Employees who perceive their employer is adopting AI report greater turnover intentions driven by identity threat; secure attachment dampens this effect while anxious attachment intensifies it.

Main Finding

Organizational AI adoption increases employees’ turnover intentions through elevating identity threat. This mediation is conditional on employees’ organizational attachment style: secure attachment weakens (buffers) the AI → identity-threat link and its indirect effect on turnover intention, while anxious attachment strengthens (amplifies) them. (The provided extract does not report a clear result for avoidant attachment.)

Key Points

• The paper integrates social identity threat theory and attachment theory to explain behavioral responses to organizational AI adoption.
• Hypothesized causal chain: Organizational AI adoption → Identity threat → Turnover intention.
• Moderators: organizational attachment styles (secure, anxious, avoidant). Secure attachment was hypothesized to attenuate identity threat; anxious and avoidant attachment were hypothesized to amplify it.
• Empirical results (from the abstract and main text): identity threat mediates the relationship between organizational AI adoption and turnover intentions. Secure attachment negatively moderates the AI → identity threat link and reduces the mediated effect on turnover. Anxious attachment positively moderates the AI → identity threat link and increases the mediated effect on turnover.
• The study highlights an identity-based “dark side” of AI adoption that complements prior work focusing on stress, job insecurity, and resource pathways.

Data & Methods

• Sample: Three-wave survey of employees (N = 312 final valid responses) from eight high-end manufacturing firms in southern China that had adopted AI devices.
• Timing: Three survey waves with 1-month intervals. T1: organizational AI adoption, attachment style, demographics; T2: identity threat; T3: turnover intention.
• Measures:
  • Organizational AI adoption: 3-item scale (Wang et al., 2016), Cronbach’s α = .91.
  • Organizational attachment styles: 18-item scale (Scrima, 2015) with three subscales — secure (α = .94), anxious (α = .83), avoidant (α = .90).
  • Identity threat: 4-item scale (Yogeeswaran et al., 2016), adapted to context.
  • Turnover intention: measured at T3 (scale details not included in the excerpt).
• Analysis approach: longitudinal survey design with mediation and moderated-mediation testing (identity threat as mediator; attachment styles as moderators). The paper reports indirect effects and interaction tests; exact estimation procedures/statistics are not provided in the excerpt.
• Limitations (apparent from methods): single-country, sector-limited sample (China, manufacturing); self-reported measures; potential attrition bias (312/401 initial, 69.3% valid rate); causality strengthened by temporal separation but still observational.

Implications for AI Economics

• Labor churn and adjustment costs: Organizational AI adoption can raise turnover intentions via identity pathways, implying higher separation rates and associated hiring/training costs—an important margin when firms weigh AI adoption benefits against labor adjustment frictions.
• Human capital and skill complementarities: Identity-related responses add a non-pecuniary channel through which AI changes labor demand. Even where tasks are complementary, perceived threats to identity may increase exits, shifting the realized net gains from AI adoption.
• Heterogeneous firm-level outcomes: Attachment-style heterogeneity implies firms with more securely attached workforces can adopt AI with fewer identity-related separations. This suggests complementarities between organizational practices (culture, communications, employee attachment) and technological adoption affect firm productivity and diffusion patterns.
• Wage and compensation dynamics: If AI adoption induces voluntary turnover among certain employee types, market wages for remaining/competing skills may adjust. Firms may need to internalize retention or retraining costs, altering the private return calculus of AI investment.
• Modeling recommendations for AI-economics research:
  • Include identity/psychological utility terms (not only task-routine substitution/complementarity) when modeling labor supply responses to automation.
  • Allow heterogeneity in worker attachment/organizational identification to capture differential exit risks and adoption externalities.
  • Link firm adoption decisions to labor market frictions (separation costs, vacancy rates) and consider the dynamic interplay between adoption, turnover, and productivity.
• Policy and managerial interventions:
  • To reduce turnover costs induced by AI, firms should invest in attachment-building practices (transparent communication, participation in adoption decisions, training framed as capability enhancement) and target interventions to anxiously attached employees who are more reactive.
  • Public policy aimed at AI-driven transitions should consider supports for identity-preserving retraining programs and incentives for firms to adopt complementary organizational practices that reduce destructive turnover.
• Research gaps for economic quantification:
  • Estimate monetary magnitudes: translate identity-driven turnover into firm-level costs, industry-level labor reallocation, and aggregate productivity effects.
  • Generalizability tests across sectors, countries, and different AI technologies (frontline service robots vs. back-office automation).
  • Causal identification: exploit natural experiments or exogenous adoption shocks to quantify identity-mediated labor responses.

If you’d like, I can (a) extract a short list of concrete managerial actions derived from the findings, (b) sketch a simple economic model that incorporates identity-threat as a utility cost of automation, or (c) propose empirical strategies to quantify the aggregate economic impact of the identity-mediated turnover channel.