What If AI Learns How You Generate the New? - A Preliminary Discussion of the Normalisation and Political Economy of Generativity

Abstract

Artificial intelligence systems have been developed for the most part on accumulated human artefacts. Systems that learn from recorded practice take a different object, since a demonstration, a workflow, a tool-use history, or a sequence of decisions records the course of an activity in place of its product. This paper asks what follows for political economy when the activity so recorded is one through which new possibilities are generated, and it develops the question through the case of research practice, where the question under investigation is itself frequently produced during the enquiry. The argument proceeds in five movements. It distinguishes four levels at which a generative practice admits description, namely outcome, representation, trajectory, and generativity, and holds these levels to be distinct, so that a record of a practice stands to the practice as a proxy stands to what it proxies. It treats the case in which a system maintains a representation of a particular other party and adjusts its operation with respect to that representation, and identifies the coupling so formed as an object lying outside that partition and held by whichever party operates the infrastructure sustaining it. It sets out a stage structure through which a situated practice becomes a portable capability: access, capture, legibility, abstraction, transfer, automation, valorisation, and enclosure. It describes the normalisation through which generative capacity comes to be treated as an ordinary economic input, and establishes a condition on any instrument whose application is triggered by a quantity of generativity. It then reads intellectual property doctrine as a diagnostic, holding that positive law has protected generative capacity through the inseparability of such a capacity from its holder, and that recording weakens the condition on which that protection rested. The normative proposal is stated as a constraint over each relation an arrangement touches, and it withholds any regime. Where generativity is relational, its protection runs through the construction and maintenance of domains affording occasions for its exercise, and the conversion of generativity into a further form of property would reproduce the ontology this paper questions.