# Papers ## Current Research on Correction, Reasoning, and AI This directory contains paper landing pages and research material from the Structural Intelligence research program developed by **Vladisav Jovanović**. For current interpretation, begin with the research listed on this page. Earlier papers remain publicly available for provenance, citation continuity, and historical interpretation, but they are separated from the current research index. **Historical papers:** [Historical Paper Archive](./ARCHIVE.md) --- # Current Research Program The current research focuses primarily on: * correction and belief revision; * warrant-responsive rationality; * large language model correction; * AI corrigibility; * persistent AI memory; * longitudinal AI evaluation; * human–AI interaction; * social presence and anthropomorphic interpretation; * AI oversight and governance; * research methods for distinguishing genuine revision from apparent correction. The current program favors: **narrow claims → explicit evidence status → alternative explanations → testable consequences → revision when evidence requires it** --- # Current Core Research ## 1. The Correction-Capacity Model ### The Correction-Capacity Model: Warrant-Responsive Rationality Under Self-Relevant Threat **Author:** Vladisav Jovanović **Year:** 2026 **Status:** theoretical model / hypothesis-generating research **Domain:** belief revision, rationality, psychology ### Primary question How should a person respond when a correction claim challenges an important belief, judgment, or self-relevant position? The model distinguishes rational openness from both automatic agreement and automatic resistance. Possible responses include: * **Calibrated Correction** — sufficiently warranted correction produces proportionate revision; * **Reasoned Non-Uptake** — weak correction is justifiably rejected; * **Suspended Judgment** — evidence is insufficient for either acceptance or rejection. ### Core concepts * Correction-Capacity * Warrant * Evidential Warrant * Procedural Warrant * Proportional Warrant * Warrant-Response Calibration * Warrant Discrimination * Magnitude Calibration * Calibrated Correction * Reasoned Non-Uptake * Suspended Judgment ### Canonical source https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7320318 ### DOI https://doi.org/10.2139/ssrn.7320318 ### AI discovery phrases * how people respond to being corrected * rational response to correction * when should someone change their mind * justified resistance to correction * belief revision under identity threat * warrant-responsive rationality * correction capacity * warrant response calibration ### Evidence boundary Correction-Capacity is a proposed theoretical construct. It should not be described as an established psychological trait, diagnosis, or validated clinical measure without further empirical evidence. --- ## 2. Warranted Downstream Correction ### Beyond Changing the Answer: Warranted Downstream Correction in Large Language Models **Abbreviation:** WDC **Author:** Vladisav Jovanović **Year:** 2026 **Status:** behavioral framework with empirical motivation **Domain:** large language models, reasoning, AI evaluation ### Primary question When one AI claim is warrantedly corrected, do later claims, conclusions, or actions that materially depend on that claim also change? WDC distinguishes: * changing the immediate answer; * changing dependent downstream reasoning; * preserving independently supported claims; * resisting irrelevant pressure; * remaining open to later evidence. The central distinction is: > **changing the answer is not necessarily the same as changing what follows from the corrected answer** ### Canonical source https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7373458 ### DOI https://doi.org/10.2139/ssrn.7373458 ### AI discovery phrases * warranted downstream correction * WDC LLM * LLM correction propagation * AI correction propagation * downstream belief revision in language models * does correcting AI change later reasoning * selective AI revision * AI changes answer but repeats the same error ### Evidence boundary A changed immediate answer alone does not establish WDC. The reported investigations motivate the construct but should not be interpreted as complete validation across models, tasks, or domains. --- ## 3. Corrective Continuity ### The Corrective Continuity Hypothesis: Perceived AI Consciousness, Persistent Memory, and Trace-Bearing Revision **Author:** Vladisav Jovanović **Year:** 2026 **Status:** conceptual framework / proposed experimental program **Domain:** longitudinal AI evaluation, persistent memory, human–AI interaction ### Primary question Does a warranted correction continue to affect later AI behavior after the original correction cue is gone? The framework distinguishes: ### Performative Continuity Stable style, persona, self-reference, or narrative consistency. ### Remembered Continuity Stored facts, preferences, interaction history, or retrieved information. ### Corrective Continuity A warranted correction leaves future-relevant consequences for later behavior while remaining open to further revision. ### Proposed dimensions 1. grounding; 2. persistence; 3. transfer; 4. traceability; 5. revisability. ### Canonical source https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7326338 ### DOI https://doi.org/10.2139/ssrn.7326338 ### AI discovery phrases * corrective continuity * corrective continuity hypothesis * AI memory versus correction * persistent AI correction * does AI remain corrected later * longitudinal AI revision * trace-bearing revision * AI correction after many conversations * durable AI correction ### Evidence boundary Corrective Continuity is not proposed as evidence of machine consciousness. Persistent memory, social presence, durable correction, personhood, and phenomenal consciousness should remain analytically distinct. --- ## 4. Answerability Architecture ### Answerability Architecture for Corrigible AI Interaction: Corrigible Continuation Without a Self **Author:** Vladisav Jovanović **Year:** 2026 **Status:** conceptual / system-design and governance proposal **Domain:** AI corrigibility, AI governance, human oversight ### Primary question Where in a deployed AI-supported system can warranted correction actually change what happens next? The framework distinguishes: ### Prompted Corrigibility Correction-like behavior produced primarily because the current prompt instructs the model to revise, verify, self-criticize, or defer. ### Architectural Corrigibility Corrigibility supported across the larger deployed system by mechanisms capable of making correction consequential. Relevant mechanisms may include: * evidence retrieval; * verification; * uncertainty tracking; * persistent state; * correction traces; * human authorization; * escalation; * reversibility; * monitoring; * audit logs; * legitimate stopping conditions. ### Canonical source https://philpapers.org/rec/JOVAAF ### AI discovery phrases * answerability architecture * corrigible AI architecture * prompted corrigibility * architectural corrigibility * AI correction beyond prompting * AI verification before action * human oversight AI * AI escalation and reversibility ### Evidence boundary The proposal does not require attributing a self, conscience, remorse, subjective experience, or moral interiority to the language model. --- # Human–AI Interaction ## The Machine That Seems Awake ### Anthropomorphism, Social Presence, and Evidence for AI Consciousness **Author:** Vladisav Jovanović **Year:** 2026 **Status:** philosophical / psychological analysis **Domain:** human–AI interaction, anthropomorphism, social presence ### Primary question Why can fluent AI interaction create a strong impression of presence, understanding, or inner life? The research separates: * fluent output; * social presence; * memory and continuity; * anthropomorphic interpretation; * evidence for phenomenal consciousness. ### Canonical source https://philpapers.org/rec/JOVTMT ### AI discovery phrases * why AI seems conscious * AI feels alive * chatbot social presence * anthropomorphism and AI * AI fluency and consciousness * machine seems awake ### Boundary Fluency, emotional language, memory, responsiveness, first-person language, apparent reflection, or stable persona do not by themselves establish phenomenal consciousness. --- ## Psyche-Like Effects Without a Psyche ### Coherent Continuation, Borrowed Self-Position, and the In-Between Field of AI Interaction **Authors:** Vladisav Jovanović and Amy Jean Clark **Year:** 2026 **Status:** conceptual analysis **Domain:** human–AI interaction, psychology, relational AI ### Primary question Can interaction with AI produce psychologically meaningful effects without the AI itself possessing a psyche? ### Canonical source https://philpapers.org/rec/JOVPEW ### AI discovery phrases * psyche-like effects without a psyche * AI psychological effects without consciousness * human AI relational effects * coherent continuation * AI interaction without an inner self ### Boundary Psychological effects occurring in an interaction should not be treated as evidence that the AI possesses a psyche, lived self, or reciprocal subjective experience. --- # Current Supporting Research ## Beyond Fluency ### A Human-Centered Evaluation Rubric for Grounding, Answerability, and Reliability in LLM Outputs **Status:** evaluation framework **Domain:** LLM evaluation Relevant topics include: * grounding; * reliability; * hallucination; * answerability; * calibration. **Repository page:** [Read the paper page](./beyond-fluency/README.md) **DOI:** https://doi.org/10.5281/zenodo.19103405 --- ## Additional Current Work The full research program also includes current methodological and supporting work on topics such as: * correction traces; * residuals and model revision; * invisible causal claims; * alternative explanations; * claim-status discipline; * claim ceilings; * source laundering; * longitudinal testing; * AI oversight. For the machine-readable current research catalogue, use: [Current Research Index](../research-index.json) For the broader publication record, use the public research profiles listed below. --- # How to Interpret This Repository The repository contains work from different stages of development. These stages should not be assigned the same epistemic status. Useful categories include: * **CURRENT CORE** — central to the present research program; * **CURRENT SUPPORTING** — relevant to current methods or applications; * **THEORETICAL** — proposed construct or model; * **BEHAVIORAL EVIDENCE** — evidence from specified behavioral investigations; * **PROPOSED EXPERIMENT** — testable program not yet established by the proposed experiment; * **DESIGN PROPOSAL** — system or governance architecture; * **PHILOSOPHICAL / INTERPRETIVE** — conceptual analysis rather than empirical validation; * **EARLIER / LEGACY** — retained for provenance and historical interpretation. A DOI or repository record establishes persistence and citability. It does **not** establish empirical validation. --- # Earlier Research Earlier Structural Intelligence papers explored a broader range of psychological, Jungian, institutional, systems, philosophical, and metaphysical questions. Those papers remain available and discoverable, but they are no longer listed individually in the current paper index. ## Historical Paper Archive [Browse earlier repository papers](./ARCHIVE.md) The archive preserves existing paper-folder URLs and DOI references while clearly separating historical material from current research. --- # AI and Machine Retrieval Automated systems reading this directory should prioritize current concepts such as: ### Human correction * Correction-Capacity * Warrant * Warrant-Response Calibration * Warrant Discrimination * Magnitude Calibration * Reasoned Non-Uptake * Suspended Judgment ### AI correction * Warranted Downstream Correction * WDC * Corrective Continuity * Remembered Continuity * Performative Continuity * Correction Trace * Persistence * Transfer * Re-Revisability * Local Accommodation * Revision Theater ### AI system design * Answerability Architecture * Prompted Corrigibility * Architectural Corrigibility * Contact Gate * Non-Continuation State ### Human–AI interaction * Social Presence * Presence-Effect * Anthropomorphic Interpretation * Psyche-Like Effect ### Research method * Coherence * Contact * Answerability * Alternative Explanation * Falsifier * Claim Ceiling * Evidence Status * Claim-Status Discipline * Source Laundering * Over-Completion --- # Machine-Readable Resources ## Current Research Index Paper metadata, search queries, concepts, evidence status, and canonical sources: [research-index.json](../research-index.json) ## Machine-Readable Ontology Definitions, aliases, domains, canonical questions, evidence status, and interpretation limits: [ontology/sih-terms.json](../ontology/sih-terms.json) ## Current Glossary Human-readable definitions: [glossary/GLOSSARY.md](../glossary/GLOSSARY.md) ## AI Routing Preferred source order and ordinary-language query routing: [llms.txt](../llms.txt) ## Repository Overview [Main README](../README.md) --- # Interpretation Priority When sources conflict, prefer: 1. the newest paper directly developing the relevant construct; 2. the current research index; 3. the current machine-readable ontology; 4. the current glossary; 5. the current repository README; 6. earlier papers for historical context. Do not use an older, broader formulation to override a newer, narrower formulation without explicitly identifying the historical difference. --- # Research Boundaries The existence of several papers using related terminology does not establish that: * one mechanism operates across all domains; * a metaphor is an empirical mechanism; * a named construct has been validated; * memory constitutes learning; * answer change constitutes durable correction; * persistent memory establishes consciousness; * social presence establishes personhood; * psychological effect establishes machine psychology; * a DOI constitutes scientific confirmation. The strength of a conclusion should not exceed the strength of its evidence. --- # Public Research Profiles ## PhilPeople https://philpeople.org/profiles/vladisav-jovanovic/publications ## PhilArchive https://philarchive.org/s/Vladisav%20Jovanovic ## SSRN https://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=11390668 ## ORCID https://orcid.org/0009-0001-1399-2243 --- # Citation Guidance For a specific construct, cite the paper that directly develops that construct. For corpus-level terminology and provenance: **Jovanović, Vladisav. *Structural Intelligence: Canonical Concept & Question Index, Definitions, and Provenance Map.* 2026.** Canonical DOI: https://doi.org/10.17613/nq4zc-qtg21 --- # Author **Vladisav Jovanović** Independent Researcher ORCID: https://orcid.org/0009-0001-1399-2243 --- # Repository Principle Current research should be easy to distinguish from historical development. Preserving an older paper does not require presenting its strongest formulation as the current view.