Artificial Intelligence’s New Clothes? From General Purpose Technology to Large Technical System

Artificial Intelligence (AI) has been quickly labelled a General Purpose Technology (GPT) for its many uses and the high expectations built around a technology that can perform tasks associated with natural intelligence. However, for now, the claim “AI equals GPT" is premature, and eventually, taking into account potential future scenarios, it can turn out to be incorrect. In fact, though every GPT is an influential technology, not every influential technology is a GPT. Checking AI against the definitional criteria of GPT, we come to the conclusion that GPT is a misspecified model of AI: what was meant to be a concept for an individual technology in this case is stretched to cover a growing infrastructural, system technology. For example, the pervasiveness featured in the GPT concept seems to be qualitatively different from the largeness that modern AI demonstrates. In this paper, we suggest an alternative framework drawn from the literature on Large Technical Systems (LTS) as more fit to represent the nature of AI. We map the building blocks of LTS on AI and describe its state-of-the-art through this novel viewpoint. This is a timely exercise, as we witness the formation of an AI industry. A correct understanding of its core technology is needed to identify mechanisms at work, problems in place and eventually the dynamics of this new industry. The LTS framework offers a broader grasp of the infrastructural nature of AI as a technology, with more convenient categories to describe AI and measures to test empirically. We investigate how the implications of AI being an LTS entail the design of adequate public policies and firm strategies.