Cross-boundary AI innovation as recombinant search in heterogeneous landscapes

Abstract

AI has rapidly penetrated various industries, hailed as a universal problem-solving tool. Scholars have studied AI innovation across the contexts of their development and implementation. As general-purpose technology, however, AI innovations need to first jump across its disciplinary boundaries before they can subsequently become useful as applications. To unpack how such jumps are made, we conceptualize cross-boundary AI innovation as an outcome of recombinant search in heterogeneous innovation landscapes that are, in turn, comprised of a set of interconnected epistemic objects. We take a dynamic network view as an analytical perspective and identify two structural attributes: structural embeddedness and junctional embeddedness, which represent its popularity and role as a bridge, respectively. To assess their impact on the likelihood of a jump by an epistemic object, we test our theory using a data set of AI-related journal and conference articles from both Computer Science and Autonomous Vehicle fields in the period between 2009 and 2020. Our results show that junctional embeddedness has a positive impact on an epistemic object’s jump particularly in the early periods of time, while the effect of structural embeddedness varies over the periods.