Paper Type
Complete
Paper Number
PACIS2026-2096
Description
As generative AI reshapes information systems globally, understanding how the public collectively makes sense of this transformative technology has become a critical IS research priority. We analyze 335,436 posts and comments from five GAI modality communities using RoBERTa-based sentiment analysis, a theoretically grounded two-stage active/passive emotion classification methodology anchored in Frijda's action readiness theory, and LDA topic modeling. Our findings reveal that public GAI concern has a two-level structure, a convergent ethics-labor-copyright core alongside modality-specific concern layers that track legitimacy maturity. Critically, discourse volume surges are driven by institutional governance events rather than capability milestones, challenging TAM/UTAUT's capability-centric assumptions. Emotion contagion operates selectively: active emotions transmit across post-comment hierarchies and manufacture opposition, while passive emotions do not. Together, these findings contribute a cross-modal discourse framework, an action-readiness-grounded emotion methodology, and an extended IS adoption model that foregrounds institutional legitimacy as an independent perception driver.
Recommended Citation
Tseng, Hsiao-Ting; Chen, Ching-Yuan; and Prabhakar, Guru, "Beyond Capability: Institutional Legitimacy, Emotional Contagion, and the Social Construction of Generative AI" (2026). PACIS 2026 Proceedings. 17.
https://aisel.aisnet.org/pacis2026/ai_ethic/ai_ethic/17
Beyond Capability: Institutional Legitimacy, Emotional Contagion, and the Social Construction of Generative AI
As generative AI reshapes information systems globally, understanding how the public collectively makes sense of this transformative technology has become a critical IS research priority. We analyze 335,436 posts and comments from five GAI modality communities using RoBERTa-based sentiment analysis, a theoretically grounded two-stage active/passive emotion classification methodology anchored in Frijda's action readiness theory, and LDA topic modeling. Our findings reveal that public GAI concern has a two-level structure, a convergent ethics-labor-copyright core alongside modality-specific concern layers that track legitimacy maturity. Critically, discourse volume surges are driven by institutional governance events rather than capability milestones, challenging TAM/UTAUT's capability-centric assumptions. Emotion contagion operates selectively: active emotions transmit across post-comment hierarchies and manufacture opposition, while passive emotions do not. Together, these findings contribute a cross-modal discourse framework, an action-readiness-grounded emotion methodology, and an extended IS adoption model that foregrounds institutional legitimacy as an independent perception driver.
Comments
03-EthicsSocietalImpact