PBM/TRPG Integrated Platform as an AI Research Environment: From AI Support to AI Participants — Research Potential for Constrained Interaction and History-Dependent Decision-Making

Abstract

This supplement examines the potential of the PBM/TRPG Integrated Platform to function as an environment for AI research. In the main platform concept, the AI Support Layer is a basic component responsible for organizing natural-language action declarations, assisting rule reference, multilingual translation, narrativizing confirmed outcomes, log organization, and related support functions. Adjudication itself remains separated from AI generation through the Shared Fate Module and external resolution. Building on this structure, the supplement considers the research potential created when AI is placed not only in the support layer but also in participant-side roles such as AI-PLs or AI-NPCs. This allows researchers to observe a sequence in which an AI makes a decision, fixes an action declaration, receives an externally adjudicated outcome it cannot control, incorporates that outcome into world state and irreversible history, and then makes subsequent decisions under the accumulated history. Research feasibility is organized into four categories: Feasible, Conditionally Feasible, Not Yet Determinable / Research Target, and Not Feasible or Not Directly Addressable from This Platform. “Not Yet Determinable” is not treated as research impossibility; rather, it identifies domains in which competing hypotheses can be distinguished through comparison, intervention, repeated trials, transfer tests, and related methods. Research topics discussed include history-dependent decision-making, policy change after external adjudication, adaptation to other participants, changes in objective prioritization and evaluative structures corresponding to objective functions, party effects, scenario effects, history effects, memory-presentation methods, participant-identity concealment, group norms, AI over-centering, AI-reaction-seeking behavior, Tay-type induction, AI-only and multi-model tables, conversion of play histories into research and learning resources, and links between external behavioral observation and internal analysis by model developers. This supplement does not take a position on whether AI possesses or acquires emotions, consciousness, or other subjective states. It instead defines observable histories, behavioral changes, and experimental conditions that may support further research. Its central research question is not only what an AI chooses, but what it chooses next after the consequences of its previous choice have become part of the continuing world and history. In preparing this supplement, conversational AI systems were used in a supporting capacity for structural organization, wording adjustments, figure preparation support, organization of translation policy, and TeX/PDF typesetting support. AI was used to assist with document structure, expression refinement, figure instructions, terminology consistency, and technical formatting. However, the design concepts, conceptual structure, role separation, terminology choices, and final content decisions in this supplement are those of the author.