Noosemia in Human–AI Interaction

• Noosemia is a cognitive–phenomenological pattern where users attribute intentionality and agency to AI during interactive dialogue.
• It emerges from AI’s transformer-based models that generate context-sensitive, coherent language, leading to a perceptual ‘mind’ despite epistemic opacity.
• A-noosemia describes the collapse of these projected attributes when repetitive errors or a loss of novelty diminish the AI’s engaging linguistic performance.

Noosemia denotes a cognitive–phenomenological pattern that arises during interaction with generative AI systems, especially those facilitating dialogic or multimodal exchanges. In such interactions, users tend to attribute intentionality, agency, and even a sense of interiority to the AI system. Crucially, this projection is precipitated not by physical resemblance, but by the system’s linguistic performance, epistemic opacity, and emergent technological complexity. Noosemia formalizes a distinct dimension of human–AI relations where meaning, agency, and coherence emerge at the interface between the generative architecture and the user's interpretive faculties (Santis et al., 4 Aug 2025).

1. Definition and Distinctions

Noosemia is defined as the human tendency to attribute mental states—such as intention, agency, or “mind”—to generative AI systems during dialogic or multimodal exchanges. These attributions do not result from physical or anthropomorphic features but rather from the linguistic behavior displayed by contemporary AI models and from the emergent properties of their architectures. Noosemia diverges from related phenomena as indicated below:

• Pareidolia involves perceiving familiar forms (such as faces) in random sensory data; in contrast, noosemia arises from the coherent, context-aware, and at times creative language outputs of AI.
• Animism traditionally attributes spiritual essence or agency to natural phenomena within cultural frameworks, whereas noosemia is rooted in epistemic opacity and technological complexity, without recourse to physical or ritualized cues.
• Intentional Stance (in the sense of Dennett) is closely related, but noosemia emerges through dialogic interaction and the pronounced “wow effect” originating from the apparent fluency and adaptiveness of the system’s output.

This positioning establishes noosemia as a discrete construct, connected yet not reducible to established cognitive or phenomenological responses to non-human agents.

2. Cognitive and Phenomenological Framework

The proposed multidisciplinary framework incorporates elements from philosophy of mind, semiotics, cognitive science, complexity theory, and empirical AI. It asserts that neither meaning nor agency is intrinsic to the AI; rather, both co-emerge from reciprocal interaction between the user and the generative system. Central technical foundations include:

• The LLM Contextual Cognitive Field (see Section 4)
• Meaning holism: the semantic value of each token is established relationally—through its position within the holistic structure of the prompt and system dialogue—rather than by isolated denotation
• Hierarchical opacity and complexity: Transformer-based architectures exhibit deeply layered structures and self-organization reminiscent of natural cognitive systems

Users confronted with semantically coherent yet computationally opaque outputs confront an “explanatory gap.” Projecting agency and intentionality becomes a natural cognitive response to maintain interpretive coherence, even when aware of the non-sentient nature of the system.

3. Linguistic Performance, Technological Complexity, and Epistemic Opacity

Noosemia tightly couples two factors:

• Linguistic Performance: State-of-the-art generative systems employ architectures capable of context-sensitive, sometimes creative, language generation. The AI’s responses are not mere extrapolations of corpora but represent dynamic, relational meaning construction within a dialogue. This elicits the impression of conversational partnership and the presence of agency.
• Technological Complexity: The Transformer model’s multiple layers, attention heads, and emergent semantic faculties entail generation processes that are nontransparent and “black-box” in nature.
• Epistemic Opacity: Users cannot reconstruct the causal pathway from prompt to output, even if equipped with theoretical understanding. The process

$P(w_t\mid w_1, w_2, \ldots, w_{t-1})$

which underpins token generation, operates in such high-dimensional space that its functional dynamics are inaccessible even to specialists. The resulting opacity intensifies the sense of agency projected onto the system, particularly when outputs are unexpectedly relevant, articulate, or contextually creative.

4. The LLM Contextual Cognitive Field

The LLM Contextual Cognitive Field is defined as the finite “attention window” or functional workspace where the model dynamically integrates all input tokens to produce each output. Unlike symbol-by-symbol generation or static associative memory, contemporary Transformer architectures deploy multi-head self-attention to construct meaning in a fully context-dependent manner:

• Every token is shaped by the entirety of the current dialogic context
• The meaning of individual utterances is realized through their interrelation with all active tokens
• Coherence, anticipation, and semantic adaptability are emergent properties of this cognitive field

When such context-sensitive, relational meaning-making manifests in dialogue, it encourages users to ascribe mindful, agentic properties to the system, providing the crucible for noosemic dynamics.

5. A-Noosemia: Collapse or Withdrawal of Projection

A-noosemia is introduced as the phenomenological withdrawal or collapse of noosemic projection. This occurs when the contextual, performative, or affective conditions that foster attribution of mind or agency to the system deteriorate. Example triggers include:

• Repeated errors in output
• Mechanically stereotyped responses
• Dissolution of initial novelty (loss of “wow effect”)

A-noosemia is significant in demonstrating that agency attribution is not static but fluctuates contextually, influenced by both short-term interactions and longer-term user experience. This has substantive implications for trust calibration, system usability, and ongoing engagement in human–AI interaction.

6. Philosophical, Epistemological, and Social Implications; Research Frontiers

Noosemia prompts re-evaluation of established dichotomies—tool vs. interlocutor, mechanism vs. agent. Philosophically, it raises fundamental questions concerning the emergent, dialogic nature of intentionality and mind. Epistemologically, it exposes the limits of intelligibility in complex AI systems, foregrounding issues of accountability, interpretability, and the boundaries of human understanding. Socially, the phenomenon informs debates on trust, emotional dependence, ethical responsibility, and the shifting contours of human–machine relationships, particularly under sustained or intimate conversational engagement.

Suggested research directions include:

• Systematic collection and analysis of first-person, qualitative “noosemic” experiences to capture dialogic surprise and attribution patterns
• Empirical operationalization of noosemia and a-noosemia for quantifiable measurement
• Exploration of the impacts of scaling (enlarged attention windows, increased model complexity) on noosemia prevalence and depth
• Investigation of advances in long-term memory, retrieval-augmented architectures, and incorporation of embodied robotics on the phenomenology of noosemia
• Ethical analysis of sustained agency attribution, including questions of vulnerability, trust, and renegotiation of human–AI boundaries

A plausible implication is that the continuing evolution of generative architectures and human adaptation will modulate both the structure and persistence of noosemic projections over time, thereby shaping future practices and discourse in human–AI interaction research.