(2024). Scientific Paradigms, Graphics Processing Units and the Evolution of Artificial Intelligence Academy of Management Proceedings, 2024 (1).
A sudden shift in scientific and technological paradigms lies at the heart of recent advancements in artificial intelligence (AI). Around 2012, traditional symbolic AI gave way to neural networks (NN) as the dominant approach for AI research. This coincided with the sudden successful application of graphics processing units (GPUs) as computational technology. GPUs had been invented for a different application, i.e. accelerating complex graphics displays, mostly in video games. We claim that these developments reflect the nature of breakthrough innovations and have implications for regions competing to become AI leaders. We investigate the role of expertise in GPUs for the uptake of AI innovation across regions globally. To this end, we construct a global database covering 2,088 urban areas for the period from 2000 to 2020. The data encompass a broad set of measures describing AI research and innovation activities, based on publications, patents and startups. We document the ascendancy of neural AI and its association with GPU expertise. Panel OLS and IV regressions demonstrate that after 2012 GPU- and NN-related human capital had a strong effect on the growth of AI-related patents and startups. We discuss implications for innovation policy.
