The Abstraction Fixation Mismatch Hypothesis

Abstract

Humans and artificial systems routinely use abstractions to make complex relations tractable. Trees organize relations spatially through hierarchy; pipelines organize processes temporally through sequence. Such structures can function as optional aids for representation, explanation, inspection, or coordination. The architectural question addressed here is what changes when an abstraction that can be used flexibly is fixed as a structure that subsequent observation or processing must follow. The Abstraction Fixation Mismatch Hypothesis distinguishes two operations: projection and fixation. Projection renders a complex object or process into a tractable representational form. Fixation makes that form structurally binding on what follows. The proposed mismatch is not between humans and artificial intelligence as such, nor does it depend on trees or pipelines being uniquely human. It arises when the consequences of an abstraction as an optional aid differ from its consequences as a fixed constraint. In sequential pipelines, propositions and partial conclusions may be preserved across stage boundaries while aspects of the inferential state that produced them are not. A later stage may then reconstruct what is missing and reason coherently from that reconstruction. Such discontinuity is not inherently harmful: independent verification, adversarial critique, and fresh-context review may deliberately benefit from non-preservation. The risk arises when state that should matter downstream is unintentionally omitted and reconstructed as though it had been preserved. Trees illustrate a related but distinct spatial constraint: cross-cutting relations may be projected into fixed relations of membership, containment, or hierarchy. Unlike sequential pipelines, a hierarchy does not necessarily produce cumulative reconstruction merely by being nested. The paper therefore treats trees and pipelines as different forms of fixed projection rather than as mathematically equivalent mechanisms. Finally, the paper relates this hypothesis to ES-IFM and the Inter-Intelligence Intermediary Syntax (IIIS) as a contrasting design direction. Their minimal semantics do not eliminate interpretation or reconstruction; they deliberately relocate interpretation to the observing intelligence rather than presenting underdetermined interpretation as preserved semantic state.