pith. sign in

arxiv: 2511.00654 · v1 · submitted 2025-11-01 · 💻 cs.HC

Measuring Machine Companionship: Scale Development and Validation for AI Companions

Jaime Banks This is my paper

Pith reviewed 2026-05-18 01:09 UTC · model grok-4.3

classification 💻 cs.HC
keywords machine companionshipAI companionsscale developmentfactor analysisconstruct validationhuman-AI interactionmeasurement scale
0
0 comments X

The pith

A new scale measures machine companionship with AI companions through Eudaimonic Exchange and Connective Coordination.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper addresses inconsistent ways of studying relationships with companionable AI agents by building and testing a dedicated measurement scale. Machine companionship is defined as an unfolding, self-directed, positive, and coordinated connection between a person and a machine. An item pool covering coordination, positive feelings, and related aspects was reviewed by experts and then answered by 467 users of AI companions. Exploratory factor analysis produced two dimensions that largely aligned with expected patterns when checked against other measures in a second sample of 249 users.

Core claim

The central claim is that a systematically generated and expert-reviewed item pool, when responded to by people using AI companions, yields two factors—Eudaimonic Exchange and Connective Coordination—that function largely as anticipated in construct validation analyses. These factors capture the intended experience of machine companionship, and post-hoc examination of the data points to two distinct templates users may follow: one socioinstrumental and one autotelic.

What carries the argument

The two-factor structure of Eudaimonic Exchange and Connective Coordination induced from the expert-reviewed item pool covering dyadism, coordination, autotelicity, temporality, and positive valence.

If this is right

  • Machine companionship experiences with AI companions can now be measured consistently across studies.
  • The two factors largely perform as predicted when compared to related measures of well-being and connection.
  • Post-hoc results indicate users may follow either a socioinstrumental or an autotelic template for machine companionship.
  • The scale offers a practical tool for assessing ongoing relationships between humans and customizable AI agents.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Designers of AI companions could apply the scale to test whether specific features promote one template more than the other.
  • Longer-term studies might check if higher scores on either factor predict continued engagement or emotional attachment.
  • The approach could be extended to measure companionship in other human-machine settings such as robotics or virtual agents.

Load-bearing premise

The systematically generated and expert-reviewed item pool adequately and without bias represents the full intended construct of machine companionship rather than reflecting only the researchers' initial framing of its components.

What would settle it

A replication with new users of AI companions in which the Eudaimonic Exchange and Connective Coordination factors fail to show the expected correlations with established measures of personal growth or social coordination would indicate the scale does not validly capture the construct.

read the original abstract

The mainstreaming of companionable machines--customizable artificial agents designed to participate in ongoing, idiosyncratic, socioemotional relationships--is met with relative theoretical and empirical disarray, according to recent systematic reviews. In particular, the conceptualization and measurement of machine companionship (MC) is inconsistent or sometimes altogether missing. This study starts to bridge that gap by developing and initially validating a novel measurement to capture MC experiences--the unfolding, autotelic, positively experienced, coordinated connection between human and machine--with AI companions (AICs). After systematic generation and expert review of an item pool (including items pertaining to dyadism, coordination, autotelicity, temporality, and positive valence), N = 467 people interacting with AICs responded to the item pool and to construct validation measures. Through exploratory factor analysis, two factors were induced: Eudaimonic Exchange and Connective Coordination. Construct validation analyses (confirmed in a second sample; N = 249) indicate the factors function largely as expected. Post-hoc analyses of deviations suggest two different templates for MC with AICs: One socioinstrumental and one autotelic.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims to address inconsistent conceptualization and measurement of machine companionship (MC) by developing and initially validating a novel scale for the unfolding, autotelic, positively experienced, coordinated connection between humans and AI companions (AICs). After systematic item-pool generation and expert review covering dyadism, coordination, autotelicity, temporality, and positive valence, exploratory factor analysis on N=467 AIC users induced two factors (Eudaimonic Exchange and Connective Coordination); construct validation largely supported the factors in this sample and was replicated in a second sample (N=249), with post-hoc analyses suggesting socioinstrumental and autotelic templates for MC.

Significance. If the central claim holds, the work supplies an initial empirical instrument for measuring MC experiences, helping to reduce the theoretical and empirical disarray identified in prior systematic reviews of companionable machines. The cross-sample replication of the two-factor structure and the pattern of construct-validation correlations provide a concrete starting point for future studies of socioemotional human-AIC relationships; the post-hoc template distinction, while exploratory, offers a falsifiable hypothesis that could be tested in subsequent work.

major comments (2)
  1. [Item Generation and Expert Review] The description of item-pool generation (systematic generation plus expert review) does not detail how the 'unfolding' aspect of the intended MC construct is operationalized or ensured to be represented independently of the a priori framing elements (dyadism, coordination, autotelicity, temporality, positive valence). Because the EFA factors are induced from this pool, any under-representation of unfolding risks the factors simply reproducing the input structure rather than validating the broader construct; full item wording, generation protocol, and expert-review criteria should be supplied to allow assessment of domain coverage.
  2. [Construct Validation Analyses] The manuscript reports that construct-validation analyses 'function largely as expected' and are 'confirmed in a second sample,' yet the available text provides neither the full correlation matrix nor the exact pattern of hypothesized versus observed relations for the two factors. Without these details it is difficult to evaluate whether deviations are minor or whether they undermine the claim that the factors capture the defined MC experience; a table or appendix listing all validation correlations with effect sizes and significance levels is needed.
minor comments (2)
  1. [Abstract and Results] The abstract and results sections use 'induced' for the EFA outcome; standard terminology is 'extracted' or 'retained' factors. Consistent use of psychometric terminology would improve clarity.
  2. [Post-hoc Analyses] The post-hoc interpretation of two MC templates is presented without a pre-registered analysis plan or correction for multiple comparisons; while exploratory, a brief note on the risk of over-interpretation would be helpful.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback, which helps us improve the clarity and transparency of our scale development work. We address each major comment below and will incorporate the requested details in the revised manuscript.

read point-by-point responses

  1. Referee: [Item Generation and Expert Review] The description of item-pool generation (systematic generation plus expert review) does not detail how the 'unfolding' aspect of the intended MC construct is operationalized or ensured to be represented independently of the a priori framing elements (dyadism, coordination, autotelicity, temporality, and positive valence). Because the EFA factors are induced from this pool, any under-representation of unfolding risks the factors simply reproducing the input structure rather than validating the broader construct; full item wording, generation protocol, and expert-review criteria should be supplied to allow assessment of domain coverage.

    Authors: We agree that greater detail on the item generation process will strengthen the manuscript and allow better evaluation of construct coverage. The unfolding aspect was operationalized through items capturing the temporal progression, evolving quality, and ongoing development of the connection (integrated with but not reducible to the temporality framing element), drawing from literature on dynamic dyadic relationships to distinguish it from static elements like coordination or autotelicity. In the revised version, we will expand the Methods section to provide the full item generation protocol, all original item wordings (in an appendix), and the specific expert-review criteria used to ensure balanced representation across the a priori elements. revision: yes

  2. Referee: [Construct Validation Analyses] The manuscript reports that construct-validation analyses 'function largely as expected' and are 'confirmed in a second sample,' yet the available text provides neither the full correlation matrix nor the exact pattern of hypothesized versus observed relations for the two factors. Without these details it is difficult to evaluate whether deviations are minor or whether they undermine the claim that the factors capture the defined MC experience; a table or appendix listing all validation correlations with effect sizes and significance levels is needed.

    Authors: We acknowledge that the current summary of construct-validation results limits full assessment of the hypothesized versus observed patterns. While the manuscript indicates that the factors functioned largely as expected with replication across samples, we agree that providing the complete data would improve transparency. In the revision, we will add a table (or appendix) presenting the full correlation matrix for both factors with all validation measures, including effect sizes, significance levels, and explicit notes on hypothesized versus observed relations for each sample. revision: yes

Circularity Check

0 steps flagged

Standard empirical scale development with no circular derivation or self-referential reduction

full rationale

This is a conventional scale-development study in human-computer interaction research. The process starts with a conceptual definition of machine companionship, followed by systematic item generation tied to that definition and related dimensions (dyadism, coordination, autotelicity, temporality, positive valence), expert review, data collection from two samples, exploratory factor analysis to induce two factors, and construct validation against external measures. No mathematical equations, derivations, or fitted parameters are present that would make the final scale or factors equivalent to the inputs by construction. Any self-citations (if present in the full text) are not load-bearing for the core claims, and the factor structure emerges from empirical data rather than being presupposed. The outcome is therefore self-contained against external benchmarks and does not reduce to a renaming or tautological fit.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The work rests on standard psychometric domain assumptions rather than new axioms or invented entities; no free parameters are fitted beyond the data-driven factor extraction itself.

axioms (1)
  • domain assumption Machine companionship can be operationalized through self-report items covering dyadism, coordination, autotelicity, temporality, and positive valence.
    This premise directly justifies the initial item pool generation and expert review process described in the abstract.

pith-pipeline@v0.9.0 · 5719 in / 1253 out tokens · 30765 ms · 2026-05-18T01:09:13.617691+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

  1. [1]

    International Journal of Social Robotics1(1), 71–81 (2009)

    https://doi.org/10.1007/s12369-008-0001-3 Bayor, L., Weinert, C., Maier, C., & Weitzel, T. (2025). Social-oriented communication with AI companions: Benefits, costs, and contextual patterns. Business & Information Systems Engineering, 67 , 637-655. https://doi.org/10.1007/s12599-025-00955-1 Bowman, N. D., & Goodboy, A. K. (2020). Evolving considerations a...

    work page doi:10.1007/s12369-008-0001-3 2025

  2. [2]

    https://doi.org/10.1080/10447318.2024.2375797 Seibt, J., Vestergaard, C., & Damholdt, M. F. (2020). Sociomorphing, not anthropomorphizing: Towards a typology of experienced sociality. Proceedings of Robophilosophy 2020. 335, pp. 51-67. IOS Press. Shank, D. B. (2025). Providing companionship. In The machine penalty (pp. 171- 191). Springer. Sherry, A., & H...

    work page doi:10.1080/10447318.2024.2375797 2024