The Machinic Plane: a topology for working with Artificial Intelligence in universities

This paper offers a methodology for grounding considerations of the uses of Artificial Intelligence, aligning different forms of machine learning with the traditions and conventions that legitimate the diverse forms of knowledge that characterise contemporary universities. This requires reconsidering the nature of agency, given that the founding definition of Artificial Intelligence is a machine that behaves in ways that would be called intelligent if a human were behaving that way. This paper follows the school of thought that sees AI as "agency without intelligence"; it generates knowledge forms by interacting with its environment, changes its state without direct external instructions, and modifies its behaviour based on inputs. As such, "artificial agency" massively extends human capabilities, but without applying unique characteristics of human intelligence, such as empathy, emotional understanding or moral reasoning. Deciding how best to deploy this new form of agency at a granular level within a university, enhancing teaching, research, and administration across disciplinary and professional contexts, requires protocols that enable the design and purposes of specific AI applications to be evaluated. A candidate for this process is Legitimation Code Theory, a widely used framework that offers a set of dimensions for exploring different forms of knowledge creation and use. However, Legitimation Code Theory's potential to deepen understanding of the implications of Artificial Intelligence for knowledge creation is limited by its founding assumption that a "knower" is always a socialised human with human intelligence. To overcome this limitation, this paper offers the "Machinic Plane" as a translation device that allows individual instances of AI to be understood as a balance of power between human and machine, moderated by the extent to which the workings of a specific application have been made transparent by its developers, enabling a user to understand the consequences of their actions. The practical utility of the Machinic Plane is tested by examining in detail the four states it generates: instrumental practices, in which machine agency is dominant and the level of transparency and insight into how an artificial agent is designed and programmed is negligible; dependent practices, a state characterised by naive human use of AI applications; engaged practices, where the human in the partnership is both in control and has sufficient working knowledge to extend their capabilities beyond their physical limitations; and detached applications, the rapidly emerging world of fully automated artificial agents that are testing longstanding conventions of responsibility and accountability. As both an analytical device and a canvas for making decisions, the Machinic Plane is a topology rather than a typology, in that the positioning of a given AI application in a specific quadrant is as much a matter of human choice as it is of the design and characteristics of the model's design and programming. Consequently, the ability to make informed decisions on the acceptable uses of Artificial Intelligence is critical. The device of the Machinic Plane contributes to this decision-making process, and to the wider objective of reclaiming the purpose of higher education.