“Thinking About AI” (edited volume)

The edited volume Penser l’intelligence artificielle – Enjeux philosophiques, politiques et culturels de l’automatisation numérique (Thinking About Artificial Intelligence: Philosophical, Political, and Cultural Issues in Digital Automation) is out in French. It is edited by Anne Alombert, Alban Leveau-Vallier, and Baptiste Loreaux, published by Presses du réel. I had the pleasure of contributing to the book through a short chapter, alongside those of Grégory Chatonsky, Yves Citton, Judith Deschamps, Matthieu Garling, Pablo Jensen, Frédéric Kaplan, Jean Lassegue, Sofia Taipa, Pierre Cassou-Noguès and Gwenola Wagon, among others.

Here is the presentation of the book:

How can we think about artificial intelligence technologies in a way that is neither blind fascination nor irrational rejection? This book offers a critical and transdisciplinary reflection on the subject. Indeed, in the context of the rapid deployment of generative AI, this book looks back at the history of computer science and cybernetics and questions the philosophical, political, and cultural issues surrounding contemporary digital devices. What are the theoretical assumptions underlying the industrial innovations known as “artificial intelligence”? How does the current acceleration of technology affect institutions and societies? What are the effects of automatic text and image generators on linguistic and artistic practices? To answer these questions, the twenty contributions in this book bring together perspectives from different disciplines: philosophy, mathematics, anthropology, history of ideas, computer science, political theory, literature, translation, media theory, aesthetics, and visual arts. Beyond naive fascination and ignorant rejection, this diversity of perspectives sheds light on the challenges of present and future technological and anthropological transformations.

My contribution (in French) is a chapter entitled “Artificial intelligence as an acceleration of bureaucratic and capitalist domination.” In it, I share a certain frustration with the discourse on AI in the intellectual sphere and the effects of closure that, it seems to me to, tend to disqualify the most radical critiques of AI. According to some of the most influential voices in these debates, criticizing AI “en bloc” would be meaningless.

In an attempt to clarify the terms of the debate, I propose to think of AI both as a slogan and as a dispositif of power in the Foucauldian sense—a set of statements (discourse, legal rules, philosophical or scientific statements, etc.) and unspoken elements (institutions, architectural arrangements, machines, measuring tools, etc.) intended to stabilize the power relations associated with bureaucratic and capitalist domination.

Drawing on the history of information technology, I see the “AI” mechanism as a radicalization of the process of computerization of society, and from there, I attempt to identify its logics and anticipate some of its effects. Here is an excerpt:

Exacerbating the trends already observed in computerization, AI amplifies (…) the structural discrimination embedded in the rationalities and practices of institutions. As illustrated by the case of algorithms dedicated to “fraud hunting” in welfare agencies targeting the most vulnerable, the supposed gain in efficiency is often at the expense of those most exposed to social injustice.

This is not only because of the biases induced by the datasets used to train the models, nor because of the transposition of developers’ or users’ prejudices into computer code—as in the case of predictive policing software tested by law enforcement agencies that incorporates controversial criminological beliefs. It is also and above all because, as Dan McQuillan writes, AI “forms feedback loops with the rest of society” and the institutions that deploy it: “It is a ‘structured structure that becomes a structuring structure’” writes McQuillan in reference to Pierre Bourdieu’s concept of habitus. AI’s statistical optimizations based on past data can thus be likened to a form of crystallized institutional habitus. Once baked into a computer system, the discriminatory practices of institutions become even less visible and therefore less prone ton contestation, whether because of the “black box” effect (the difficulty of explaining the result produced by the algorithm) or of automation biases (humans’ blind trust in the recommendations produced by these automated systems).

By making a qualitative leap in bureaucratic automation, AI could also lead to new “disembeddings.” In his 1944 book “The Great Transformation”, economist Karl Polanyi used this concept to refer to the emergence of a self-regulating market now positioned as an imaginary institution organizing society, and thus to a disembedding of the economy from the society it was supposed to serve. A process that, according to him, contributed to the fascist reactions of the 1930s. With many social relationships now being reconfigured by the growing importance of machines and the increasingly marginal role given to humans, are we not at risk of seeing the moral economies associated with a vast number of social relationships being undermined, in favor of a tenfold increase in bureaucratic logic and opaque statistical optimizations?

(…)

The corollary of these developments is deregulation. In the name of “simplifying procedures,” accelerating deployment, and sacrosanct innovation, “legal barriers” must be removed. In France, in the name of the “national interest,” the state can now impose data centers on local authorities and to circumvent environmental rules or those intended to implement “technical democracy.” Immediately after his return to the White House, Donald Trump was quick to reverse the few rules relating to AI enacted by the Biden administration. This approach seems to be gaining traction among European Union leaders, who appear ready to water down the principles set out in the “AI Act” adopted in mid-2024. It matters little that this text—riddled with exceptions and essentially amounting to a system of self-regulation under the control of industry—has been denounced by NGOs for its harmful impact on human rights. Rather than embodying a third way between China and the United States, Europe prefers to sink into a deadly arms race.

AI has been made possible by a tremendous accumulation of data, capital, and resources under the aegis of multinational tech companies and the military-industrial complex. Today, it serves as a technocratic watchword to fuel a new cycle of industrial capitalism and the accentuation of power relations in complete disregard of people’s rights.