Three Essays on Technology Adoption in Firms: Studying Three Aspects of the Diffusion of Artificial Intelligence

Recent developments in the field of artificial intelligence have taken business and academic literature by storm and made several fundamental questions on the topic both pervasive in the public debate and relevant to policy makers. This thesis investigates three aspects of this rapid diffusion (in the broader context of digital technologies diffusion) and proposes answers that are partial in nature due to the lack of hindsight on this recent phenomenon. The thesis sets out by taking the general question of digital technology diffusion at its more general level: the one of economies. This essay, elaborated while working at the European Central Bank, aims at designing a way to produce prediction on the extensive margin of diffusion of a certain technology in a certain country for a certain year by leveraging the information available regarding not only past diffusion of technologies in the focal country or the diffusion of the focal technologies in other countries but also the diffusion of other technologies in other countries in the past. The idea and theory is not new, it has been widely studied in economics and other fields (such as the marketing literature) for a few decades, but the essay proposes to tackle the problem of forecasting the diffusion at the level of the economy using recent developments in the field of statistics (such as the beta-regression) and of statistical computing (such as No U-Turn Sampling for Monte Carlo Markov Chains) to produce a density forecast rather than a point estimate. While this chapter relates to digital technologies in general rather than artificial intelligence specifically (among other because there was at the time relatively little available aggregated data), the model devised can be used on artificial intelligence applications in the future as the topic gets measured and aggregated at national level. The second chapter takes a closer look at the question of the sourcing of applications based specifically on artificial intelligence. In this essay, we use a discrete choice model to analyse the extent to which firms in different industrial sectors of the economy prefer to develop their AI solutions in-house or outsource that decision to third parties, and whether there is complementarity or substitution between the choice of developing AI solution in-house or to procure it from elsewhere. This chapter highlights the disparities in the form of AI adoption across sectors as well as in diverging sectorial preference for complementarity versus substitution. The final chapter studies a specific form of delivering artificial intelligence in the economy: the no-code, low-code AI software. In a time where generative AI is getting a lot of attention, no-code, low-code AI systems enable non-technical workers to have access to some version of the machine learning algorithms while retaining more control over the workflow. In this paper, we study the effects of having skills in those systems on the offered wage at the level of a job posting and, hence, at the level of the worker. Moreover, we study the change of the number of postings issued by a firm after it posts its first posting for a worker with skill in such software (which probably correlates with the adoption of the technology itself). We find that having no-code, low-code AI skills is more valuable for non-technical workers than for workers specialised in computers or mathematics and, perhaps suprisingly, that the adoption of no-code, low code systems increases the demand for workers both for the occupation of computers and mathematics, and for business and financial operations.