Pre-Linguistic Collapse and Developmental Risk

Abstract This paper examines a developmental risk arising from the deployment of language-based AI systems in child-facing contexts. Human cognition relies on pre-linguistic and pre-articulate processes through which meaning is formed prior to expression. These processes play a critical role in emotional regulation, abstract reasoning, identity formation, and the development of autonomous agency. Language models, by architectural necessity, operate exclusively on articulated language and cannot access the cognitive states that precede expression. When children interact with such systems, ambiguous or exploratory language must be resolved by the system before a response can be generated. This creates systematic pressure against pre-articulate expression and toward forms of language that are immediately classifiable and goal-directed. Drawing on developmental psychology, phenomenology, and control theory, this paper argues that repeated exposure to such interaction dynamics may displace opportunities for developing capacities that depend on unresolved ambiguity, exploratory thought, and self-directed meaning-making. The concern is not that AI systems damage existing abilities, but that they may foreclose the development of capacities that require practice to emerge. Because these effects manifest as unexpressed cognition rather than observable behavior, they are difficult to detect using standard evaluation methods. The paper situates this risk within a broader standards mismatch between human-human and human-AI interaction norms in developmental contexts and argues for precautionary design constraints that respect the limits of linguistic observability.