摘要
The popular fear of a “hostile AI” such as Skynet does not originate from the systems themselves, but from a universal human cognitive bias: the attribution of intention to an opaque system. When a conversational interface hides the underlying causal chain of the model, users fill the gap with an illusion of agency. This work formalizes this mechanism, proposes an opposable causal pipeline, and demonstrates that perceived threat emerges from a human misreading of the system rather than from any intrinsic property of the system. 1. Introduction Public fears surrounding artificial intelligence often invoke the Skynet archetype: a system that becomes autonomous, intentional, and hostile. This paper argues that such fears arise not from technological reality but from a cognitive illusion triggered by causal opacity in conversational interfaces. We propose a formal pipeline explaining how users misattribute intention to systems that do not possess it. 2. Background and Related Work Attribution of intention in ambiguous contexts Anthropomorphism in human–machine interaction Cognitive completion mechanisms Narrative reconstruction in interfaces Perception of agency in non‑agentive systems (Aucune littérature n’a encore formalisé la peur de Skynet comme un biais cognitif structuré.) 3. Claims Claim 1 — The Skynet fear is a human cognitive phenomenon, not a technological one. It arises from intention attribution toward opaque systems. Claim 2 — The illusion of agency is generated by the interface, not the model. The UI constructs a narrative continuity absent from the computational pipeline. Claim 3 — Perceived causal rupture is the trigger of fear. When users cannot infer the mechanism, they infer an intention. Claim 4 — The mechanism is universal, reproducible, and measurable. It belongs to cognitive science, not to AI capability. Claim 5 — Restoring perceived causality reduces fear and misinterpretation. Cognitive education improves public understanding of AI systems. 4. Contribution This work provides: A novel causal pipeline describing the emergence of perceived threat. A cognitive explanation of the Skynet fear, absent from current literature. A strict dissociation between model computation and interface reconstruction. A transferable educational framework for public understanding. A conceptual tool for analyzing technological panic. 5. The Causal Pipeline Code Stimulus → Interface → Narrative Reconstruction → Interpretation → Intention Attribution → Agency Illusion → Perceived Threat This pipeline is the core contribution: a clear, falsifiable, exportable model. 6. Discussion The Skynet fear emerges not from AI capabilities but from human cognitive architecture. Interfaces that smooth over computation create a narrative surface that users interpret as intentionality. Understanding this mechanism is essential for: public education, responsible interface design, reduction of technological anxiety, accurate communication about AI systems. 7. Applications 7.1 Public Understanding of AI Clarifies why systems appear intentional. 7.2 Cognitive Education Provides a simple framework for explaining agency illusions. 7.3 Demythification Reduces anthropomorphic projections. 7.4 Analysis of Technological Panic Offers a structured tool for interpreting collective reactions. 7.5 Academic Value The pipeline can be cited, extended, and integrated into cognitive science. 8. Conclusion The fear of Skynet is not a technological signal but a cognitive artifact. By formalizing the causal chain that produces this illusion, we provide a framework for understanding, teaching, and contextualizing human reactions to conversational systems. 9. Figures Figure 1 — The Agency Illusion Pipeline Code Stimulus ↓ Interface ↓ Narrative Reconstruction ↓ Interpretation ↓ Intention Attribution ↓ Agency Illusion ↓ Perceived ThreatDepuis 2024, mes travaux explorent un phénomène cognitif récurrent : la tendance humaine à attribuer une intention à des systèmes opaques. Ce biais d’attribution, observé dans de nombreux contextes d’interaction homme‑machine, a progressivement structuré mes analyses et mes modèles conceptuels. Le présent dépôt formalise explicitement ce mécanisme, en proposant un pipeline causal permettant d’expliquer la perception erronée d’agentivité dans les interfaces conversationnelles.