Firms pursuing human‑centric AI experience lower firm‑specific stock volatility in China, though the effect varies: stronger where digitalisation and executive ownership are higher, and weaker where operational efficiency and CEOs with IT backgrounds are greater.

Main Finding

• HCAI adoption is associated with lower firm idiosyncratic risk (IR). Firms with human-centric AI strategies exhibit reduced firm-specific stock volatility, interpreted as investors perceiving lower execution/ethical uncertainty and greater long‑term value.
• Moderators: the IR‑reducing effect of HCAI is strengthened by higher firm digitalisation and greater executive shareholding, but weakened by higher operational efficiency and having a CEO with an IT background.

(Paper: Information Systems Frontiers. DOI: 10.1007/s10796-026-10759-7. Authors: Liu, Wang, Yan, Gupta, Giannakis. Data: Chinese listed firms 2015–2023; accepted 25 May 2026.)

Key Points

• Conceptual framing
  • HCAI is treated as a situated AI strategy (situated AI theory) operationalised through three constructs: human‑centred design, ethical management, and responsible implications.
  • The paper integrates Situated AI Theory (SAIT) with Socio‑Technical Systems (STS) to explain how HCAI reduces investor uncertainty by aligning AI deployments with stakeholder expectations and preserving human oversight.
• Mechanism
  • HCAI signals stronger governance, explainability, and human‑in‑the‑loop design to investors → lowers perceived firm‑level execution and ethical risk → lowers idiosyncratic volatility.
• Moderating factors (socio‑technical)
  • Strengthening moderators: digitalisation (technical) and executive shareholding (social) magnify HCAI’s risk‑reducing effect.
  • Weakening moderators: operational efficiency (technical) and CEO IT background (social) attenuate the effect—authors argue throughput pressures and a technology‑first orientation can dilute human oversight and increase perceived execution risk.
• Robustness and checks
  • Results robust to alternative IR measures, time‑lag specifications, propensity score matching (PSM), and subsample regressions.

Data & Methods

• Sample: Multi‑source panel of Chinese listed firms, 2015–2023; 16,461 firm‑year observations.
• HCAI measurement: Patent text mining—identification of HCAI‑related innovations from ~2.873 million patent texts, following prior methodology (Liu et al., 2025) to proxy firm HCAI strategy at an early/innovation stage.
• Dependent variable: Firm idiosyncratic risk (IR) — operationalised via firm‑level stock volatility net of systematic market factors (standard idiosyncratic volatility approach in finance literature).
• Empirical strategy: Panel econometric models with controls; additional robustness through alternative specifications, lagged models, propensity score matching, and subsample analyses. Moderation tested by interaction terms between HCAI and socio‑technical variables (digitalisation, operational efficiency, executive shareholding, CEO IT background).

Implications for AI Economics

• For capital markets and valuation
  • Investors discriminate between mere AI adoption and the design/governance of AI. HCAI can lower firm‑specific risk, potentially reducing cost of equity and improving firm valuation if signalled credibly.
• For firm strategy and governance
  • Firms should treat HCAI as a strategic signal—not only to employees and regulators but to investors. Embedding human‑centred design and ethical governance into AI development can mitigate financial risks associated with AI deployment.
  • Complementary investments matter: digital transformation and aligned executive incentives (shareholding) enhance the market payoff from HCAI; however, focusing narrowly on throughput/operational efficiency or appointing technically oriented CEOs can blunt investor confidence if they suggest deprioritised human oversight.
• For policy and regulation
  • Policy incentives or disclosure standards that encourage HCAI features (transparency, human‑in‑the‑loop, ethical guardrails) can have real financial stability benefits by lowering firm‑level risk and limiting negative investor reactions to AI missteps.
• For empirical research in AI economics
  • Suggests shifting some analyses from AI exposure (adoption) to AI design/governance (how AI is embedded and governed). Patent/text‑based indicators of AI design are useful early‑stage proxies but should be complemented with operational and disclosure measures in future work.
  • Potential extensions: cross‑country comparisons, causal identification of channels (e.g., event studies around HCAI disclosures), and investigation of systemic risk implications when many firms adopt similar HCAI practices.