LLM Simulation as Affective Lesion: Decomposing Cognitive-Discursive and Motivational Layers in Group Regulation Through Simulation–Human Triangulation
News
LLM Simulation as Affective Lesion: Decomposing Cognitive-Discursive and Motivational Layers in Group Regulation Through Simulation–Human Triangulation
Abstract
We introduce a demarcation-oriented framework for LLM multi-agent simulation: rather than asking whether LLM agents can replace human participants, we ask what group-level processes they reproduce and what they systematically omit. We report 480 five-agent deliberation sessions in a 2×2×2×2 factorial design crossing perceived environmental uncertainty, group need for cognitive closure, group temporal orientation, and subject-matter involvement (SMI). The simulation selectively reproduced cognitive-discursive dynamics while producing zero affective destabilization across all 480 sessions. SMI showed an inverted effect – accelerating rather than impeding convergence – consistent with the distinction between outcome-relevant and value-relevant involvement. Triangulation with a human survey (N=271, K=50 work groups) confirmed that only the defensive component (SMI_def) predicts destabilization (r=.623, β=.724), while the cognitive component (SMI_rel) does not (r=.133, n.s.). Cross-architecture verification on ChatGPT 5.2 (40 sessions, identical prompts) replicated all three boundary conditions and additionally dissociated convergence from closure, empirically separating the seizing and freezing components of closure urgency. We propose that systematic patterns of reproduction and omission function as an "affective lesion," converting known LLM limitations into diagnostic instruments for decomposing group-level psychological constructs.
Keywords
Edition
Proceedings of the Institute for System Programming, vol. 38, issue 6, part 1, 2026, pp. 139-160
ISSN 2220-6426 (Online), ISSN 2079-8156 (Print).
DOI: 10.15514/ISPRAS-2026-38(6)-9
For citation
Full text of the paper in pdf
Back to the contents of the volume