I gotta use words when I talk to you: Primed suspension of disbelief in views on agency in relation to Artificial Intelligence

Abstract

The public launch of generative AI systems in 2022 has prompted considerable popular interest in their potential to transform work and to achieve the long-sought goal of machine intelligence. While the uncanny abilities of Large Language Models to produce fluent text on almost any subject lends these ideas some plausibility, they are based on debatable, if not erroneous, claims about the capabilities of generative AI. Although rarely presented in these terms, these claims may be seen to reflect a limited, substantive conception of the agency of AI systems. Alternative conceptualisations of agency from other disciplines are presented that may offer potentially more fruitful ways of thinking about agency in relation to AI. The way that conceptions of the agency of AI systems is shaped by the language used to describe their behaviour is highlighted and opportunities for alternative conceptions of agency to inform a more critical analysis of generative AI are identified.