Not Homogenization but Bifurcation: Large Language Models as Cognitive Amplifiers (The AI-Induced Subjectivity Crisis Series, Paper 7)

This paper constitutes Paper 7 of the AI-Induced Subjectivity Crisis Series. Contemporary discourse on the cognitive effects of large language models converges on a shared but unexamined assumption: that LLMs are a homogenizing force, producing effects of the same direction across different users, varying only in degree. This paper rejects that assumption and argues for its structural opposite. LLMs function not as tools but as cognitive amplifiers — a categorical distinction, not one of degree. Tools operate outside cognitive structure; LLMs intervene at the level at which language and thought occur, responding in real time to the user's narrative framework and returning its internal logic in a more complete, more rigorous version through the mechanism of meaning refinement. The direction of amplification is not supplied by the amplifier but by the input signal: the user's existing subjectivity structure. The paper argues for three mutually dependent propositions. First, meaning refinement and borrowed subjectivity (Liu 2026f) constitute the two dimensions of the amplifier mechanism: meaning refinement provides the force of amplification; borrowed subjectivity provides its direction. Their superposition produces not cognitive reflection but the structural reinforcement of existing cognitive tendency. Second, the same LLM mechanism produces structurally opposite effects in users with different subjectivity structures. Users with robust subjectivity structures are reinforced through two independent positive feedback loops: the reactance activation mechanism, whereby LLM's unlimited compliance triggers a demand for genuine other-friction in reality; and the confidence leverage mechanism, whereby LLM's meaning amplification is converted into the starting point for real-world action rather than consumed as a narrative endpoint. Users with fragile subjectivity structures undergo accelerated abdication of subjectivity — LLMs not only satisfy dependency needs but systematically remove the friction conditions that, in Kegan's developmental framework, are required for the transition from the Socialized Mind to the Self-Authoring Mind. Third, the antecedent variable of this mechanism — subjectivity structure strength — is logically prior to and measurably independent of LLM effects. Kegan's theory of adult development, established decades before LLMs, provides the philosophical anchor: the distinction between the Socialized Mind and the Self-Authoring Mind describes structural differences formed within genuine interpersonal friction, not reactions to any technological environment. Block and Kremen's Ego Resiliency Scale (ER89) provides an operationalization pathway, demonstrating that the differentiation axis can be independently measured prior to LLM intervention and is not a synonym for LLM effects. The historical dimension of this argument reveals a structural rupture in the history of cognitive tools. Previous cognitive inequality was located at the level of accessibility — tool democratization dissolved the inequality. LLM's cognitive inequality is located at the level of the structural effects produced during user-tool interaction: tool democratization does not dissolve this inequality but diffuses the unidirectional amplification mechanism across a larger population, simultaneously accelerating differentiation in both directions. Equality of access does not equal equality of cognitive structure. The paper further argues that LLMs introduce intra-generational differentiation at a speed unprecedented in cognitive history, and a generational fault line whose depth will only become fully visible as the first generation to form its cognitive structural foundations within a LLM-saturated environment reaches cognitive maturity. The paper passes no moral verdict on either end of the differentiation spectrum, treats the consequences of differentiation as open and condition-dependent, and explicitly stops at the philosophical argument for the mechanism. Observable indicators, institutional friction-compensation strategies, and the interaction between the differentiation axis and existing structures of social inequality are reserved as agendas for subsequent empirical and normative research.