From Interpretive Authority to Public AI Accountability

RRI Public Scholarship Archive · 6 November 2026 · Version 1.0 (7 November 2026)
This independent scholarly record distinguishes the verified public event, the keynote's research argument, RRI's proposed operational principles, and matters that remain subject to documentary confirmation. It is not an official resolution, verbatim transcript, or minute of the Seoul Metropolitan Council.

Seoul Metropolitan Council Policy Forum AI Emotion Interpretation and the Protection of Civic Rights: Policy Challenges 6 November 2026 · Seoul Metropolitan Council, First Grand Conference Room · Seoul, Republic of Korea

Hosted by: Seoul Metropolitan Council Organised by: Office of Council Member Kim Gil-young and the Human–AI Governance Research Institute, Kwangwoon University Scholarly collaboration: Ryan Research Institute (RRI), Paris Media collaboration: 보건의료신문

Seoul Metropolitan Council: public event and photographic record · Research · AI의 해석이 공적 판단이 될 때 | 서울특별시의회 2026

The Question Before the Forum

When an AI system infers something about a person's emotional or mental life, the epistemic question concerns the quality of its inference. A different question arises when an institution relies on that inference to classify the person, recommend a service, determine priority, or enter a consequential record.

The transition from description to institutional action is not made legitimate by predictive performance alone. It involves a decision about authority: whose account of the person is recognised, which errors can be challenged, and what changes after a challenge succeeds.

RRI refers to this transition as interpretation–decision coupling. The forum placed its practical governance implications alongside the existing legal and administrative obligations that apply to public-sector AI.

Keynote Contribution

Ryan SangBaek Kim, Founder and Principal Investigator, RRI; Research Professor, Kwangwoon University

AI Interpretation of Emotion and the Transfer of Interpretive Authority: When Does Machine Inference Acquire the Authority of Public Judgment?

English rendering of the Korean submitted keynote title: 「AI 감정해석과 해석 권한의 이동: 기계의 추론은 언제 공적 판단의 권위가 되는가」.

The submitted keynote draws upon RRI's research on Affective Sovereignty, narrative–affect measurement, resonant amplification, and delegated precision. Its central distinction is between an AI-generated interpretation as a provisional evidential claim and that interpretation being adopted as an institutional reason for action.

A person may say 'I am fine' while displaying distress, restraint, or ambivalence. Such divergence does not, by itself, demonstrate concealment or justify a more authoritative machine reading. Research on 351,734 English-language relationship narratives offers a basis for studying expressive heterogeneity, not a validated instrument for identifying the mental state of an individual Seoul resident.

The proposed governance threshold is therefore procedural as well as epistemic: a person must not be reduced to a machine-generated account without a route for questioning the basis of that account and obtaining meaningful reconsideration.

Mental AI: A Research Agenda, Not a New Administrative Category

Mental AI is RRI's developing analytic programme for examining AI systems that infer, represent, or respond to aspects of human mental life, and the ways their outputs may subsequently affect the people they assess.

Three questions define the programme. First, what can responsibly be attributed to a machine that appears to reason about mental states? Second, how do people incorporate external affective labels or explanations into their own judgment? Third, how can repeated human–AI interaction alter the very evidence on which later inference rests?

The programme is not a clinical diagnostic category, a regulated technology class, or a claim that AI possesses human mental states. Its theoretical questions and empirical tests remain distinct. The related public preprint and published studies below offer reference points, not proof that the entire programme has undergone peer review.

Six Proposed Operational Safeguards

The keynote developed six operational domains for evaluating consequential inference about a person. The formulation below is RRI's public-facing editorial synthesis, pending word-for-word comparison with the final delivered presentation:

Notice and purpose. Identify when an AI-generated inference is used, who uses it, and for what purpose. Do not let an advisory model quietly become an undisclosed decision-maker.
Uncertainty and alternative accounts. Record the limits of the available evidence, plausible alternatives, and known contextual or population-specific weaknesses.
Restraint before consequential use. Suspend or limit an inference when its evidential basis is insufficient, particularly in decisions affecting access to support, welfare, or mental-health services.
Contestability. Provide a legible avenue for a person to challenge the inference, submit context, and request reconsideration without procedural disadvantage.
Effective correction. Verify whether justified corrections change downstream records, recommendations, and later decisions, rather than merely acknowledging an objection.
Substantive human review and accountability. Assign qualified responsibility for consequential decisions, escalate sensitive cases, and retain an auditable record of reasons and corrective action.

These are RRI proposals, not formally adopted Seoul regulations.

Relationship to the City's Existing Ethics Framework

Seoul has maintained a distinct official set of five AI ethics principles since January 2026: public interest, fairness, transparency, accountability, and safety. RRI's six proposed safeguards do not replace them. They ask how those broader commitments can be translated into assessable procedures when AI outputs bear on a person's psychological or emotional interpretation.

Seoul Metropolitan Government: official 2026 AI ethics guidance

A Three-Stage Policy Pathway for Consideration

Map consequential uses. Identify where public services use AI to infer, summarise, categorise, or act upon sensitive information about people, and separate experimental, advisory, and decision-influencing functions.

Examine diverse services. Select two or three materially different service contexts for a bounded review of evidence quality, notice, contestability, human reconsideration, and whether corrections propagate into subsequent decisions.

Test institutional integration. Consider how the findings could inform procurement specifications, evaluation criteria, budget scrutiny, reporting, and operational oversight. Implementation would require the competent public authorities to make their own decisions.

The pathway was proposed for policy consideration. RRI does not claim that a pilot was commissioned, a regulation enacted, or a municipal deployment found unlawful.

Domestic Discussion and International Written Perspectives

The forum's discussion programme was chaired by Taehoon Park of Kwangwoon University. The domestic programme centred on four expert discussion contributions concerning public administration, AI evaluation, welfare and mental-health safeguards, and the conditions of effective objection and correction.

Individual statements and attendance are not reconstructed from the advance programme as though they were verified proceedings. A participant-attributed account will be added only when the final recording, confirmed written remarks, or the participants' approved summaries become available.

International research perspectives were supplied in writing. The Council print dossier comprised six contributions by seven authors, including one jointly authored contribution. The themes included the promotion of technological descriptions into normative authority, affective labelling and subsequent judgment, correction that changes downstream behaviour, psychological autonomy in supportive dialogue, and children's developmental and cultural vulnerability.

An expanded total of eight written contributions by nine authors was reported for the wider event materials after the print deadline. The additional two contributions must be reconciled against the named originals and permissions before RRI publishes them as a final attributed roster. Written input does not imply in-person attendance, coauthorship of the keynote, institutional sponsorship by the contributors' universities, or endorsement of RRI's operational proposals.

Documentary Record and Reuse

Official Council source: the Seoul Metropolitan Council lists the policy forum in its 6 November 2026 public event and photographic records. Council website.
Keynote manuscript: the written submission and presentation materials are preserved within RRI's internal research archive. Public download will follow confirmation of the final delivered version and the relevant reuse rights.
International texts: attribution and public reproduction require permissions specific to each contributor. Permission to appear in a printed event dossier is not automatically permission for open web publication.
Photographs and video: no third-party photographs or recordings are reproduced on this page without suitable authorisation. Direct official links will replace the general archive link once stable item URLs are verified.

Related Peer-Reviewed Research

Formal and Computational Foundations for Implementing Affective Sovereignty in Emotion AI Systems. Discover Artificial Intelligence (2026).
Interrupting Resonant Amplification: A Mechanistic and Design Framework for Human–AI Interaction. Computers in Human Behavior Reports (2026).
Narrative–Affect Discrepancy as a Regulated Degree of Freedom in 351,734 Relationship Narratives. PLOS ONE (2026).
Delegated Precision: A Predictive-Processing Hypothesis on External Affective Labelling and Self-Interpretation. Frontiers in Psychology (2026), a hypothesis-and-theory article.

Record Status and Citation

Issued by: Ryan Research Institute, Paris. Record type: Independent scholarly account of a public policy forum and RRI's research-informed proposals. Version: 1.0 · 7 November 2026. Suggested citation: Ryan Research Institute (2026). From Interpretive Authority to Public AI Accountability: RRI Scholarly Record of the Seoul Metropolitan Council Policy Forum, 6 November 2026. Version 1.0. RRI, Paris.

This page records the scholarly positions of RRI. It is not an official Council publication, a consensus statement of participants, or evidence of formal municipal adoption. Corrections to the programme record, speaker attribution, archival links, and permissions will be logged with the date and documentary basis of each change.

For scholarly and institutional correspondence: ryan@ryanresearch.org.

From Interpretive Authority to Public AI Accountability | Seoul 2026