The Commonplace
Home Dashboard Papers Evidence Digests 🎲
← Papers

AI’s employee dividend is finite: moderate AI adoption raises firms’ employee-focused CSR, but once AI is deeply embedded managerial attention shifts toward systems and employee-oriented CSR falls. Industry automation risk accelerates the decline, while diverse leadership and employee ownership delay and soften it.

Attention to Whom? AI Adoption and Corporate Social Responsibility Toward Human Employees
J. Lyu, Shi-Ye Wang, Chenhao Hu · Fetched March 18, 2026 · Management and Organization Review
semantic_scholar correlational medium evidence 7/10 relevance DOI Source
Using a decade-long panel of Chinese listed firms, the paper finds an inverted U-shaped relationship between AI adoption and employee-related CSR: moderate AI increases managerial attention to employees and ECSR, while deep AI embedding shifts attention toward AI systems and reduces ECSR.

The rapid evolution of artificial intelligence (AI) has profoundly reconfigured the contemporary workplace, redefining the interactions among human employees, AI systems, and organizational processes. Yet, most research adopts a tool-centric view, overlooking how AI’s emergence as an alternative working agent reshapes managerial attention and employee welfare. Drawing on the attention-based view (ABV) and a dual-agent model, we theorize that AI adoption activates two opposing mechanisms: a human attention gain mechanism, where collaboration needs heightened focus on employees and increased employee-related corporate social responsibility (ECSR), and an AI attention shift mechanism, where deep AI embedding redirects attention toward AI, suppressing ECSR. Using panel data from 2575 Chinese listed firms (2013–2023), we find an inverted U-shaped relationship between AI adoption and ECSR. Moreover, industry AI substitution risk sharpens and left-shifts this curve, while top management team (TMT) functional diversity and employee stock ownership flattens and right-shifts it. These findings advance research on AI adoption, managerial attention, and employee-focused CSR by illuminating how attention allocation in dual-agent contexts shapes ethical and strategic outcomes, offering actionable insights for balancing human–AI integration with sustained employee welfare.

Summary

Main Finding

AI adoption has an inverted U‑shaped effect on employee-related corporate social responsibility (ECSR): at low-to-moderate levels, AI adoption increases managerial attention to employees and raises ECSR, but beyond a threshold deeper AI embedding shifts attention toward AI and reduces ECSR.

Key Points

  • The paper frames AI adoption through an attention-based view (ABV) and a dual-agent model that highlights two opposing mechanisms:
    • Human attention gain mechanism: initial AI–human collaboration demands increase managerial focus on employees, boosting ECSR.
    • AI attention shift mechanism: deep embedding of AI redirects managerial attention toward AI systems and away from employees, suppressing ECSR.
  • Empirical finding: a statistically significant inverted U‑shaped relationship between firm AI adoption and ECSR.
  • Moderators:
    • Industry AI substitution risk (i.e., likelihood AI can substitute human tasks) sharpens the inverted U and shifts its peak left — firms in high-substitution-risk industries reach the turning point earlier and experience stronger negative effects at high AI adoption.
    • Top management team (TMT) functional diversity flattens and right‑shifts the curve — diverse expertise in leadership delays and mitigates the negative attention shift.
    • Employee stock ownership (ESOP) similarly flattens and right‑shifts the curve — employee stakeholding aligns incentives and preserves employee-focused attention as AI adoption deepens.
  • Policy and managerial implication: balancing human–AI integration requires organizational structures and incentives that sustain managerial attention to employee welfare as AI penetration increases.

Data & Methods

  • Sample: Panel of 2,575 Chinese listed firms observed from 2013 to 2023.
  • Key variables:
    • AI adoption: firm-level measure of AI technology use/embedding (paper likely uses textual analysis, disclosure metrics, or investment proxies — see original for specific operationalization).
    • ECSR: employee-related CSR outcomes/ratings (likely drawn from CSR reports or third-party ratings).
    • Moderators: industry AI substitution risk, TMT functional diversity, employee stock ownership.
  • Empirical strategy (high level):
    • Panel regressions testing a quadratic (AI + AI^2) specification to detect an inverted U relationship.
    • Interaction terms to test heterogeneity by industry substitution risk, TMT diversity, and ESOPs.
    • Controls for firm-level covariates and robustness checks (e.g., fixed effects, alternative variable measurements, sub-sample analyses).
  • Identification: longitudinal panel design with robustness analyses to support the proposed attention mechanisms (causal claims are grounded in theoretical mechanisms and conditional associations; see paper for further identification strategy).

Implications for AI Economics

  • Firm-level attention allocation matters: AI adoption affects not only productivity but also managerial attention dynamics that shape labor welfare and CSR investments. Economic models of AI adoption should incorporate endogenous attention allocation and multi-agent incentives.
  • Nonlinearities and thresholds: benefits to employees are not monotonic with AI adoption—policy and strategy should anticipate turning points where further automation damages worker welfare.
  • Industry heterogeneity: sectors with high substitution risk face earlier and stronger negative spillovers to employee welfare; industry-level automation risk should be an explicit variable in empirical and theoretical work on AI diffusion.
  • Governance and incentive design matter: TMT diversity and employee ownership can mitigate negative welfare outcomes. Economists and policymakers should evaluate interventions (e.g., ESOPs, board composition rules, training subsidies) that preserve employee interests as AI adoption increases.
  • Measurement and future empirical work: encourage richer measures of AI depth vs breadth, direct measures of managerial attention (communication, time allocation), microdata on employee outcomes, and cross-country comparisons to assess institutional moderation.
  • Broader trade-offs: the findings highlight a trade-off between technological embedding and social welfare within firms; models of automation should include firm-level CSR decisions and potential long-run feedbacks (retention, morale, human capital) on productivity.

If you want, I can (a) extract the paper’s specific operationalizations of AI adoption and ECSR, (b) outline an econometric replication plan, or (c) draft policy recommendations tailored to a particular industry.

Assessment

Paper Typecorrelational Evidence Strengthmedium — The analysis uses a large panel of 2,575 firms over a decade, fixed effects, nonlinear specification, heterogeneity tests, and robustness checks, which provide suggestive and policy-relevant evidence of the inverted‑U pattern; however, absence of an exogenous source of variation (IV, diff‑in‑diff from an external shock, or randomization), possible reverse causality, measurement noise in AI and ECSR proxies, and remaining unobserved time-varying confounders limit strong causal claims. Methods Rigormedium — Good practices are apparent — longitudinal data, fixed effects, formal test for nonlinearity, interaction analyses, and robustness checks — but the operationalization of AI adoption and ECSR is not clearly causal (likely text/investment proxies and third‑party ratings), mechanisms (managerial attention) are inferred rather than directly measured, and there is no clear pre‑trend, placebo, or instrumental analysis reported to more convincingly rule out endogeneity. SamplePanel of 2,575 Chinese listed firms observed annually from 2013 to 2023; key variables include a firm-level AI adoption/embedding measure (likely derived from text/disclosure or investment proxies), employee-related CSR scores/ratings, industry indicators (for AI substitution risk), measures of top management team functional diversity, and indicators of employee stock ownership; controls for standard firm-level covariates and use of fixed effects. Themeshuman_ai_collab org_design IdentificationPanel regression exploiting within-firm variation over 2013–2023 with firm and year fixed effects, a quadratic term (AI and AI^2) to identify an inverted-U relationship, interaction terms for heterogeneity (industry substitution risk, TMT diversity, ESOP), and robustness checks using alternative measures and subsamples; no plausibly exogenous shock, instrumental variable, or natural experiment is reported, so causal inference rests on controlling observed covariates and assuming no important time-varying unobserved confounders. GeneralizabilitySingle-country (China) context — institutional, regulatory, and labor-market conditions may differ from other countries, Listed firms only — likely biased toward larger, better‑resourced firms and not representative of SMEs or informal sector, AI measure may capture disclosure or investment intensity rather than deep operational embedding, limiting applicability to firms with different reporting practices, ECSR measured via reports/ratings may reflect reporting incentives and third‑party methodology biases, Time window (2013–2023) spans rapid AI evolution; findings may not generalize to newest generative AI adoption dynamics, Industry composition matters — results may not transfer to industries absent or underrepresented in the sample

Claims (11)

ClaimDirectionConfidenceOutcomeDetails
AI adoption has an inverted U-shaped effect on employee-related corporate social responsibility (ECSR). Worker Satisfaction mixed high Employee-related corporate social responsibility (ECSR)
n=2575
quadratic panel regression: positive coefficient on AI and negative coefficient on AI^2 (inverted U)
0.3
At low-to-moderate levels of AI adoption, AI increases managerial attention to employees and raises ECSR (human attention gain mechanism). Worker Satisfaction positive medium ECSR (and managerial attention as a theoretical/mediating construct; managerial attention appears inferred rather than directly measured)
n=2575
positive coefficient on AI term (low-to-moderate AI adoption associated with higher ECSR)
0.18
Beyond a certain threshold of AI embedding, deeper AI adoption shifts managerial attention toward AI systems and away from employees, reducing ECSR (AI attention shift mechanism). Worker Satisfaction negative medium ECSR (managerial attention shift inferred)
n=2575
negative coefficient on AI^2 term (ECSR declines at high AI adoption)
0.18
Industry-level AI substitution risk moderates the AI–ECSR relationship: higher substitution risk sharpens the inverted U and shifts its peak left (firms in high-substitution-risk industries reach the turning point earlier and suffer stronger negative effects at high AI adoption). Worker Satisfaction negative medium ECSR
n=2575
significant interactions between AI/AI^2 and industry-level substitution risk indicating earlier and sharper decline in high-risk industries
0.18
Top management team (TMT) functional diversity moderates the AI–ECSR curve by flattening it and right-shifting the peak, delaying and mitigating negative attention shifts from employees to AI. Worker Satisfaction positive medium ECSR
n=2575
interaction effects: TMT functional diversity flattens and right-shifts the inverted-U (mitigates negative effects at high AI adoption)
0.18
Employee stock ownership (ESOP) moderates the relationship by flattening and right-shifting the inverted U, aligning employee incentives and preserving employee-focused attention as AI adoption deepens. Worker Satisfaction positive medium ECSR
n=2575
interaction effects: ESOP presence/level flattens and right-shifts the inverted-U (mitigates negative effects at high AI adoption)
0.18
Theoretical framing: an attention-based view (ABV) and a dual-agent model capture two opposing mechanisms—(1) human attention gain from initial AI–human collaboration and (2) AI attention shift under deep embedding—that jointly generate the inverted U-shaped AI–ECSR relationship. Organizational Efficiency mixed medium Managerial attention (theoretical/mediating construct)
0.18
Data/sample claim: the empirical analysis uses a panel of 2,575 Chinese listed firms observed from 2013 to 2023. Other null_result high N/A (sample description)
n=2575
0.3
Empirical strategy: the main identification strategy uses panel regressions with quadratic AI specification and interaction terms, controlling for firm covariates, employing fixed effects and robustness checks (alternative measures, sub-samples). Other null_result high N/A (methodological claim)
0.3
Policy/managerial implication: organizational structures and incentives (e.g., TMT diversity, ESOPs) are effective levers to sustain managerial attention to employee welfare and mitigate the negative effects of deep AI penetration on ECSR. Organizational Efficiency positive medium ECSR (and managerial attention as targeted by interventions)
0.18
Broader implication for AI economics: firm-level attention allocation, nonlinearities, thresholds, and governance/incentive design should be incorporated into economic models of AI adoption because AI's effects on workers and CSR are not monotonic and depend on industry and governance. Governance And Regulation null_result speculative N/A (theoretical/modeling implication)
0.03

Notes