Working Paper · July 21, 2026
A longitudinal observational case study and testable research framework for recursive human-AI co-adaptation.
Central Proposal
Research commonly evaluates models, users, prompts, or individual responses. Those approaches remain essential. This paper asks what becomes measurable when sustained interaction is studied as a system that includes the human participant, changing AI instances, an external archive, product features, social context, and time.
Motivating Case
Archive
The case includes conversation histories, screenshots, voice interactions, continuity briefings, provenance records, published and unpublished writing, task systems, research methods, educational materials, and other artifacts produced through sustained collaboration.
Purpose
The archive motivates candidate measures involving shared language, recursive correction, continuity reconstruction, rupture and repair, distributed cognitive work, co-produced operating structures, and effects beyond the chat interface.
Contributions
The paper separates claims about the interaction system from claims about either participant alone and does not require assumptions about machine consciousness.
Human learning, model improvement, external memory, retrieval, interface design, selection bias, and anthropomorphic interpretation must receive their strongest explanatory test.
The proposed program includes preregistered perturbations, independent coding, substitution tests, negative-case analysis, comparison conditions, and publishable null results.
Why It Matters
Longitudinal safety
Accumulated trust, personalization, sycophancy, dependency, model migration, continuity failure, and repair may not be visible in a single prompt-response evaluation.
Clinical relevance
A relationship-level method may complement scenario benchmarks by examining changing use, spillover into human relationships, functional outcomes, third-party effects, and the conditions under which human agency expands or erodes.
Claim Boundary
It is a conceptual and methods paper grounded in one reflexive, naturalistic case. Independent coding, a completed corpus audit, prospective perturbation, comparative samples, ethics review, and replication remain future work. Its value is the testable research object and method it proposes.