AI studies meet comparative epistemology

In Western modernity, the concepts of machine, organism, and language have played fundamental roles in the formation of science and the humanities, recombining into paradigms of increasing complexity at each turn of their evolution. They emerged as abstract forms and then as scientific objects claiming their own discipline. They developed into philosophical paradigms, ideological worldviews, political mentalities, and cultural metaphors, in which their role exceeded the scale of the originary observation and conception. To clarify the underlying paradigms of AI, this essay navigates the contested and shifting status of these epistemic mediators in the history of knowledge, in a way similar to previous observations made by Ernst Cassirer (1923) on symbolic forms, Ludwik Fleck (1935) on thought styles, and Lorraine Daston (2015) on epistemic images, yet resisting emulating illustrious predecessors such as Hegel (1807) on the figures of thought (Gestalten des Denkens). Summarily, one may say that machines inspired mechanicism, organisms organicism, and languages structuralism. Yet one has still to explain the circulation, contamination, and convergence of these paradigms, and the way they eventually contributed to the making of the technical mentality of AI.

The modern concepts of machine, organism, and language ramified across disciplines: for instance, in between engineering and biology, as well as in between linguistics and anthropology; and then, in the developments across cybernetics, information theory, and, ultimately, AI. In this latter case however, they have been rarely analysed together. As is well known, the belief that a machine could be designed as an organism was a key tenet of cybernetics, whilst information theory established the first material synthesis between language and machine, at the same time structuralism sought to describe kinship as a language. The philosophy of science and technology has often addressed the hybridization and contamination of the machine and organism paradigms, whilst leaving language aside. This essay attempts to interweave again the lineage of language into the one of the machine and organism, understanding this process as a political translation across disciplines.

Contemporary AI models, such as Large Language Models, can be seen as the convergence of the genealogy of machine, organism and language paradigms—a philosophical conurbation that emerged to encode, control and automate social relations into a new morphology. Although AI is commonly perceived as a mathematical achievement, it should instead be investigated through the paradigms that shaped its development and the worldviews these paradigms mediated. This essay, however, does not frame the paradigms of machine, organism and language as historical a priori: rather, as social mirrors diffracting other social mirrors, that is as ‘solutions to the problem of knowledge [that] are embedded within practical solutions to the problem of social order’ as Steven Shapin and Simon Schaffer (1985: 15) once registered in their study of seventeenth century natural philosophy.

Why is this exercise of political and epistemic translation, or comparative epistemology, necessary? Globalisation and late modernity scholarship have witnessed the rise of macro-paradigms at the crossroads of several disciplines that require increasingly complex research plans (see the Anthropocene debate, amongst others). In another season of intellectual complexity, at the peak of structuralism, Michel Foucault (1966/1994) attempted a similar synthesis of the modern episteme around the paradigms of life, labour and language. In these, one can recognise the traditions of natural sciences, political economy and linguistics respectively. Foucault’s main shortcoming was an excessive preoccupation with detecting isomorphism, rather than causality, in their developments: at the time, structuralism was advocating a focus on synchronic rather than diachronic configurations as a way to emancipate both philosophy and linguistics from the weight of historicism. Looking carefully, the categories of life, labour and language carry different historical and political weight and cannot be read on the same abstract continuum. Labour plays a fundamentally different role in historical dynamics than language. Although Foucault was concerned with rendering the two equal, philosophical inquiry today should acknowledge that the sciences of language rarely include labour in semiosis, whilst the sciences of labour rarely include language in production. Foucault’s concerns about isomorphism should be turned into concerns of translation, contamination and mutation.

As Marx (1858/1973) argued in the introduction to the Grundrisse, the modern category of labour is not a given, but a construction—the final articulation of deep historical processes of social abstraction. Indeed, due to its historical magnitude and complexity, the politics of labour cannot be left to the same Wunderkammer of fossil skeletons, as Foucault did. Foucault’s operation, nevertheless, can be read at the same level as Marx’s, as the intention to make the politics of labour a legitimate concern by positing labour as a scientific object. The aim of the present essay is similar, but not identical: it is about tracing the social and economic roots of key paradigms such as machine, organism and language, in order to illuminate them as mirrors and indices of the social order.

Foucault’s analogy of life and language failed to credit the last lecture that Ernst Cassirer (1945) gave in New York just before his death. His lecture consolidated the term ‘structuralism’ in English-speaking academia, framing it as a transformation of German organicism, which first rendered language as an (organic) system. Cassirer identified then a parallel between ‘structures’ in Saussure’s linguistics and Cuvier’s anatomy, whilst recapitulating his life-long project of a comparative epistemology of modern thought. In his philosophy of symbolic forms, Cassirer (1923, 1925, 1929) had already investigated science, art and mythology as emerging all from the same disposition of human civilisation toward abstraction. He defined, for this reason, the human as a symbolic animal (Cassirer 1944: 26). His inception of Kulturphilosophie legitimates the study of cultural forms beyond science and philosophy, predating the rise of cultural and media studies. Yet what his analysis (as well as Foucault’s) missed is an account of technology (Freudenthal 2017).

The present study advances a comparative, historical and political epistemology of AI, reading the genealogy of notions such as machine, organism and language that contributed to grounding AI research and design. It should be remembered that concepts often change their political polarity. For instance, at the end of the eighteenth century, Kant (1790) defined the concept of organism also to contrast the hegemony of theology on free will and the new social space that the new middle class was occupying. In the mid-twentieth century, cyberneticians (Wiener 1948) mobilised their ideal of the organism to foster the design of new machines and the agenda of military automation. Similarly, the machine was the form of social discipline in the industrial age (Marx 1867; Berg 1980), but after 1968, counter-culture and critical philosophy adopted it as a metaphor of empowerment against logocentrism: see, its role in French post-structuralism (Deleuze and Guattari 1972) as well as in the cyborg manifesto debate (Haraway 1985). Again, in information theory (Shannon 1948), the statistical analysis of language was key to improving the capacity of communication channels with no reference to the problem of meaning, whilst, in structuralist anthropology, the analysis of language was considered a point of access to the social unconscious (Lévi-Strauss 1949). Writing a historical epistemology of philosophical, scientific and technical notions means venturing into such contested territory.

The current essay deconstructs the making of AI as a composition of these three foundational paradigms that for historical reasons are presented in the following order: machine, organism and language. It questions two important analogies built upon these paradigms, namely the organism–machine and language–machine analogies, leaving aside, for reasons of space, the language–organism analogy (Fig. 1).

Fig. 1: Diagram illustrating the genealogy of the paradigms of organism, machine, and language in relation to AI studies