Seminar: Glen Berman (University of Melbourne, Australian National University)
Epistemic extractivism in GenAI
Some Generative AI (GenAI) platforms are introducing advertising, others are not. This paper argues that this apparent divergence in monetisation strategies obscures a deeper convergence. All GenAI platforms accumulate records of users' interactions through observability infrastructure, and all are positioned to develop AI capability markets from those records. Through close analysis of OpenTelemetry's semantic conventions for GenAI, which underpin observability infrastructure, the paper surfaces interaction telemetry as a distinct data object in the political economy of GenAI -- one that captures not only user states but the teleological labour through which users cajole GenAI platforms into completing desired tasks. This telemetry feeds two forms of extraction, distinguished at the point of commodification. Behavioural extraction commodifies user states for advertising markets, intensifying the logics of surveillance capitalism. Cognitive extraction commodifies teleological labour for capability markets, stripping it of its conditions of production and refining it into platform capabilities that progressively automate the labour from which they were derived. Together, these forms of extraction place users in a double bind: behavioural extraction deepens dependency on the platform, while cognitive extraction depletes users' economic value. Drawing on Grosfoguel (2019), the paper theorises this conjunction as epistemic extractivism -- a mode of accumulation that is naturalised by the operational framing of observability as debugging, monitoring, and system improvement.
Glen Berman is a Postdoctoral Research Fellow at the School of Historical and Philosophical Studies at the University of Melbourne, and a Senior Researcher at Australian National University. Across these roles, Glen studies AI and the science system, focusing particularly on how actors within the system construct and engage in evaluations of AI technologies.
Event Details
- Date:
- 14 September 2026 at 12:00 pm – 1:30 pm
Description
Epistemic extractivism in GenAI
Some Generative AI (GenAI) platforms are introducing advertising, others are not. This paper argues that this apparent divergence in monetisation strategies obscures a deeper convergence. All GenAI platforms accumulate records of users' interactions through observability infrastructure, and all are positioned to develop AI capability markets from those records. Through close analysis of OpenTelemetry's semantic conventions for GenAI, which underpin observability infrastructure, the paper surfaces interaction telemetry as a distinct data object in the political economy of GenAI -- one that captures not only user states but the teleological labour through which users cajole GenAI platforms into completing desired tasks. This telemetry feeds two forms of extraction, distinguished at the point of commodification. Behavioural extraction commodifies user states for advertising markets, intensifying the logics of surveillance capitalism. Cognitive extraction commodifies teleological labour for capability markets, stripping it of its conditions of production and refining it into platform capabilities that progressively automate the labour from which they were derived. Together, these forms of extraction place users in a double bind: behavioural extraction deepens dependency on the platform, while cognitive extraction depletes users' economic value. Drawing on Grosfoguel (2019), the paper theorises this conjunction as epistemic extractivism -- a mode of accumulation that is naturalised by the operational framing of observability as debugging, monitoring, and system improvement.
Glen Berman is a Postdoctoral Research Fellow at the School of Historical and Philosophical Studies at the University of Melbourne, and a Senior Researcher at Australian National University. Across these roles, Glen studies AI and the science system, focusing particularly on how actors within the system construct and engage in evaluations of AI technologies.