Selected Papers of #AoIR2024: The 25th Annual Conference of the Association of Internet Researchers Sheffield, UK / 30 Oct - 2 Nov 2024 Suggested Citation (APA): Quelhas Freire, S. (2024, October). Simulating Subjectivity - Baudrillard and the Political Economy of LLMS. Paper presented at AoIR2024: The 25th Annual Conference of the Association of Internet Researchers. Sheffield, UK: AoIR. Retrieved from http://spir.aoir.org. SIMULATING SUBJECTIVITY - BAUDRILLARD AND THE POLITICAL ECONOMY OF LLMS Sebastião Quelhas Freire University College Dublin Introduction Although the work of Baudrillard is semi-frequently evoked in descriptive critical assessments of new technologies, be it in the context of the risk of simulated social systems (Andrejevic, 2020) or in the context of an oversaturated media environment demanding increasing participation (Della Ratta, 2020), there is a lack in a productive integration of Baudrillard's work in contemporary studies of AI. This paper aims at contributing in this direction, by showcasing the utility of Baudrillard’s concepts for an understanding of immaterial free labour in the context of Large Language Models (LLMs). Context The popularisation of AI technologies and their proliferation in daily uses has sparked a discussion on the social and political risks of the offloading of social systems to automated ones (Andrejevic, 2020). Concurrently, debates over the political economy of digital systems are still divided among the lines of free digital labour (Fuchs, 2023) and digital rentier capitalism (Christophers, 2020). Since the launch of ChatGPT in November 2022, generative AI systems have become the hallmark of current and future technological developments, with LLMs being applied to innumerable digital applications and it being hailed as one of the next big steps in consumer products for daily use. Discussions on the political economy of artificial systems and their integration has neglected an in-depth appreciation of the problem of subjectivity and its relationship with processes of valorisation. Although, this is in part warranted as the hypothesised sentience of LLMs is merely apparent and ascribable only as user perception, the rise of LLMs as human-like communication generators presents an opportunity of re-introducing the problem of subjectivity in the political economy of digital media, as the promise of LLM implementation is that of generating human-like content without the requirement of actually relying on human subjects. The problem of subjectification has been examined extensively by the post-operaist tradition, through discussions of immaterial labour in the context of digital systems (Terranova, 2000; Virno & Hardt, 2010), wherein it is argued that the digital economy relies on the capture of the social knowledge of the networked interactions of subjects acting therein. However, the concept of immaterial labour, both in its precise definition and its theoretical utility, has yet to be clarified (Gandini, 2021), and the discussion of the political economy of digital media still neglects that the monetisation of digital media is concurrent with the distribution of subjects in such a way as to render them of use to processes of valorisation (Foucault, 1990). As such, scholarship on free labour currently relies on an under- theorised conception of digital subjectivity. Theoretical Framework - Baudrillard In order to bring to light this issue in the context of LLMs, I propose to recover the work of Jean Baudrillard, relevant in light of both its anticipatory dimension in relation to AI (Andrejevic, 2020) and its description of a specific subject of communication (Baudrillard, 2019; Baudrillard & Maclean, 1985), which anticipates the rise of communication- intensive social relations. To trace this description, I first highlight Baudrillard's concept of simulation, understood as logic of programmed anticipation and production of 'the real' by technical systems based on models without origins (Baudrillard, 2016, p. 80, 2019, p. 1). Following, I describe how both individual subjects and the social as "the masses" function in this system of simulation, explain how they are the locus of social demand for communication (Baudrillard, 2019) and how the social is conceptualised as the locus of confirmation of the functionality of an overall system of control (Baudrillard, 1983, p. 76). In a third instance, I pinpoint Baudrillard's sparse discussions of robots and AI, which he hypothesises could replace this functional role in the logic of simulation he identifies in humans (Baudrillard, 1995, 2016). Contending that free user labour in general carries the implied notion that subjects fulfil an operational and functional role in the system of data production and manipulation, I argue that this role is precisely what conversational AI systems automate. To grapple with this phenomenon, I propose the concept of ‘subject function’ and of 'the automatic writing of the world' (Baudrillard, 1995, pp. 101–102, 2000, p. 64) to describe the simulated output of LLMs, in view of their function of automating the generation of information. By examining Baudrillard’s notions of subject operability in the circuitry of capitalist value production, I aim to integrate Baudrillard’s discussion of subjects of communication, which anticipate debates of the role of users as data sources (Della Ratta, 2020), with the novelty offered by the communication automating role of conversational LLMs. Application - LLMs Drawing from the recent phenomena of the sale of Reddit communications content to OpenAI as training data (Orland, 2024), I propose the notion of digital common as the pre-trained collected and recorded data of actual human communication through digital systems. The digital common I argue, constitutes the pre-condition for LLMs output generation and consequentially, for its subsequent monetisation through subscriptions and licensing fees. Additionally, the inclusion of collective human communication, located outside of strictly economic spheres, into processes of valorisation such as these, demand increased attention to be paid to role of subjectivity in the context of AI. Heading this demand, I propose the framework of the subject function as expounded in the work of Baudrillard, in both its individual and collective aspects, as a necessary conjuncture to understand how commercial applications of conversational LLMs fit into the broader landscape of digital political economy. I suggest that the role played in this specific application derives from the appropriation of freely generated user-data in how it constitutes the digital common and as carrying a specific conception of subjectivity. Baudrillard's concepts of subjects, masses, and society as functional to the social logic of a capitalist mode of domination, wherein the commutability of signs takes over production as the main source of value production, serve here as probing elements in evaluating this newfound role of the digital common in new processes of valorisation. Finally, through an analysis of Baudrillard's hypothesis that the functionality of subjectivity can be replaced by automatic systems, I explain the context of an already simulated reality of automated social systems and anticipate what the popularization of these systems might entail for the disappearance of a political subjectivity of communicating users in the face of fully automated subjectivities. Thus, I propose that Baudrillard's dense and polemical conceptual apparatus serves critical AI studies by placing the onus on the identification of human social agency in the context of processes of valorisation Baudrillard himself could not fully anticipate. References Andrejevic, M. (2020). Automated Media. Routledge. Baudrillard, J. (1983). In the Shadow of the Silent Majorities, or, the End of the Social, and Other Essays. Semiotext(e). Baudrillard, J. (1995). The Virtual Illusion or the Automatic Writing of the World. Theory, Culture & Society, 12, 97–107. Baudrillard, J. (2000). The Vital Illusion. Columbia University Press. https://doi.org/10.7312/baud12100 Baudrillard, J. (2016). Symbolic Exchange and Death (Rev. edition). Sage Ltd. Baudrillard, J. (2019). Simulacra and Simulation (S. F. Glaser, Trans.). Univ. of Michigan Press. Baudrillard, J., & Maclean, M. (1985). The Masses: The Implosion of the Social in the Media. New Literary History, 16(3), 577. https://doi.org/10.2307/468841 Christophers, B. (2020). Rentier Capitalism: Who Owns the Economy, and Who Pays for It? Verso. Della Ratta, D. (2020). Digital Socialism Beyond the Digital Social: Confronting Communicative Capitalism with Ethics of Care. tripleC: Communication, Capitalism & Critique. Open Access Journal for a Global Sustainable Information Society, 18(1), 101–115. https://doi.org/10.31269/triplec.v18i1.1145 Foucault, M. (1990). The History of Sexuality. 1: An Introduction. Vintage Books. Fuchs, C. (2023). Media, Economy and Society: A Critical Introduction (1st ed.). Routledge. https://doi.org/10.4324/9781003391203 Gandini, A. (2021). Digital Labour: An Empty Signifier? Media, Culture & Society, 43(2), 369–380. https://doi.org/10.1177/0163443720948018 Orland, K. (2024, February 23). Reddit cashes in on AI gold rush with $203M in LLM training license fees. Ars Technica. https://arstechnica.com/ai/2024/02/reddit- has-already-booked-203m-in-revenue-licensing-data-for-ai-training/ Terranova, T. (2000). Free Labor: Producing Culture for the Digital Economy. Social Text, 18(2), 33–58. Virno, P., & Hardt, M. (Eds.). (2010). Radical Thought in Italy: A Potential Politics (Nachdr.). Univ. of Minnesota Press.