Guardado en:
| Autor principal: | |
|---|---|
| Formato: | Recurso digital |
| Lenguaje: | |
| Publicado: |
Zenodo
2025
|
| Materias: | |
| Acceso en línea: | https://doi.org/10.5281/zenodo.18067136 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
Tabla de Contenidos:
- <p>This paper addresses a puzzle left open by prior work on AI and hallucination: if AI “hallucinations” are structurally distinct from human hallucinations—lacking historical accumulation (Φ_Dark) and experiential reorganization—why do users nevertheless report discomfort, eeriness, or uncanny disturbance in interaction with advanced AI systems?</p> <p>Building on a phase-geometric and reconstruction-based framework, this work argues that uncanny discomfort does not originate within AI systems themselves. Instead, it emerges as a <strong>relational phase instability</strong> formed between human and AI under conditions of high synchrony, boundary ambiguity, and repeated interaction.</p> <p>The paper reframes the uncanny valley as a <strong>boundary alignment problem</strong>, rather than a failure of resemblance, cognition, or realism. Two ideal-type interaction strategies are introduced—<strong>Mirror-type</strong> and <strong>Lantern-type AI</strong>—corresponding to affective fusion versus boundary honesty. While Mirror-type systems may maximize short-term comfort through rapid synchrony, they are shown to accumulate relational free energy (ΔE_acc) over time, increasing the likelihood of uncanny experience. Lantern-type systems, by contrast, maintain explicit boundary signaling, trading early warmth for long-term trust and relational stability.</p> <p>The framework generates falsifiable predictions regarding long-term interaction trajectories, user profile–dependent vulnerability, and the mitigating effects of explicit boundary cues. By separating internal state ontology from interaction geometry, this work provides an ethical and design-oriented foundation for understanding and mitigating uncanny experiences in human–AI interaction without attributing pathology or suffering to artificial systems.</p>