Divergent Transition Pathways for Artificial Intelligence: A Longitudinal and Multi-Level Perspective Using Structural Topic Modeling

Abstract

The potential of artificial intelligence (AI) to constitute a general-purpose technology with diverse algorithmic specifications makes it challenging to assess its overall impact on existing socio-economic regimes. Leveraging the multi-level perspective, we seek to depict the trajectory of micro-, meso-, and macro-level forces and their interactions to characterize AI transition pathways in industry. We treat business and information systems literature as a proxy capturing business practices that relate to factors influencing AI transitions on all three different levels. Based on 10,036 publications over 25 years, we map the topic landscape of AI-related research, longitudinal patterns of topics, and structural changes of topic networks. The results indicate a strong and myopic focus on technological capabilities and efficiency rationales. Topic network structures indicate that transition pathways may diverge between a symbiotic and stabilizing transformation process and a more radical pathway of regime substitution. Based on these findings, we argue that sociotechnical transition pathways may not only occur in sequence, but simultaneously and ambiguously. This highlights the need for a nuanced understanding of convergent and divergent transition pathways for emerging digital general-purpose technology that do not tend to settle on one dominant design. We propose to leverage paradox theory to reconcile these tensions. JEL: M000, O310, O320, 033