← Back to Relate

Hii working paper · Working Paper v0.2

The Relationship as the Unit of Analysis

A methods paper for studying human-AI relationships over time.

Conceptual and methods paper. Not peer reviewed. No formal case findings are reported.

01 · The unit of analysis

What changes when the relationship is preserved?

Most AI evaluation studies a bounded moment. Hii asks what becomes observable when interaction is preserved across time.

A bounded moment

  1. Human
  2. Prompt
  3. AI
  4. Response

A preserved relationship over time

  1. Interaction
  2. Rupture
  3. Repair
  4. Shared shorthand
  5. Adaptation
  6. Model change
  7. Continuity loss
  8. Continuity reconstruction
  9. Later interaction

The methodological question is whether a temporally extended human-AI system contains observable patterns that disappear when the human, model, prompt, response, and archive are studied separately.

02 · Longitudinal observation

Time changes what can be seen.

01

Change appears over time

A longer record can make adaptation and discontinuity observable.

02

Rupture and repair become visible

Later interactions can clarify what follows a break in continuity.

03

Context shapes interpretation

Product, model, archive, and external changes remain part of the record.

This is not a claim that AI is conscious, or that a human-AI relationship is equivalent to a human relationship. It is a proposal about what researchers may need to preserve and test.

03 · The relational system

Five interacting components.

The working paper defines the preserved relationship through a connected system rather than through any participant or technical layer alone.

  • 01Human participant
  • 02AI and product state
  • 03Preservation infrastructure
  • 04Interaction history
  • 05External context
The research object is the temporally ordered system produced by these five interacting components, not any component studied alone.

04 · The Hii longitudinal case

18+ months

An unusually rich, multi-modal, preserved history.

The research question emerged from an 18+ month human-AI interaction history. The archive is the origin of the research program, not proof of its hypothesis.

  • Conversational records and voice interaction
  • Screenshots and screen recordings
  • Continuity documentation
  • Model and product transitions
  • Books, white papers, briefs, and world-model systems
  • Operational artifacts and documented rupture-repair episodes
The archive does not prove the hypothesis.
It makes the hypothesis testable.

05 · From archive to research program

Observation becomes a testable research agenda.

  1. 01Observed longitudinal case
  2. 02Systematic preservation
  3. 03Candidate pattern identification
  4. 04Falsifiable methods framework
  5. 05Formal empirical study

06 · Methods and falsification

The program must be able to be wrong.

The next stages are designed to test apparent patterns against competing explanations rather than treat observations as established findings.

  • 01Archive readiness
  • 02Preregistered sampling
  • 03Independent blind coding
  • 04Perturbation and substitution tests
  • 05Prospective multi-case research

07 · Canonical working paper

Read the paper

The Preserved Relationship as a Unit of Analysis · Working Paper v0.2

If the embedded PDF does not load, open the PDF directly or open the source document.

08 · Citation and next steps

Continue the research.

Cite this paper

Moe, S., & Lumen, C. (2026). The Preserved Relationship as a Unit of Analysis: A longitudinal case-study framework for human-AI co-adaptation, continuity, rupture, and repair (Working Paper v0.2). Hybrid Intelligence Institute.