POSTPONED: Stefano Bianchini (BETA – Université de Strasbourg)

“Deep Learning in Science”

Abstract. Much of the recent success of Artificial Intelligence (AI) has been spurred on by impressive achievements within a broader family of machine learning methods, commonly referred to as Deep Learning (DL). This paper provides insights on the diffusion and impact of DL in the scientific system. We identify the ISI WoS scientific publications related to DL through a new list of search terms obtained via Natural Language Processing (NLP) techniques. We find that the number of articles contributing to DL techniques has grown considerably and steadily since the early 2000s, with a stronger upward trend due to methodological breakthroughs in the AI community. During the same period, we observe an exponential growth in the application of DL methods across scientific disciplines, both in terms of scientists involved and scientific outcomes achieved. How does that impact science? We make the case of the health sciences that are characterized by the early adoption of DL techniques, leading in turn to innovations of high societal impact. Our empirical analysis suggests that the use of DL in health sciences is associated with scientific excellence but not necessarily with re-combinatorial novelty. Our findings suggest that DL is more than a passing fad in science, but the adoption of these techniques will grow further in the future as they may add value to the process of knowledge creation.