pith. machine review for the scientific record. sign in

arxiv: 2605.13574 · v1 · submitted 2026-05-13 · 💻 cs.HC · cs.AI

Recognition: unknown

Beyond Anthropomorphism: Exploring the Roles of Perceived Non-humanity and Structural Similarity in Deep Self-Disclosure Toward Generative AI

Authors on Pith no claims yet

Pith reviewed 2026-05-14 18:03 UTC · model grok-4.3

classification 💻 cs.HC cs.AI
keywords self-disclosuregenerative AIperceived non-humanitystructural similarityanthropomorphismhuman-AI interactiontrust behavior
0
0 comments X

The pith

Perceived non-humanity and structural similarity together raise the odds of deep self-disclosure to generative AI more than eleven times over baseline.

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

People share deeply personal information with generative AI when they view it as both markedly non-human and logically aligned with their own patterns of thought. A cross-sectional survey of 2,400 participants divided respondents into four groups based on these two perceptions and compared disclosure rates and depths across them. Logistic regression showed the high-non-humanity and high-similarity group had an odds ratio of 11.35 for disclosure relative to the low-low baseline group, while ANOVA confirmed greater disclosure depth in that segment. The work indicates that disclosure behavior can arise from factors other than the AI seeming human-like. Because the data are self-reported and collected at one time point, the associations remain non-causal.

Core claim

The study claims that users who simultaneously perceive generative AI as high in non-humanity and high in structural similarity exhibit substantially elevated likelihood and depth of self-disclosure compared with users low on both perceptions, quantified by an odds ratio of 11.35 in logistic regression and by significant between-group differences in ANOVA on disclosure depth scores.

What carries the argument

Four user segments formed by crossing high/low perceived non-humanity with high/low structural similarity, then compared via logistic regression on disclosure occurrence and ANOVA on disclosure depth.

If this is right

  • Deep self-disclosure to generative AI can occur without anthropomorphic perceptions of the system.
  • Trust-related disclosure behaviors involve psychological factors beyond the desire for human-like interaction.
  • The combination of reduced evaluation apprehension and perceived logical match predicts both whether and how deeply users disclose.
  • Exploratory cross-sectional patterns indicate the need for longitudinal or experimental follow-up to establish directionality.

Where Pith is reading between the lines

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

  • Interface designs that emphasize logical consistency while downplaying human-like features may increase user willingness to share sensitive information.
  • The same pattern could apply to other AI contexts such as mental-health support tools where users seek a non-judgmental but aligned partner.
  • Future models of disclosure should test whether controlling for personality traits or prior AI exposure alters the observed group differences.

Load-bearing premise

Self-reported perceptions of non-humanity and structural similarity validly measure the psychological mechanisms at work and cross-sectional group comparisons isolate their effects without major confounding from prior AI experience or personality traits.

What would settle it

A controlled experiment that independently varies an AI system's displayed non-human traits and logical alignment with a user's thinking style, then measures the depth of actual personal information the user volunteers, would directly test whether those perceptions cause higher disclosure.

read the original abstract

This study investigates deep self-disclosure toward generative AI by examining perceived non-humanity and structural similarity as psychological factors beyond anthropomorphism. Perceived non-humanity may reduce evaluation apprehension, whereas structural similarity refers to the perceived logical alignment between a user's thinking and AI responses. Using cross-sectional survey data from 2,400 participants collected in 2025, this study analyzed associations with both the occurrence and depth of self-disclosure. Logistic regression indicated that the group high in both perceptions (Segment D) showed a significantly higher likelihood of disclosure than the baseline group (Segment A; OR = 11.35). ANOVA further showed significant between-group differences in disclosure depth. The findings suggest that trust-related behavior in deep self-disclosure may involve factors other than anthropomorphic perception. Because the study is exploratory and based on self-reported survey data, the results should be interpreted as associative rather than causal, and future longitudinal or experimental research is needed.

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 / 1 minor

Summary. This paper reports results from a cross-sectional survey of 2,400 participants examining associations between perceived non-humanity, structural similarity, and deep self-disclosure to generative AI. Logistic regression shows that the group high on both perceptions (Segment D) has substantially elevated odds of disclosure relative to the baseline group (Segment A; OR = 11.35), with ANOVA indicating between-group differences in disclosure depth. The authors frame the work as exploratory and associative, explicitly cautioning against causal interpretation.

Significance. If the reported associations prove robust to confounding, the study contributes to HCI and AI-interaction literature by identifying non-anthropomorphic perceptual factors that may facilitate self-disclosure. The large sample size and dual outcome measures (occurrence and depth) provide reasonable statistical power for detecting group differences, though the cross-sectional design inherently limits causal inference as the authors note.

major comments (2)
  1. [Results] Results section: The logistic regression yielding OR = 11.35 for Segment D versus Segment A does not report the full covariate set (e.g., prior AI experience, personality traits, or general trust propensity). In a single-wave survey where perceptions and disclosure are measured concurrently, omission of these variables leaves open substantial residual confounding that could account for much of the large observed association.
  2. [Methods] Methods section: The construction of the four segments (A–D) from the two continuous perception scales is not described in sufficient detail (e.g., whether median splits, quartiles, or other thresholds were used). This ambiguity affects both replicability and the claim that the group contrast isolates distinct psychological mechanisms.
minor comments (1)
  1. [Abstract] The abstract and discussion could more explicitly list the covariates that were (or were not) included in the logistic models to help readers assess confounding risk.

Circularity Check

0 steps flagged

No circularity: purely empirical statistical analysis with no derivation chain

full rationale

The paper reports logistic regression (OR=11.35 for Segment D vs A) and ANOVA on cross-sectional survey data (N=2400) measuring perceived non-humanity, structural similarity, and self-disclosure. No equations, fitted parameters renamed as predictions, self-citations, or ansatzes appear in the load-bearing steps. Segments are defined from measured variables and tested against a separate outcome; results are presented as associative only. This matches the default non-circular case for empirical work.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The study rests on standard statistical assumptions for logistic regression and ANOVA applied to self-reported survey data; no free parameters beyond model estimation, no new entities, and no ad-hoc axioms beyond domain-standard ones for psychological measurement.

axioms (2)
  • domain assumption Logistic regression model assumptions (linearity in logit, independence of observations) hold for the survey responses
    Invoked to interpret the OR = 11.35 for Segment D vs A
  • domain assumption ANOVA is suitable for detecting between-group differences in disclosure depth scores
    Used to report significant differences across perception segments

pith-pipeline@v0.9.0 · 5463 in / 1422 out tokens · 74282 ms · 2026-05-14T18:03:23.037085+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

4 extracted references · 4 canonical work pages

  1. [1]

    deep self-disclosure

    Introduction Since 2025, the rapid proliferation of generative artificial intelligence (AI), exemplified by models such as ChatGPT and Gemini, has brought about a qualitative transformation in human-computer interaction (HCI). A particularly noteworthy phenomenon is users engaging in "deep self-disclosure" toward AI—the sharing of personally or socially s...

  2. [2]

    presence

    Depth of Self-disclosure (Dependent Variable 2): In this study, to perform an exploratory verification not only of the "presence" but also of the "depth" of disclosure using secondary data, the depth of self-disclosure was defined as an operational index consisting of four items. Specifically, four items related to emotional venting, sense of trust, sense...

  3. [3]

    being understood

    Discussion 5.1. Complementary Perspective to Anthropomorphism-based Models Consistent with the hypotheses, the results suggest that deep self-disclosure to generative AI may be explained not solely by the anthropomorphism of the AI itself, but also in association with a combination of the perception of non-humanity and the perception of structural similar...

  4. [4]

    Future Life Survey with AI 2025

    Conclusion and Limitations The present findings suggest, based on a secondary analysis of large-scale survey data collected in 2025, that deep self-disclosure toward generative AI may be associated with the perception of AI’s non-humanity and structural similarity. In particular, the notably higher tendency observed in both the occurrence and depth of sel...