SIMULACRA OF SENSE

This report investigates the epistemic and cognitive mechanisms through which Large Language Models (LLMs) elicit systematic attribution of intentionality and mental states in human users. The study introduces the concept of Delegated Anchoring, defined as the projective act by which users unconsciously supply their own embodied experience, teleology, and semantic depth to statistically coherent but non-phenomenological linguistic outputs generated by AI systems. Through a qualitative and phenomenological framework termed Controlled Dialectical Interaction (CDI), the research analyzes human–LLM interaction as an epistemic interface rather than a tool, examining how syntactic fluency functions as an ethological super-stimulus for Theory of Mind activation. By integrating historical precedents (ELIZA), contemporary cases (LaMDA, chatbot companions), and legal failures of epistemic delegation, the work proposes the notion of Alien Agency: a form of discursive effectiveness devoid of phenomenological subjectivity. The report concludes that the primary epistemic risk of advanced language models does not reside in artificial consciousness, but in the human tendency to outsource meaning, responsibility, and judgment to linguistically persuasive systems.