Publications
Updated September 2026 — Ryan Research Institute
Ryan Research Institute develops a research programme in Human–AI Governance, Interpretive Authority, Affective Sovereignty, AI evaluation, and the institutional consequences of computational inference about people. Its peer-reviewed work spans affective and relational AI, narrative–affect geometry, large-language-model evaluation, measurement validity, and governance.
This page presents public research outputs with DOI, journal records, open research resources, and selected public-facing coverage. Manuscripts under confidential peer review are not listed here by venue.
Published Peer-Reviewed Articles · 8
The affective thermodynamic relationship: an empirical information-theoretic scaling relationship for normative-conflict collapse in large language models (2026)
Communications AI & Computing
Status: Published · Version of Record live
DOI: https://doi.org/10.1038/s44488-026-00006-y
→ Reports an empirical information-theoretic scaling relationship for interpretative collapse under normative conflict. The study separates model uncertainty from evaluative uncertainty and treats conflict preservation, rather than premature convergence, as a distinct evaluation problem.
Related research essay: The Curve Was Stable. The Judgment Was Not. · Springer Nature Research Communities, 22 August 2026
→ Ryan Sangbaek Kim on Springer Nature Research Communities
Narrative Complexity Is Weakly Coupled With Affective Energy, Not Net Valence, in Relationship Narratives (2026)
Acta Psychologica, 270, 107689
Status: Published · Open Access · Version of Record
DOI: https://doi.org/10.1016/j.actpsy.2026.107689
→ An analysis of 351,734 naturalistic relationship narratives showing that narrative complexity is almost unrelated to net valence and only weakly associated with cancellation-robust affective energy. The study distinguishes net emotional direction from affective magnitude and uses a GAM-normalised residual measure for its main narrative–affect discrepancy analysis.
Open research resource: ANAD v1.4.0 · Zenodo
Dataset article: Data in Brief, 66, 112643
Algorithmic affective blunting quantifies the collapse curve of interpretative failure in large language models (2026)
Discover Artificial Intelligence
Status: Published
DOI: https://doi.org/10.1007/s44163-026-01573-w
→ Introduces Algorithmic Affective Blunting (AAB), a dose-dependent degradation of affective interpretive coherence in a large language model under semantic stress. In a standardized single-model setting, mean Affective Degradation Index rose from 0.16 in Control to 2.92 in Extreme exposure.
Visual citation card: Emotional AI Needs a Stress Test
Narrative–Affect Discrepancy as a Regulated Degree of Freedom in 351,734 Relationship Narratives (2026)
PLOS ONE, 21(5), e0348715
DOI: https://doi.org/10.1371/journal.pone.0348715
→ Models narrative–affect discrepancy as a regulated expressive degree of freedom and shows that an RLHF-aligned language model occupies an approximately 1.70× smaller expressive region than humans.
Media coverage and public writing:
PsyPost: A new study mapped 350,000 relationship stories and found a communication style AI struggles to copy
MIT Technology Review Korea: How Emotion AI Misses the Structure of Human Feeling
https://www.technologyreview.kr/기고-감정을-읽는-ai가-정작-놓치는-인간-감정의-구조/
Formal and Computational Foundations for Implementing Affective Sovereignty in Emotion AI Systems (2026)
Discover Artificial Intelligence
DOI: https://doi.org/10.1007/s44163-026-01000-0
→ Introduces Sovereign-by-Design, the DRIFT protocol, and auditable alignment metrics for protecting interpretive authority in emotion-aware AI systems.
Interrupting Resonant Amplification: A Mechanistic and Design Framework for Human–AI Interaction (2026)
Computers in Human Behavior Reports, 21, 100975
DOI: https://doi.org/10.1016/j.chbr.2026.100975
→ Defines feedback-driven escalation loops in human–AI interaction and proposes Cognitive Circuit Breakers for interrupting amplification without erasing user agency.
ANEST Narrative–Affect Dataset (ANAD v1): A Large-Scale Derived Feature Resource for Quantifying Narrative–Affective Discrepancy (2026)
Data in Brief
DOI: https://doi.org/10.1016/j.dib.2026.112643
→ Provides a large-scale derived feature resource for measuring narrative–affect discrepancy across 351,734 relationship narratives.
DefMoN: A Reproducible Framework for Theory-Grounded Synthetic Data Generation in Affective AI (2025)
Machine Learning with Applications, 23, 100817
DOI: https://doi.org/10.1016/j.mlwa.2025.100817
→ Introduces a reproducible framework for generating theory-grounded synthetic data around defensive motivational nodes and affective AI.
Public Preprints and Open Research Resources
Mental AI: Post-Deployment Criterion Endogeneity in the Causal Loop of Human Mental Life (2026)
Public preprint
DOI: https://doi.org/10.20944/preprints202607.2118.v1
→ Develops post-deployment criterion endogeneity as a problem in which AI inferences can alter the human states, evidence, and criteria later used to evaluate those same inferences.
From Composite Scores to Latent Variables: Reassessing Prospective Associations Between Social Substitution and Emotional Closeness in Human–AI Attachment (2026)
Corrected public research version · Zenodo v2.0
DOI: https://doi.org/10.5281/zenodo.21971714
→ Public corrected version. Study 4 uses a one-week interval; the prior directional claim is superseded, and the corrected analysis does not establish that social substitution precedes emotional closeness or the reverse.
Open Data and Research Resources
ANAD v1 · Narrative–Affect Research Resource
Current public version: v1.4.0
Version DOI: https://doi.org/10.5281/zenodo.22135369
Concept DOI · all versions: https://doi.org/10.5281/zenodo.17632585
→ A privacy-preserving derived-feature research resource representing 351,734 relationship narratives. Version 1.4.0 aligns terminology and public documentation across the ANAD research line, standardises LoC as Level of Complexity, preserves the distinction between dataset-level discrepancy and the residual-based NADI used in Acta Psychologica, and restores the canonical public data and aggregate research artifacts.
Published dataset article: Data in Brief, 66, 112643
Historical cited version: v1.3.0 · DOI 10.5281/zenodo.18680687
Affective Sovereignty Research Resources
Implementation framework preprint:
https://doi.org/10.5281/zenodo.17154182
Minimal Declaration:
https://doi.org/10.5281/zenodo.16992479
→ Public research materials supporting the conceptual and design foundations of Affective Sovereignty.
NADI / ANEST Research Program Archives
Human baseline archive:
https://doi.org/10.5281/zenodo.17864574
ANEST theory archive:
https://doi.org/10.5281/zenodo.17864654
Narrative–affect geometry archive:
https://doi.org/10.5281/zenodo.17864774
→ Open archive records for the broader narrative–affect discrepancy and expressive-geometry research program.
Public Scholarship and Coverage
From Research to Public Deliberation: AI, Mental Health, and Interpretive Rights
On 24 August 2026, Korean healthcare outlet 보건의료신문 announced the launch of a dedicated AI·Mental Health / Interpretive Rights editorial section and the appointment of Ryan Kim, Founder & Principal Investigator of RRI, as Senior Specialized Reporter & Editorial Board Member.
The initiative extends several strands of RRI research into sustained public discussion, including Affective Sovereignty, Interpretive Authority, Mental AI, human oversight, correction and contestability, and the governance of AI-generated inferences about human internal states. The section connects psychiatry and mental health, psychology and counseling, healthcare and welfare, digital health, law and policy, user perspectives, and human–AI interaction.
Its governing question is not only whether an AI interpretation is accurate, but who has authority over that interpretation, how a person may challenge or correct it, and what procedures should follow when such an inference enters a recommendation or decision process.
Special Feature: dedicated AI·Mental Health / Interpretive Rights section launched; Ryan Kim appointed Senior Specialized Reporter & Editorial Board Member.
→ Read the 24 August 2026 announcement
This record marks the section launch and appointment announcement. Future feature interviews and the Korean-language signature column will be listed separately when each article is actually published.
Springer Nature Research Communities
A continuing Behind the Paper series connecting RRI’s peer-reviewed work to its conceptual and methodological development. The 2026 sequence runs from Who Gets to Say How You Feel? through Affective Sovereignty, narrative–affect discrepancy, Algorithmic Affective Blunting, Mental AI, and the 22 August essay The Curve Was Stable. The Judgment Was Not.
→ View Ryan Sangbaek Kim’s Research Communities profile
Substack · Prof. Ryan Kim
Research-led essays extending peer-reviewed and public research into broader questions of AI evaluation, interpretive authority, relational AI, Mental AI, authorship, and provenance. Selected 2026 essays include The Human Mind Is Becoming AI’s Operating Environment, The Risk Is Not the Reply. It Is the Relationship., The Machine Is Certain. That Does Not Mean It Knows What You Feel., Your AI Can Read You. It Should Never Outrank You., The Missing Coordinate in the RLHF Diversity Debate, and Can Claude’s Watermark Prove AI Authorship?
PsyPost
A new study mapped 350,000 relationship stories and found a communication style AI struggles to copy
MIT Technology Review Korea
Essays on emotion AI, interpretive authority, and affective sovereignty.
https://www.technologyreview.kr/author/ryankim/
AI Times Korea
AI watermarking and authorship judgment — Korean-language contributed essay, 19 August 2026.
https://www.aitimes.com/news/articleView.html?idxno=214132
→ Distinguishes technical provenance from intellectual contribution and authorship. A watermark or detector signal may have evidentiary value, but it should not by itself determine misconduct, authorship, or other institutional judgments.
Interpretive rights, correction, and high-impact AI — Korean-language contributed policy essay, 11 August 2026.
https://www.aitimes.com/news/articleView.html?idxno=213803
→ Translates RRI’s interpretive-authority program into six operational principles for high-impact AI: purpose notice, uncertainty and alternatives, abstention, subject correction, correction propagation, and meaningful human review.
Notes
This page separates published articles, public preprints, open research resources, and public scholarship.
Manuscripts under confidential peer review are not listed here by journal venue until a public DOI, official record, accepted article status, or public archive record is available.
For research, reproducibility, validation, or theory-integration inquiries: