Human-centric AI and firm idiosyncratic risks, 2015–2023

Human-centric AI strategies correlate with lower firm-level idiosyncratic risk in a panel of Chinese listed firms from 2015 to 2023, according to an arXiv submission. The paper, arXiv:2606.24224, was submitted on 23 Jun 2026 by Zhen-Yuan Ralph Liu (CUMT) and four coauthors and uses a multi-source panel dataset to test the relationship and its moderators.

What did the authors study and how?

The authors examined whether human-centric AI, conceptualised as a situated AI strategy that reduces AI-related ethical risks and fosters AI-human synergies, relates to firm idiosyncratic risk, using a multi-source panel dataset of Chinese listed firms spanning 2015 to 2023. They integrate situated AI theory with socio-technical systems theory and operationalise moderators including digitalisation, operational efficiency, executive shareholding, and CEOs with IT backgrounds.

The paper frames idiosyncratic risk, IR, as firm-level stock volatility isolated from systematic factors, and treats HCAI as an organizational strategy expected to align AI deployment with diverse stakeholder expectations. The dataset and period are fixed: Chinese listed firms, 2015 to 2023, and the submission appears on arXiv as arXiv:2606.24224.

How does human-centric AI affect firm idiosyncratic risk?

HCAI is associated with lower firm idiosyncratic risk in the authors' empirical tests; the relationship is not uniform, it is moderated by four socio-technical factors. Specifically, the study finds that digitalisation and executive shareholding strengthen the risk-reducing effect, while operational efficiency and CEOs with IT backgrounds attenuate it.

The paper interprets these moderating patterns through socio-technical lenses: greater digitalisation and aligned executive incentives appear to amplify the market's positive response to human-centric AI strategies, whereas higher operational efficiency and technical CEO backgrounds may reduce the marginal risk benefit linked to HCAI. The authors present these findings as empirical evidence connecting ethical, human-centered AI choices with measurable financial outcomes.

Why it matters

Lower idiosyncratic risk tied to HCAI suggests that investors respond to firms that intentionally reduce AI-related ethical risks and promote AI-human collaboration, not merely to technical performance gains. The moderating role of executive shareholding points to governance channels where insiders' stakes matter for how markets price human-centric AI. At the same time, the attenuation by operational efficiency and CEOs with IT experience signals that technical competence or streamlined operations do not automatically substitute for human-centred governance in the eyes of investors.

Those patterns matter for corporate managers and boards deciding where to place scarce attention: ethics and stakeholder alignment can carry measurable risk implications, and governance arrangements such as executive ownership may amplify those effects.

What to watch

Look for the paper's peer-reviewed publication; the arXiv entry notes a comment line linking the work to Information Systems Frontiers, 2026. Observers should also watch whether other samples or markets replicate the same moderator pattern, in particular whether the attenuation by CEOs with IT backgrounds appears outside Chinese listed firms.

Methods, authors and provenance

The paper is authored by Zhen-Yuan Ralph Liu (CUMT), Yu-Ting Wang (NFU), Jia-Jia Yan, Shivam Gupta (NEOMA), and Mihalis Giannakis. It was submitted to arXiv on 23 Jun 2026 as arXiv:2606.24224 and carries a DOI link via arXiv's DataCite entry. The authors position their contribution at the intersection of ethical AI governance and firm financial risk management, and they describe HCAI as a situated strategy that reduces ethical risk while promoting AI-human synergies.

The dataset, empirical approach, and the list of moderators are all taken from the paper's abstract and metadata. Readers interested in the statistical models, robustness checks, and exact variable constructions should consult the full PDF linked on arXiv.