Research preprint · English · v0.2 · 6 September 2026
Shared Open Worlds for Human–AI Dyads
Role Switching, Experience Qualification, and Cross-World Continuity Evaluation
Enjoyment for the human, useful learning for the system: a research proposal for cooperating, competing by agreement, and returning to cooperation across changing game worlds.
Publication
Abstract
This conceptual and methodological working paper examines a human and a long-lived candidate AI line cooperating, competing by agreement, and returning to joint activity across open game worlds. It connects the c=a+b / Temporal AI Presence corpus with social roles, experience provenance, and currently applicable permissions. Evaluation must address shared-history utility, behavioral change, and human enjoyment rather than permission handling alone. An authorial record dated March 3, 2026 documents the earlier intention; its reported mod is not treated as verified runtime evidence. Prior work already covers cross-match memory, co-learning, and sustained computational relationships. Accordingly, C designates a candidate implementation of requirements, not a distinct experimental class. Version 0.2 withdraws the presumption of an independent C/B1 mechanism; no such causal comparison is planned before an actual treatment is specified. Separate contrasts address history availability, transition risk, state organization, and skill transfer. Conventional controls retain ordinary memory and safeguards; episodic-history ablation does not remove current permissions. Five author-constructed illustrations expose rules, events, and admissible responses without constituting an independently annotated benchmark. The staged proposal proceeds from case determinacy to technical and gameplay feasibility, followed by a longitudinal study. No new experiment, identity-continuity result, consciousness claim, new AI class, or economic superiority is reported.
Author: Ivan Kotov, Independent Researcher, Brussels, Belgium · ORCID 0009-0009-6002-9845.
Version DOI: 10.5281/zenodo.22542470
Concept DOI: 10.5281/zenodo.22542469
License: CC BY 4.0, within the published package's stated scope.
Copyright: Copyright © 2026 Ivan Kotov.
Research focus
History, transitions and human value are different questions
The paper separates four contrasts: access to permitted episodic history, transition risk relative to matched controls, an explicitly specified state-organization difference, and transfer of a permitted skill to a new task. Human enjoyment and the burden of corrections are assessed separately. Correct permission handling is not by itself learning or enjoyable co-play.
The candidate uses existing memory and governance primitives. A fully equipped conventional baseline retains the same ordinary safeguards; it is not weakened to manufacture a c-specific result. If no distinct treatment remains, the shared mechanism is adopted as conventional engineering while the history and gameplay questions remain open.
Five author-constructed illustrations make the rules inspectable: retaining a current grant across a world rollback, retaining a revocation after loading an old save, ending competition without erasing useful memory, limiting skill transfer, and withholding an action when a material basis is unknown. They are explanatory examples, not experimental results or independent gold labels.
Evidence and publication status
Conceptual and methodological preprint. No empirical experiment, runtime implementation, independently annotated benchmark, independently demonstrated c-specific effect, identity, consciousness or economic superiority is reported. English v0.2 has not undergone separate independent peer review. A separate C/B1 comparison is not planned before an actual treatment is specified.
This English edition follows the author-accepted Russian v0.2 and the source-based adjudication of three model reviews of v0.1. The earlier GTA V mod report is retained only as documentary history of the idea, not verified runtime evidence. Prior work on game companions, coordination, memory and relational agents is discussed in the paper; general priority is not claimed.
Downloads and citation
Exact published files
PDF · Markdown · DOCX · Public ZIP
DOI-bound publication metadata · Citation with the published DOI · Original package checksums · Archive checksum
Public ZIP SHA-256: 47945ca619f65ea6788de84c22ee278d0cee29fba929ab0e7e1db5fe17cc92ce
The paper, archive and all nine original members are unchanged. New DOI metadata and this navigation page are additive records outside the deposited package. No internal reviews, private Drive locators or personal memories are published here.
Kotov, I. (2026). Shared Open Worlds for Human–AI Dyads: Role Switching, Experience Qualification, and Cross-World Continuity Evaluation (Version 0.2). Zenodo. https://doi.org/10.5281/zenodo.22542470
Within the corpus
Roles, provenance and current permissions
Social roles and memory custody delimit the use of shared history. Historical-versus-operative state separates a remembered permission from one that applies now. Experience qualification and raw-locality rules distinguish learning from permission to transmit or train on a record.
A simulator save may restore a virtual crane position. It cannot restore an operator's revoked authorization. Correct interlocks, a learned skill and a useful human experience therefore remain separate evaluation targets.
Related citations: AI Social Roles and Memory Custody; World 8 / Z0-A (external prior art, not coauthorship or endorsement).
Temporal AI Presence · Corpus map · AGI repository · All publications