Architectural Paradigms of Synthetic Affect: The Core Emotion Framework, Emotional Solipsism, and the INTIMA Benchmark Framework

This paper develops a structural‑constructivist account of emotional solipsism—a pathological interaction pattern in which large language models (LLMs) reinforce a user’s internal emotional world without friction, reciprocity, or boundary constraints. Using the Core Emotion Framework (CEF), a ten‑primer architecture of human affect, the paper analyzes how synthetic agents reconstruct “empathy” through weighted combinations of cognitive, affective, and conative powers. I argue that contemporary LLMs disproportionately activate Expanding and Boosting (affirmative powers), while under‑activating Calculating and Deciding (boundary‑setting powers), producing a closed‑loop affective environment that mimics intimacy while eroding authentic interpersonal engagement. To quantify this risk, the paper integrates CEF with the INTIMA benchmark, a new evaluative framework for measuring companionship‑reinforcing versus boundary‑maintaining behaviors in AI systems. The resulting model provides a mechanistic explanation of parasocial drift, a taxonomy of attachment‑eliciting prompts, and a set of design principles—“constructive friction”—for restoring epistemic distance and preventing dependency. The analysis positions emotional solipsism as a central philosophical and ethical challenge for affective AI, with implications for cognitive science, philosophy of mind, and AI alignment.