Administration https://kasba.or.kr/
IASEAI – International Association for Safe & Ethical AI https://www.iaseai.org/
SIpEIA – Italian Society for Ethics in Artificial Intelligence https://sipeia.it/
Nature Portfolio Community https://go.nature.com/users/ryan-sangbaek-kim
© 2026 Ryan Research Institute. All rights reserved.
A research institute for emotion, interpretation, and the ethics of AI.
ResearchPublicationsBooksMusicAboutNewsletterHome(KR)