top of page

Knowledge Accumulation and Artificial Intelligence: A Marxian Perspective

19 March 2024

Knowledge Accumulation and Artificial Intelligence: A Marxian Perspective
Article PDF

Humanmechanical, Hans Ruedi Giger; Image credit: Medium

The article presents a thesis about the reciprocal relation between the histories of knowledge and that of labour. It argues that from the medieval period knowledge was used to separate labour from the intellectual act. The article posits that in the current age the knowledge-capital-labour relation is pivoting on the capabilities of AI through a process similar to the retention of dead labour as capital as described in the classical marxian approach.

We live in an age in which information grows exponentially. A large part of our daily toils depends on feedback loops where existing data generate more data. We take part in this accumulation when driving to work and when shopping online, when engaging in social media and when working—be it in software development or at a grocery store. In all cases, we typically leave digital traces, and to a large extent these traces are used to generate more data. Large language models such as Chat GTP are but the latest high-level manifestations of this ubiquitous runaway tendency. And while accumulation of data, information and knowledge is most evident in the high-tech sectors of the rich world, it plays a role also in perpetuating 19th-century-style exploitation in most of the planet. We should be aware that breakthroughs in artificial intelligence (AI) and cyber warfare go hand in hand with super-exploitation, WW1-like trenches and an overall rise in slave labour. 

The following pages focus on the mutual dependency between accumulation of knowledge and accumulation of wealth under capital. A symbiotic relationship between two self-accretion processes - knowledge and capital - which is very evident in the leading-edge sectors of the economy. The aim here is to propose a critical and historical perspective on this mutual dependency. We are constantly reminded that technological innovation is driven by the prospect of financial returns and that economic growth is ever more dependent on breakthroughs in basic science and engineering. What is lacking is a broader view capable of looking at this dependency through a historical lens and providing insights on what is at stake not only in high tech sectors, but in the economy at large. 

For millennia - including the first two centuries of capitalist accumulation (1) - production of knowledge and production of wealth have coexisted and expanded not because of their closeness, but because of their separation. Labour produced wealth; philosophers and theologians - later managers and engineers - produced the knowledge necessary to perpetuate the separation of labour from thought—a separation guaranteed by the church in the middle ages and by the Fordist assembly line under capital.

It is instructive to look back at the Greek origins of abstract thought. (2) The first philosophers contributed only indirectly, if at all, to the wealth of their city-states. On one side there was Theory, an end in itself with no immediate application; on the other Poiesis, the material production and reproduction of society. The context of the first abstract speculations was one of separation. A minority (free citizens) could discuss on an equal footing thanks to the invention of self-government and the flourishing of trade; and this was made possible by the labour of others—slaves, women and foreigners. It is in this privileged setting that a new mode of knowledge production began for the first time to emerge. A production arising from conversations no longer aimed exclusively at reproducing existing knowledge and power structures but also capable of innovating, of feeding off their own results, of addressing abstract topics such as justice, numbers, beauty and love. It will be argued that these early intellectual practices are the founding seeds which have led to the science and technology of today; in other words, that the spectacular growth of human knowledge (admittedly: with countless crises, setbacks and regressions) is ultimately based on the same positive feedback dynamic which first enabled, two-and-a-half millennia ago, the output of one conversation to become input for the next. 

So, back to the present, the position will be defended that what makes our times truly unique is not knowledge accumulation per se (the process that has led from philosophical dialogues to the discovery of black holes); instead, it is the blending of this accumulation with a relatively more recent and, prior to the information age, well distinct form of cumulative growth: the valorisation of capital. Put differently, the historical novelty lies not in self-referential accumulation of knowledge but in the fact that this accumulation is in many sectors no longer separate and distinct (as it was until recently, also under capitalism) from the practical aim of producing wealth; knowledge, information and data are now fully ‘productive’ also from a capitalist standpoint; in many sectors, these two forms of growth (knowledge and capital) operate simultaneously, within the same context and making use of the same ‘mental’ labour. We are now faced with a DNA-like double helix where two spirals (accumulation of capital and accumulation of knowledge) have multiple connections and continually overlap and reinforce one another. A self-reinforcing symbiosis capable of influencing the entirety of social reproduction—old economy sectors included. 

The following pages investigate this simultaneous accumulation making use of categories derived from the Marxian critique of political economy. The outlook is both synchronic and diachronic. The first four sections set the stage, proposing a Marxian interpretation of knowledge accumulation and introducing two categories – code and context - that will play a key role throughout the article. The following two (5 and 6) test these conceptual tools historically, first looking into the origins (in ancient Greece) and then into the intensification (with modern science) of knowledge accumulation. The final three sections propose an interpretation of knowledge accumulation in relation to capital with the aim of providing new insights into ‘cognitive capital’ (section 8) and AI (section 9). 

1. What is accumulation?

In Das Kapital, a century before cybernetics (but with a profound assimilation of Hegelian philosophy), Marx was able to conceive accumulation not as a summation of unrelated parts, but as a system, endowed in itself with a principle of reproduction and growth. Under capital, in Marx’s view, the product of past, or ‘dead’, labour survives, conditions and informs present, or ‘living’ labour, and is thus able to direct it to its self-reflexive goal—the continuous growth of the social surplus taken away from consumption and reinvested in production. (3) Similarly, in the perspective of this article, accumulation of knowledge is seen as a system, the specific product of which coincides with the conditions necessary for its future expansion; where the output, in other words, is continuously reinvested in the process that generates it and must therefore - precisely for this reason - be of the same ‘substance’ of the input.  

Production is the specific object of Marx's investigation; not distribution and exchange, which (although indispensable for understanding the capitalist totality) are more relevant in pre-capitalist social formations, aimed at promoting consumption (of the dominant classes), not accumulation. In particular, Marx investigates the mode of production. That is, those characteristics and conditions, both technical (tools, machinery, productive forces in general) and social (the position of the social classes with respect to the ownership and use of productive forces) which, in a capitalist society, impose the constant reference of living labour to past labour and therefore fuel - from within production - a process of cumulative growth. Having reached, driven by the reality of capitalist accumulation, this mode of production perspective, Marx is then able to look back to the history of past social formations and identify the 'specific differences' (in particular, the systematically reproduced separation of labour from the objective conditions of its reproduction) that make the capitalist mode of production an object (in the strongest sense) of critical analysis.

Karl Marx in Trier; Image credit: DW
Karl Marx in Trier; Image credit: DW

Similarly, in these pages knowledge will be viewed from the internal perspective of the mechanisms that preside over its production. Not so much the content and use (and therefore ‘consumption’) of ideas but rather the conditions that enable ideas to be used productively in the future, in a self-sustaining accretion process. The analysis will be centred - as is the Marxian one - on the particular social relations and technical conditions that enable a constant reference of living (and, in this case, mental) labour to dead (mental) labour, thus generating, as in the capitalist realm, cumulative and potentially unlimited growth. With an aim also of shedding light (as does the Marxian perspective, albeit in a different context) not only on knowledge accumulation but also on the characteristics of those forms of knowledge that, though constantly reproduced, do not trigger (in this like pre-capitalist social formations) compounding growth. ‘The human anatomy contains a key to the anatomy of the ape’, as famously expressed by Marx in the Grundrisse.

2. The four quadrants of knowledge production

A very general perspective is therefore proposed. A perspective capable of relating, within this ‘mode of production’ framework, forms of knowledge as diverse as production techniques on one side and, on the opposite extreme, scientific abstractions. 

The table below presents two distinctions. On the vertical axis the modes - non-accumulative and accumulative - in which knowledge can be produced. On the horizontal, the aims of knowledge production: either an end in itself, with no immediate practical application or, on the opposite side, practically motivated towards the reproduction of society and the creation of wealth. The resulting four quadrants are, obviously, ‘ideal types’. In real-life, knowledge is typically generated across quadrants. These quadrants do, however, provide a good schematisation in view of  the arguments that will follow. 

The first quadrant (non-accumulative production of knowledge with an end in itself) has its most typical examples in language and play. If often directed to specific goals, these activities nonetheless owe their existence precisely to the rules they are made up of and which they continually reproduce. (4) They are thus, in an important sense, ends in themselves. And, although we sometimes ‘make up the rules as a game goes on’, (5) they are essentially repetitive and, therefore, not cumulative activities.

In the second quadrant (accumulative production of knowledge with an end in itself), theoretical knowledge is at work, from its 'philosophical' origins in ancient Greece right up to the activities pursued in today's universities and research institutions. The distinguishing feature of this form of knowledge is the absence of practical aims and immediate linkages with action (6). The production of theoretical knowledge typically consists of confirmations, refutations, additions or improvements to previous knowledge. Its starting point is the stock of existing knowledge. This implies that the subjects taking part in its production must undergo long periods of training. Thus, whereas to take part in the knowledge production of the first quadrant (language and play), it is sufficient to be part of a community of speakers, in the second quadrant knowledge has traditionally been the exclusive privilege of classes and social contexts (for example convents and universities) separated from the world of necessity and material production.

The third quadrant (non-accumulative production of practically oriented knowledge) is symmetrically opposite to the second. It is where the know-how that accompanies the creation of wealth and, more generally, of all the material and cultural aspects involved in the reproduction of society is at play. It is a form of knowledge often indistinguishable from the actions in which it is embedded. While it is frequently generated in the same context in which it is used (working skills), starting from the industrial revolution it is also sourced from the 'separate' science produced in the second quadrant (hence the rise of technology); in all cases, it is adapted to the specific, practical needs of the contexts in which it is applied. 

Only in the fourth quadrant (accumulative production of knowledge aimed at the production of wealth) do the two forms of accumulation take place within a single process. Here, as in the second quadrant, the outcome of knowledge production is the generation of ever more knowledge. However, with an aim not of producing knowledge per se but of contributing to the wealth-creation process for capital. This form of knowledge production, typical of the information age, closes the cleavage between academia and the workplace. Capital is no longer compelled to ‘buy’ knowledge from the second quadrant (or receive it as a 'free gift') before 'putting it to work' in the third: it can appropriate innovation directly within the production process, that is, within the same context of its valorisation. Following this argument, there is a sense in which knowledge can itself be seen as a form of capital. Here it will be defined - using a term introduced in the Italian Marxian debate by Lorenzo Cillario - 'cognitive capital'. (7) 

Artificial intelligence (AI) is also located in the fourth quadrant (it accumulates data, information and knowledge in the context of the capitalist production of wealth). However, by removing labour, its relation to capital is more problematic. On the one hand, like machines, AI is a form of fixed capital; its purpose (like all forms of fixed capital) is not to facilitate production processes per se but to reproduce capital—which, as we shall see, in the context of knowledge production implies skewing the knowledge produced (its use value) toward the preservation of dominant relations of production. But there is more to AI, it will be argued. It can also be thought of in analogy to fictitious capital: an attempt, like credit and finance in the economic sphere, to overcome all internal barriers to accumulation, with the associated risks of periodic and ever harsher confrontations with reality.  

3. Contextual and codified knowledge


The following distinction - between 'contextual' and 'codified' knowledge - embraces all four quadrants and cuts through traditional distinctions such as ‘know what’ and ‘know how’, or ’theoretical' and 'practical' knowledge. (8) Contextual is any form of knowledge embedded in the context in which it is generated and used. It is the outcome of the actions and experience of the subjects that possess it. It is not necessarily articulated in words. It is personal and informal and retains these attributes also when it is socialised and enters into the life and actions of a community. Since its existence depends on the individuals who possess it - on their memory, experience, actions, and traditions - it is not simple to transfer beyond the context (time, space and culture) in which it originates.

Codified knowledge is defined instead as a form of knowledge translated, through an explicit set of rules, into a form, or code, that can be used and retain its meaning in different epochs, locations and cultural settings. To codify knowledge is to relocate portions of living, meaningful activities into a space external to living bodies—an inanimate space that can outlive these activities and project their meaning beyond the contexts in which they originate. This definition of code excludes spoken language, which can only exist thanks to speakers, that is, the interaction of living bodies and brains. But it does include writing. The latter relocates speech into a space  external and independent of speakers. It thus enables language to be conserved and produce effects also when speakers are no longer present—or alive, as with Latin or Sanskrit. Similarly, gestures can be codified. Like speech, gestures can be made independent of the living bodies that originally perform them. A potter’s wheel, for instance, transfers the gesture of shaping a vase into the regular movement of a simple machine: a meaningful activity (shaping a vase, in this case) assumes a reality independent from the artisan—just like language, through writing, exists independently from living speakers. Tools and machines - even of the simplest, analogical kind - all enable, to varying degrees, similar actions to be carried out in similar fashions by different people, in different locations and at different times; actions that (just like speech acts) would disappear, absent codification, the very instant they are performed. 

In general, codified knowledge is therefore at play wherever skills, competencies and knowledge are externalised and rendered reproducible beyond their context of origin. (9) Reproducibility is typically achieved by reducing the degrees of freedom within which actions (wether ‘communicative’ or ‘productive’) can be performed. Meaningful actions thus become more homogeneous and predictable. Their underdetermination is reduced. The differences with which individuals and cultures perform them are either erased or reduced to a minimum. 

installation view, ‘Cybernetics of the Poor’, Phonosophia, Camila Sposati, 2020, Tabakalera, San Sebastián. Photograph: Mikel Eskauriaza; Image credit: FRIEZE
installation view, ‘Cybernetics of the Poor’, Phonosophia, Camila Sposati, 2020, Tabakalera, San Sebastián. Photograph: Mikel Eskauriaza; Image credit: FRIEZE

Crucially, codification can also be applied to codified knowledge itself. The printing press, for instance, codifies handwriting, the latter being a codification of speech (10); similarly, the spinning wheel codifies the pinching, twisting and winding movements performed with the drop spindle—itself a codification of the gestures originally involved in producing yarn without external aids. In our age, software, algorithms, and, more generally, information and data typically codify knowledge that has already undergone multiple codification processes. 

Codified knowledge is thus a means of exchange of cognitive contents. It is a form of knowledge that can be understood outside of the context in which it originates and can thus encourage the transmission and exchange of information between different and often distant subjects and situations. Since it does not have to rely on living memory and repetition in order to be preserved (as is the case, for example, with the muscle memories of artisans or myths in pre-literate civilisations), it can also be thought of as a store of information or, to introduce Marxian terminology, as dead mental labour.  

This brings us to the dialectical nature of the relationship between codified and contextual knowledge. Consisting of dead labour, codified knowledge, by itself, is inert and useless. Codes are not self-interpreting. To live and operate, they need to point to contexts. In their written form, they often require indexicals and deictics (‘here’, ‘this’, ‘I’, ‘now’ and the like). They must be learned, adapted to an environment, become an object of communication and thus enter into the actions, life and work of individuals and communities. (11) To be used, codified knowledge must, in other words, be converted back into living, informal, contextual knowledge. 

Knowledge production can thus be seen as a circle in which the two forms, contextual and codified, are continuously converted into one another. On the one hand, the concrete, contextual and essentially vague knowledge originating in life, actions and experience is transformed, to be exchanged and preserved, into abstract, reproducible, codified knowledge. It is rendered explicit, formalised, purified of references to people, situations and facts. At the same time, to be used, codified knowledge has to be re-immersed into the specificity of a form of life and a context. It must, in other words, be once again transformed, through learning and use, into contextual knowledge.

This brings us back to the subject of accumulation.

4. Simple circulation and accumulation


The circle described above, from contextual, to codified, back to contextual, also admits, being a circle, a different point of departure and a different end—no longer contextual, but codified knowledge. This leads to a very different outcome. In the first case (contextual-codified-contextual), the 'sense' of the circle is to facilitate the reproduction and consumption of contextual knowledge. The aim of the process is use. Knowledge does not expand; it is preserved. Codified knowledge is only an intermediate phase, subordinated to the concrete horizon of contextual knowledge. Wittgenstein's 'language games' - not by chance embracing the first and third quadrants - are good examples of this non-accumulative (or simple, as we shall define it) circulation of knowledge. (12) When, on the contrary, the circle opens and ends in the codified form, its sense is no longer use, but self-reflexive and compounding growth. The function of contextual knowledge becomes to give life to codified knowledge by interpreting it, introducing it into new contexts and thus enabling the creation of more and enriched codified knowledge at the end of the process. From a mere means of exchange and store of information, dead mental labour (codified knowledge) here becomes the true subject of the process. A subject that can condition and direct the use of living mental labour (contextual knowledge), subordinating it to its goals. This is what takes place in the second and fourth quadrants. The only difference between the two being that in the former (scientific and academic activities), knowledge production is separated from action; in the latter (cognitive capital, artificial intelligence), this production is an integral part of the processes of wealth creation.

Marx confronts us with the consequences of the inversion of a similar circular conversion process—that of commodities into money. In one case, 'simple circulation', the commodity is situated at the beginning and end of the circle. Here, as in the simple (or non-accumulative) circulation of knowledge (from contextual to codified back to contextual), the horizon is of quality and use. Money is but an intermediate phase, the purpose of which is to promote the circulation of goods aimed at satisfying concretely human needs through its function of a means of exchange. It is through the inversion of this circle, from money to commodities back to money, that capitalist accumulation finds its most appropriate definition. In this process, money is transformed from a mere means to an end. Its function as store of value is here crucial. Human purposes lose relevance. More than circular, the motion becomes that of an expanding spiral. The conversion of money into commodities, which in simple circulation was aimed mainly at satisfying needs, now becomes an intermediate step in the potentially never-ending growth of abstract value.

In both realms, knowledge and the economy, two distinct conversion processes are thus possible: simple circulation and accumulation. In one case, the aim is to satisfy concrete needs through consumption and use; in the other, the compounding growth of abstract exchangeability. This leads to the fundamental question: What conditions enable this change in dominance - from concrete to abstract - to occur? How is it possible for what in simple circulation are mere means (codified knowledge, money) to become ends in themselves? 

For economic theory, there are two possible answers. The distinction between them coincides, in Marxian terminology, with the distinction between 'bourgeois' economics (both classical and neoclassical) and the critique of political economy. 

According to the first answer (bourgeois economics), the transition from simple circulation to accumulation can be placed within the narrative of an evolution from simple to progressively more complex social formations. Accumulation (assuming it is recognised as such) can be seen as the coherent development of premises already implicit in the movement of simple circulation, from which it can, so to speak, be analytically 'deduced'. In this interpretation, there is no need for history; logic and mathematical modelling can suffice. Not by chance, the privileged field of investigation for bourgeois economics (that is, contemporary economic theory) is a timeless representation of circulation (the market), not production, let alone its cumulative nature. 

For the second interpretation (critique of political economy), accumulation is instead a critical transition; it is the contingent outcome of a historical rupture, first produced by external events and then continuously reproduced by the mechanism of accumulation itself. It cannot, therefore, be ‘deduced’ from simple circulation; on the contrary, it can be understood only by extending the investigation to those historically disruptive and improbable elements, which are at the same time pre-conditions and results of the process. (13)

For Marx, accumulation can be understood only by recognising the importance of a decisive historical event: the separation of workers from their means of subsistence and production; their reduction, in his terms, to mere ‘working capacity’; to potential labour, capable of being actualised - of becoming living labour - only within the relation (formally free, but substantially lacking alternatives) with the dead labour (machines and organisation) already accumulated by capital. This separation – the ‘so-called primitive accumulation’ (14) - can be historically traced back to the enclosures and the consequent expulsion of peasants from the land they previously ‘belonged to’. But, even more important, it is then continuously reproduced by capitalist production itself. This is possible through capitalist means of production. Though products of labour, machines and organisation belong to capital—both juridically and (crucially, as will be argued) from the point of view of the knowledge they incorporate. This implies that to survive (to reproduce their working capacity), workers depend on productive forces from which they are divorced and cannot control without capital. It is capitalist production itself that reproduces, in this way, the separation of labour from the conditions necessary for its existence. The passage from simple circulation to accumulation can thus be understood, reading Marx, as a social process of separation turned into a system through the technical appropriation, by capital, of the instruments of labour.

In the following two sections, a similar approach will be applied to the production of knowledge. The idea will be tested that also in the realm of knowledge the transition from simple to accumulative circulation can be seen as a historical rupture (not a mere evolution) systematically reproduced, both socially and technically, by the process of accumulation itself.

5. The ‘philosophical’ origins of knowledge accumulation 

Why did dialogues on abstract matters such as the nature of justice, love, numbers and ideas first originate in ancient Greece? Why not elsewhere? What made Greek city-states so different from other ancient civilisations? To the point of being able to host rational discourses aimed not at immediate practical gains but capable of systematically generating new ideas? This is the vexed question of the conditions which led to the development of rational, or 'philosophical', thought in Greece. It is also - I suggest - an investigation into the origins - both social and technical - of knowledge accumulation. (15)

The answer tested here is that the first philosophers were able to benefit from a unique form of separation from material needs. They also lived in a culture that saw the rise of the most powerful codification technique ever invented for natural languages: the alphabet. This highly improbable combination of two conditions - the first social, the second technical - occurred for the first time precisely in Greek city-states. This historical improbability should be taken seriously. Naive views that see the first philosophers as simply expanding wisdom imported from the East, or as isolated scholars who suddenly and miraculously started reflecting on the wonders of their inner and outer worlds are no longer convincing. (16) (Just as the idea that industrial capitalism stems from the logic of free trade, or the genius of individual entrepreneurs, is hardly plausible from a Marxian perspective). 

Origin of Alphabet, Phoenician; Image credit: Britannica
Origin of Alphabet, Phoenician; Image credit: Britannica

A few words on the unique situation of Greek city-states. The separation of intellectuals from material needs is symmetrically opposite to the forced separation, discussed above, of labourers expropriated from their means of subsistence. Here we have a privileged separation. Not labour power, but a class of free citizens living off the labour of peasents, foreigners, women and slaves. But what is peculiar to fifth century B.C. Greece [Ed.: Phoenecian] is that this privileged separation derived not only from direct domination through force, taxes and inheritance—as was the case in all ancient civilisations. The true novelty of the Greek city-states - Athens in particular - consisted of unprecedented wealth derived from a combination of commerce, minted money and a dominant position on the seas. Such was the distance of free citizens from the material concerns of life - even from the practical question of how to appropriate manual labour - that a purely social reality based on free communication, negotiation and trade emerged. (17) It is precisely this 'second nature' that saw the birth of democratic self-governement in Greece: an innovation that further contributed to separating the privileged few from the practical preoccupations of material reproduction. (18)

But this privileged separation from labour is only part of the story. It does not explain why speculations were not confined, as is often the case with non-productive (but rational, or even 'philosophical') uses of language, to the first quadrant. To continue the parallel with capitalist accumulation, the forced separation of workers from their livelihood is not sufficient alone to explain the origins of economic growth. Millennia of dispossessions, thefts and enslavements have passed without the results of these expropriations being systematically reinvested - as is typical of industrial capitalism - in the process of their generation. Following Marx, we can only understand capital by looking at the conditions that allow the surplus extracted from living labour to be preserved in an abstract form so as to be used again in successive expansion cycles. Similarly, if a non-productive cognitive activity is not only to exist but also to expand, it must be possible to conserve it on a technical level. Conversations in the here-and-now must be able to crystallise in a decontextualised form to be readily available for future interventions. And the citizens taking part in them must be freed not only from manual labour, but also from the burden of continuous oral repetitions. Yet another Greek innovation, which adds on to the commercial power of city-states and to democracy, made this possible. This was a means of codification: alphabetical writing. We should not underestimate the importance of writing in general and the power of the alphabet in particular. 

In oral cultures, there is a tendency for knowledge not to grow. Every innovation is, so to speak, ‘uneconomical’. Besides being produced, new ideas must also be recalled; they must therefore compete with the contents already present in social memory; and since the latter, in the absence of external aids (writing), coincides with individual memory, innovations soon run into natural limits. Though expandable through mnemotechnics, human memory is, inevitably, finite. Hence the static, conservative nature of oral cultures. And hence, also, the necessity for all members of an oral community to share, through continuous and collective repetitions, the entire 'encyclopaedia' of practical, ethical and religious knowledge necessary for social reproduction. (19)

By breaching the link between social and individual memory, writing overcomes this natural limit to expansion. It frees individual minds from the need to repeat identical content continually. It thus opens up the possibility - of course, only to classes separated from manual labour - of systematic cognitive innovation. (20) More. While the spoken word vanishes in the very act of pronouncing it, writing survives the act of its production; just like an artefact, it can be looked at from without, it can become a 'thing', an object of reflection; it can be transformed, in other words, into the product of intellectual labour. A product potentially within every reader's reach and therefore universal and expandable; a product no longer dependent on the times, places and contexts - always specific and non-repeatable - of life, action and use. 

Moreover, as a means of codification of the spoken word, alphabetic writing has an essential advantage over previous forms of writing. It does not function - as do hieroglyphs and ideograms - by visually reproducing the meaning of words. Instead, it goes all the way to represent the word itself or, more precisely, the sequence of sounds that make up its signifier. (21) A codification of no small importance. It provides access to the most intimate structure of natural languages: the possibility to generate a potentially infinite number of signifying units (morphemes) by combining a limited number of elements lacking meaning in themselves (phonemes). Language is the most economical, flexible and productive form of communication precisely because of this 'doubly articulated' (phonemes and morphemes) structure. (22) And the visual reproduction of this double articulation (a finite number of letters is combined to form a potentially infinite number of words) allows alphabetic writing to preserve, fixing it in space, this extraordinary potential of spoken language. And, in so doing, it enables writing to become more accessible, further reinforcing the Greek second nature made up of a purely social and communicative reality. (23)  

6. Intensification of knowledge accumulation: the scientific revolution

If the combined influence of separation and codification is plausible as an interpretation of the conditions that led to the origin of the cumulative mode of knowledge production, it becomes even more so when we consider the full-scale regime of knowledge accumulation inaugurated by modern science. With the seventeenth century scientific revolution, codification comes to affect not only the spoken word but experience itself; for this reason, the separation of intellectual labour comes to imply, as is taught by the Husserlian tradition, a separation even more radical than the privileged distance from the toil of material reproduction: that of intellectual experience from the ‘lifeworld’.(24)

At the origin of modern science is the unification of what, for Aristotelian physics, was strictly distinct: the incorruptible regularity of celestial spheres and the contingent disorder of the 'sublunar world'. The ancient cosmos, marked by this unbridgeable separation, is replaced by a homogeneous universe—a geometric space equally measurable and with no hierarchical or qualitative caesura. This unification was a genuinely disruptive conceptual revolution. Its consequences on knowledge production last to these days. 

Still from Schism music video, Tool; Image credit: You Tube.
Still from Schism music video, Tool; Image credit: You Tube.

It should be stressed that unlike the celestial spheres, it is not easy to come across order and regularity on planet Earth. Phenomena do not spontaneously lend themselves to being interpreted as particular instances of universal laws. If we were to rely exclusively on our senses, Aristotle's qualitative physics would be far more plausible than Galileo's mathematisation of nature. To be experienced, regularity and uniformity - both necessary conditions for mathematisation - must, on our planet, first be artificially constructed. Therefore, contrary to what is commonly believed, modern science's origin lies not in the valorisation of experience but rather in its systematic manipulation. (25) Science involves producing the desired observable phenomena. It involves interrogating nature, using a language aimed at obtaining precise and quantifiable answers. No longer philosophy's language of common sense; instead, a language artificially constructed elsewhere and subsequently forced upon nature. In Galileo's words, a 'mathematical language', a language made up of 'triangles, circles and other geometric figures’. (26)

Modern science is only conceivable when one-off, individual and contextual experiences can be subsumed under standard protocols; when the contents of experience can be transformed by making it exchangeable, comparable, measurable, controllable, and therefore expandable by a cosmopolitan community of specialists; in other words when the continuous, vague and ever-changing flux of everyday experience can be transformed into reproducible, identical and therefore verifiable and falsifiable, experiments. This is possible only when a preliminary codification takes place. That is: when knowledge is produced by making use of standard means of manipulation, both conceptual (mathematical, logical, geometrical…) and material (mechanical, optical, magnetic, electrical, acoustic…). It is only thanks to this forcing upon experience of codes - and to the precision, the order and the regularity that go with them (27) - that modern science can produce data capable of acting not only as points of arrival but also as starting points for new cycles of knowledge expansion.

Here again, we have a parallel with the capitalist mode of production, which also experiences a powerful intensification process at a particular stage of its development. Marx calls it the 'real subsumption of labour under capital', opposing it to the merely formal subsumption that occurs when working processes, though finalised to capitalist growth, undergo no alteration in their contents and methods. en attendant que la liste me soit communiquée, voici le courriel reçu par un étudiant concernant sa demande de visa. (28) With real subsumption, which Marx associates, in Capital, with the advent of machinery and large scale-industry, the working processes themselves - no longer only their aim - have the effect of concretely enhancing accumulation. Capital enters directly into their technical and organisational configurations; it forces times and methods upon them through the codes incorporated in systems of machines and organisational procedures. 

Likewise with modern science. Through the knowledge codified in its instruments and findings, the very contents of intellectual activity is transformed. Like labour processes, real-life experiences - the raw material of scientific research - come to lose their singularity; they become reproducible; they get channelled into alternatives predetermined by 'universal and necessary' categories; they pass, we could say, from being formally to being really subsumed under the impersonal imperative of self-referred and cumulative production of decontextualised knowledge. 

The separation of intellectuals remains, of course, privileged because it is based, as in the past, on the productive labour performed by working classes elsewhere. However, while for the first philosophers, the main pre-requisite to take part in knowledge production - class privilege apart - was simply the ability to read and write, the dependence on external means of codification becomes crucial with modern science. The hierarchical organisation of scientific research comes to assume, with academies and universities, increasing relevance. The competencies of scientists come to distance themselves from the objects of their research, focusing more on the processes and technologies involved in their investigations. The stock of past knowledge crystallised in scientific instruments and theories (the paradigms of Kuhn's 'normal science’) (29) make the scope for genuinely disruptive individual interventions ever narrower. As past codified knowledge becomes ever more relevant, the emergence of new figures the likes of Galileo, capable of producing the instruments of their own innovations, becomes less and less probable.

7. Capital and knowledge

With real subsumption, capitalist production is no longer restricted to the techniques and processes it inherits from the past. Instead, it begins to impose new techniques itself, disengaging from the specificities of local contexts and replacing workers' skills and tacit knowledge - previously controlled by workers (or by guilds, or tradition) - with new practical knowledge encoded in machinery and impersonal procedures. 

This is not to say contextual knowledge is no longer required of workers in the workplace. However, this knowledge comes to depend (not unlike the real subsumption of intellectual experience under modern science's means of codification) on the knowledge already crystallised in the factory hardware and software. It would not exist independently of it. As subsumption becomes real, workers lose visibility of the relationship between their labour and the objects they produce. Their practical knowledge is confined to the outermost interfaces of their instruments of labour. In contrast, the most relevant knowledge - the knowledge governing times, methods, pauses and quantities - is incorporated into the predefined codes operating in capital's machinery and organisation. 

The cognitive circle typical of industrial capitalism can thus be described as the replacement and regeneration of contextual knowledge through the introduction of new codified knowledge. On the one hand, pre-existing contextual knowledge (controlled by labour) is replaced by new codified knowledge (controlled by capital); this substitution, in turn, causes the generation of new contextual knowledge (controlled by labour) necessary to use - and therefore dependent upon - the new codified knowledge introduced by capital. It is a cycle, as can be seen, governed by codified knowledge. Capitalist means of production determine both the contents of the workers' knowledge to be replaced, and the new skills that labour will be required to master (always temporarily, pending the next replacement cycle) to operate with new machinery in the specificity of a production context. 

It should be stressed that the repetition of this cycle - the continuous replacement and regeneration of contextual knowledge through the incessant revolution of the means of production - is a specific trait of industrial capitalism. Unlike previous modes of production, under capitalism the appropriation of a surplus by the dominant classes no longer occurs - only - through force, taxation, or religious symbolism. The specific mode of capitalist appropriation (which by no means implies that other forms of dispossession do not continue to exist and flourish under capitalism) (30) consists, rather, in the technical control of the labour process; in other words, in the control of the knowledge involved in the various phases of the transformation of input into output. This is because so long as the practical knowledge concerning production belongs to workers, expropriation may well occur, but only after production has taken place; that is, workers may well be expropriated of the products of their labour, but not of labour itself. Only by directly controlling labour processes (how much is produced, at what pace, with the consumption of how much raw materials and energy and so on) can the dominant classes get richer through (not after) production. By controlling these processes, and the knowledge they incorporate, it is, in fact, possible for capital to control not only the products but also the time it takes on average (or ‘socially’, in Marx’s terms) to produce them. And by shortening this time - while keeping unchanged (or increasing) the number of hours worked per day - it is possible for capital to appropriate a quantity of labour greater than the quantity necessary to reintegrate wages—a surplus-labour, in Marxian terms, to reinvest (having realised its value on the market) in the following cycles of production. 

Codification of knowledge is crucial in this respect. To control production times (in order to speed them up), it is not sufficient to possess the means of production. It is also necessary, and of the utmost importance, to possess the knowledge required to produce and innovate them. This takes us back to the separation of labour from the conditions necessary for its reproduction. We can now add that starting from industrial capitalism (well before the ‘knowledge economy’ and the current hype on AI) the separation of labour invests not only the ownership of the means of production but also, and to the fullest extent, the knowledge (machinery and organisation) incorporated by capital and therefore expropriated, precisely because of this incorporation, from direct producers. (31)

But key to an understanding of knowledge production under capitalism in the third quadrant is also the recognition of its necessary reliance upon external sources; that is, upon codified knowledge originating from breakthroughs in basic scientific research or, more simply, from the work of managers and engineers taking place before, and outside of, the value-producing process. This means that, unlike science, technology is not able (at least not before the microelectronic revolution, which will now lead us to the fourth quadrant) to feed on its results; its input - scientific models and abstract theories - is of a ‘substance’ different from its output: the codification of concrete segments of material production. Within the third quadrant, it is not possible, in short, to take the outcomes of innovations as inputs for more innovations and to thus give rise to a system of self-centred cumulative growth. Such a system can be triggered, we will now see, in the fourth quadrant, with cognitive capital and artificial intelligence. 

8. Cognitive Capital

The self-referential accumulation of codified knowledge within the context of capitalist production has been made possible by the ‘information revolution’. This revolution, which gathered pace in the final third of the twentieth century, has two sides, one social and one technical. The social side consists in the formation of a ‘knowledge’ labour-power separated, just like manual labour - and in contrast to the privileged separation of second quadrant intellectuals - from the conditions necessary for its reproduction. The technical side consists in the autonomisation of software from hardware made possible by the microprocessor. (32) The outcome is a system where the output of one innovation can be taken - without exiting capitalist production - directly as an input for the next. This means that the cognitive cycle discussed above, typical of industrial capitalism - workers’ contextual knowledge is continually replaced and regenerated through the introduction of ever-new codified knowledge - can itself become part of the capitalist production system. This simultaneous accumulation of capital and knowledge - or, more precisely, of capital and of a form of knowledge that originates from capital’s internal drive to control knowledge - is what is here defined ‘cognitive capital’. 

With cognitive capital, one and the same relation - that of knowledge-workers to capital’s codified means of production - is thus at the heart of two distinct cumulative dynamics: capital and knowledge. In other words: the symbolic, communicative and cognitive activities carried out daily in the so-called ‘knowledge economy’ serve both to valorise and expand codified knowledge (software) and, at the same time, to produce a surplus for capital. As bearers of contextual knowledge, workers and consumers - the distinction between work and life becoming ever more blurred - contribute to the accumulation of knowledge; as living labour they also contribute, simultaneously, to the accumulation of capital. Knowledge workers are thus located at the point of intersection between two different self-expanding accumulation processes. Each dominated by an abstract pole (codified knowledge, exchange value); and each dependent, for its expansion (and therefore for its existence, since existence and expansion, in both cases, coincide) on a concrete relationship with human life and action. 

To be clear: while the two systems (accumulation of wealth and accumulation of knowledge) overlap, they do not also coincide. They have a vital activity in common: the appropriation of contextual knowledge. But their specific outcomes - abstract wealth and codified knowledge - are different in kind. Codified knowledge is not money, and is very difficult to appropriate. Unlike gold, or oil, it is ‘non-rivalrous’. The same code, for instance, can be (and most often is) used by multiple workers, consumers and machines. And despite intellectual property laws (licensing fees, patents, royalties and the like), it is in any case difficult to attribute economic value to knowledge in its abstract form. It is no chance the trading of codified knowledge is rare. (33) 

What is typically traded is useful knowledge. Knowledge applied to contexts, rather than to knowledge per se. The more evident the contextual component of knowledge, the easier it is for it to used—and hence for new commodities which are the outcome of knowledge accumulation to be devised. Knowledge and economic value typically make contact where commodities assume the appearance of being tailored to our very specific and unrepeatable (i.e., contextual) needs. Street navigation, targeted ads, medical diagnostics, face recognition, surveillance systems and so on all point to adaptations, to specific uses, to differences in space and time.

The two accumulation systems do not coincide, therefore, in their outputs but they do, however, act in unison. They use a shared resource (mental labour) to produce different outcomes (exchange value and code); and they make contact when the concrete and intermediate phases of their respective expansion cycles (use values and contextual knowledge) latch on to each other; in other words, when knowledge gains use value because of its contextual form. 

In this sense, cognitive capital is a symbiotic relationship between two systems - two ‘self-expanding spirals’ - which reinforce each other in the pursuit of their respective impersonal goals. They reinforce each other because, as argued above, codified knowledge lies at the heart of capital’s control over working processes (and can thus be used for the typically capitalist objective of creating new cycles of worker control and displacement); also, because for capital to continually expand into new information-rich markets with an apparently ever-increasing range, diversity and personalisation of consumer choices, an accumulation of codified knowledge must be taking place in the background; and finally because the most effective way to convince the human bearers of contextual knowledge (workers and consumers) to submit to the never-ending disruptions of working practices and lifeforms is the forced separation - recursive and systematic under capitalism - of human life from the conditions necessary for its continued existence.

9. From Cognitive Capital to Artificial Intelligence

At the heart of cognitive capital lies the transformation of contextual into codified knowledge. It is here that labour encounters codified capital. Should this transformation, for any reason, run into setbacks or interruptions, so too would the production of surplus knowledge and value. A potential limit to the accumulation of cognitive capital must therefore lie in the amount of contextual knowledge available for codification. If there were no more contexts left to digitise, cognitive capital would cease to accumulate—and, therefore, to exist. 

There is a parallel argument for capital in general. If, as suggested by the Marxian perspective of this article, the source of value under capitalism ultimately resides in the encounter of capital with labour, then the amount of value that can be created at any given time must be limited by the amount of capital employable productively and by the size of the population available to be put to work. In Marx's terms, by 'the number of working days that can simultaneously be exploited’ or, more succinctly, by the 'total working day’. (34) For cognitive capital, the limit analogous to the 'total working day' is the amount of contextual knowledge available at any given time for codification. This is why cognitive capital is under pressure to continually find new areas of contextual knowledge to feed into its digitisation processes. This is why, from natural languages to emotional states, from sexual orientations to the recognition of voice, fingerprints, body language and eyeballs, we seem well on track to create digital replicas of everything human; and also to convert, through a proliferation of sensors and smart devices, all our natural and social environments into digital data. However, the fact remains: just as ever-new segments of the world population need to find their way into the wage-labour relation if capital's growth is to persevere (can we explain the global growth of recent decades without factoring in the uprooting of hundreds of millions of Chinese farmers from their land?), so cognitive capital needs to continually find new contexts. Can this carry on indefinitely? Or will codifiable contexts eventually dry up in a completely artificial world? 

There is a countervailing tendency to this 'running out of contexts’: every new codification cycle creates its own contexts. We have come across this when discussing the impact of literacy on the free citizens of Greek city-states (section 5), when looking into the effects of codification (for instance mathematics and measurement) on the contents of scientists’ intellectual activity (section 6) and when reflecting on the regeneration of contextual knowledge brought about by the introduction of new machinery on the factory floor (section 7). We can now generalise by saying that when used by humans, means of codification always bring with them the generation of new contexts and, therefore, new contextual knowledge. The boundary between codified and contextual knowledge is continually pushed, so to speak, one step forward. The pace with which, in our age, new code-dependent contexts are created (and replaced) is staggering. No sooner do we get used to one artificial environment (and build up experience, common sense and skills around it) a new one comes in to replace it. As a result, we live with a sense of constant precariousness. This acceleration in the production of new lifeworlds should be interpreted in light of the fact that for the first time in history, we are presented with two (no longer one) distinct, and in some points overlapping, impersonal imperatives to codify. With the consequence of increasing the production of codified knowledge by orders of magnitude (when compared to the second and third quadrants) and also, in the process, of generating an increasingly artificial environment (ever new code-dependent contexts—or 'metaverses', in contemporary parlance) to feed into the system. 

There is, however, another existential threat to cognitive capital and, possibly, to the capitalist mode of production tout-court. This threat comes from the codification process itself, which is under constant pressure (like all processes under capital) to reduce - possibly eliminate - the input of living contexts and labour. Which leads us to Artificial Intelligence (AI).

Today's AI results from the shift - made possible by the advent of graphic processing units (GPUs) and the development of neural networks - from expert systems to deep machine learning. While the former apply a number of rules listed by human programmers to real-life situations and examples, deep machine learning goes the opposite way, taking examples as starting points and then proceeding autonomously to create rules. This is a fundamental inversion. Programmers used to create software that was then applied to concrete cases; deep machine learning now takes concrete cases (examples, data) as starting points and then proceeds, through ‘neural networks’, to create software.

On the one hand, AI is just another form of fixed capital. Automating codification, AI lies at the heart of cognitive capital's quest to appropriate (codify) and regenerate new contexts. It can thus be seen as the form of fixed capital most adequate to cognitive capital. And its use value, as with all capitalist machines, can be fully comprehended only when considering its function for the reproduction of the system. In the case of cognitive capital, the function is both to control labour, reproducing its subordinate relation to capital and, simultaneously, to create forms of knowledge that can generate more knowledge. This double aim (accumulation of capital and knowledge) directly influences the content of the knowledge produced. First, AI’s outputs tend to reinforce the capitalist relations under which production takes place. If biased knowledge (e.g., relative to class, gender and race) is fed into the data involved in the automatic knowledge-creating process, such biases will tend to amplify after each expanding cycle. (35) Second, the automatic and expanding nature of the process tends to accentuate the subordination of qualitative considerations to the quantitative imperative of accumulation. This is strikingly evident in academia and research-related professions, where AI proves to be an ideal tool for subordinating decisions about what content ends up being produced (and funded) to the quantitative logic of citation numbers, university rankings, impact factors and the like: a recipe for transforming research into the solving of conundrums related to existing paradigms, rather than an intellectual activity aimed at disrupting paradigms—the latter, as we know, being the only path leading to genuine scientific breakthroughs. (36) 

There is another equally troubling aspect of AI, however. Deep learning emulates human learning by starting with concrete (contextual) examples and then finding ways to fit them into general patterns and underlying rules. But while examples used by humans are inextricably tied to the time, space and actions of living communities, there is nothing to stop the examples fed into neural networks - big data - from severing their bond to living contexts. Big data do, it is true, have contextual components indicating specific times, places, people, or objects, either directly or indirectly (through their meta-data); but unlike human examples, the data which are fed into machines need to undergo a preliminary abstraction process. They enter the system in the form of QR-codes, sensors and bots - not innocent recordings of life and action. And there is nothing to stop these data from being outcomes of previous cycles of artificial production of information through data. (37) Take translation software. It works by accessing myriad examples of language used in context. How many of these 'contextual' examples are already the result of artificial translators? Or take large language models (LLMs) like Chat GTP. They are trained on text taken from the web. This text may have been predominantly human-generated to start with, but this is no doubt changing very rapidly, as LLM’s contribute themselves, self-reflexively, to generate the text which will be available to train them. (38) The same logic can apply to the generation of sounds, images, videos, 3D models and so on.

In other words, the frontier we are now facing is an accumulation of knowledge based on examples that originate not only (and not so much) from the workplace or the realm of consumer behaviour but from knowledge accumulation itself. In this sense, AI is not only fixed capital. It is also a knowledge-creating system that goes beyond cognitive capital. An accumulation system - so it appears - capable of exponentially increasing knowledge while doing away with the defining relationship of cognitive capital: that of capital's codified means of production to living contexts. Or, seen from an economic perspective, a system apparently capable of self-valorisation without having to engage in the capital-labour relationship; a system capable of extracting economic value no longer through the exploitation of labour, but through a form of return based, like rent or interest, on pure ownership. (39) In short, we are no longer looking at the code-context-code expanding spiral which, as argued so far, is the condition behind cognitive capital—and knowledge accumulation tout court. Instead, we are facing the possibility of an accumulation proceeding from codified knowledge to (surplus) codified knowledge directly, thus doing away with the intermediate, contextual phase. 

By way of a provisional (and provocative) conclusion, I will push the analogy with Marx's critique of political economy one step further. While Marx identifies the 'total working day' as an unescapable limit to accumulation, he also provides a very sophisticated (though sadly incomplete) analysis of a form of capital that can go beyond this limit, generating money from money without engaging in the capital-labour relationship. This form of capital is interest-bearing capital, 'a mysterious and self-creating source … of its own increase … self-valorising value, money-breeding money’. (40) A form of capital behind institutions familiar in Marx's days and even more in ours: banking and credit, private and public debt, an array of financial products, highly complex derivatives and so on. It is important to stress that this form of capital is no epiphenomenon. Investments in machinery and infrastructure - at the heart of capitalist production - would not be possible without credit; nor would the equally crucial synchronisation of the very different circulation times of fixed and circulating capital. (41) In the same way, AI is also central to our era's simultaneous accumulation of knowledge and value—the intelligence it produces, as argued, is at the heart of cognitive capital's intensification of the appropriation and regeneration of new contexts.

However, both finance and AI can also be viewed from a different angle. Freed from the process of valorisation-through-production (having broken off all connection with the 'total working day'), interest-bearing capital is also endowed with an inescapable 'fictitious' nature. It can represent a claim on exploitation that has yet to take place. Once ways are found, with this fictitious capital - but also, if the analogy holds, with AI, or fictitious knowledge - to magically reach out and grab more value than can be produced, there is no limit to how many fictitious working days - or fictitious contexts - can be claimed. Indeed, once the possibility exists, the incentive to claim more value than is even remotely plausible (at given levels of productive capital and labour) becomes irresistible. 

And, as we know too well, there is also no limit to the violence of the crashes that ensue when the system eventually comes to terms with its limits. Or, possibly, there is no limit to the opportunities to imagine new social formations capable of going beyond built-in imperatives to compounding growth.



1. That is, until the first two-thirds of the 20th century; as will be discussed below, things started to change in the 1970s, with the introduction of the microprocessor and the emergence of a new 'mental' workforce. 

2. Alfred Sohn Rethel was the first author to reflect on this topic making use of Marxian categories, as will be done, from a different perspective, in this article (see Alfred Sohn Rethel, Intellectual and Manual Labour, Atlantic Highlands, New Jersey 1978). 

3. Karl Marx, Capital: A Critique of Political Economy. Volume 1, translated by Ben Fowkes, Penguin, Harmondsworth 1976, pp. 247 ff. and passim.  

4. In John R. Searle’s words, these rules are ‘constitutive: they do not merely regulate, they create or define new forms of behavior’. The rules of football or chess, for instance, (and the same applies to language) ‘do not merely regulate playing football of chess, but as it were they create the very possibility of playing such games’ (John R. Searle, Speech Acts. An Essay in the Philosophy of Language, Cambridge Univ. Press,  Cambridge 1969, p.33). 

5. Ludwig Wittgenstein, Philososphical Investigations, Basil Blackwell, Worcester 1953, § 83.

6. Universities and research institutions obviously depend on funding, which is generally contingent on political and economic goals (more on this later).  But, as specified above, we are discussing ideal types here, and to this end we make the naive assumption that basic research really is an end in itself. 

7. Lorenzo Cillario, ’Il capitalismo cognitivo. Sapere, sfruttamento e accumulazione dopo la rivoluzione informatica’, in Trasformazione e persistenza. Saggi sulla storicità del capitalismo, Franco Angeli, Milan 1990. See also Lorenzo Cillario, L’”uomo di vetro” nel lavoro organizzato, Editoriale Mongolfiera, Bologna 1990 and Lorenzo Cillario, Roberto Finelli (eds.), Capitalismo e Conoscenza, Manifestolibri, Roma 1998. Since its introduction, the term has subsequently been used in Italian and French autonomist Marxism literature mostly in conjunction with ‘post-workerist’ categories such as ‘immaterial labour’, ‘general intellect’, ‘biopower’ and ‘multitude’ (see, for instance, Yann Moulier-Boutang, Cognitive Capitalism, Polity, Cambridge 2008; Carlo Vercellone, ‘From Formal Subsumption to General Intellect: Elements for a Marxist Reading of the Thesis of Cognitive Capitalism’, Historical Materialism, vol. 15, no. 1, January 2007); these theoretical frameworks, inspired by the work of Toni Negri and other Italian ‘post-workerists’, do away with essential parts of Marx’s theory, including the labour theory of value. In contrast, the relation, under capital, of value to labour continues to play a central role in the perspective of this article, as it did in the works that first introduced the term. For an insightful critique of ‘cognitive capital’ as currently used, see George Caffentzis, In Letters of Blood and Fire: Work, Machines, and the Crisis of Capitalism, PM Press, New York 2013, pp. 95-123.

8. This distinction draws from Michael Polanyi's distinction between 'tacit' and 'explicit' knowledge (see Michael Polanyi, Personal Knowledge. Towards a Post-Critical Philosophy, The Univ. of Chicago Press, Chicago 1958), which was introduced into economics literature by Richard R. Nelson and Sidney G. Winter in An Evolutionary Theory of Economic Change, Belknap, Cambridge M.A. 1982. To my knowledge, the terms ‘codified’ and ‘contextual’ were first used to refer to this distinction in economics literature in Barbara di Bernardo, Enzo Rullani, Il management e le macchine. Teoria evolutiva dell’impresa, Il Mulino, Bologna 1990.

9. The concept of 'externalisation' figures prominently in the works of André Leroi-Gourain. In Gesture and Speech, The MIT Press, Cambridge MA 1993, the author posits that tools and objects form an impersonal and collective memory that is distinct and 'external' with respect both to the individually acquired (epigenetic) memories of our brain and to the biological (phylogenetic) memory we inherit from our ancestors. This 'third kind' of memory is a cornerstone of Leroi-Gourain's reflections on hominisation. It is also the starting point of many of Bernard Stiegler's reflections on technics (for instance, Technics and Time, 1: The Fault of Epimetheus, Stanford Univ. Press, Stanford 1998). The definition of 'codification' proposed here also bears a strong resemblance to Sylvain Auroux's concept of grammatisation (Sylvain Auroux, La révolution technologique de la grammatisation, Mardaga, Liège 1994), which was also taken up by Bernard Stiegler in the context of his reflections on the 'correlationist models' of the automated processing of big data (see Automatic Society, Volume 1: The Future of Work, Polity, Cambridge 2016). The arguments proposed in the following pages resonate with numerous assumptions and conclusions of Stiegler's prolific work.

10. This is not to imply writing originates as a codification of spoken language. On the contrary, all archaeological evidence points to the fact that the first forms of writing were used to codify entirely different kinds of meaningful actions: from counting and bookkeeping to keeping track of ancestry and honouring the dead, from predicting the future to identifying the best seasons for planting crops and waging wars. More than two millennia were to pass before the alphabet - the most efficient means for codifying spoken language - was devised. And even in its alphabetical form, writing codifies only a fraction of the complexities involved in spoken language. 

11. Should it lose all connection to a living environment, codified knowledge would not even be, strictly speaking, knowledge—if we accept a broad definition of ‘knowledge’ as a form of conservation of what originates (often passing through incredibly complex mediations) from adaptive behaviour (for an interpretation in these lines, see Henry Plotkin, Darwin Machines and the Nature of Knowledge, Penguin, London 1994). Further on (Section 9), knowledge with no connection to contexts will be defined as ‘fictitious’, in analogy to Marx’s ‘fictitious capital’. 

12. Ludwig Wittgenstein, Philosophical Investigations, cit. 

13. On the impossibility of circulation to consist in itself and on the consequent necessity to move from the analytical approach of classical and neoclassical economics to the synthetic and dialectical perspective of Marx’s Grundrisse and Capital, I rely on Roberto Finelli's groundbreaking work, starting from Astrazione e dialettica, dal romanticismo al capitalismo, Bulzoni, Rome 1987. See, in English, Roberto Bellofiore and Roberto Finelli, ’Capital, Labour and Time: The marxian monetary labour theory of value as a theory of exploitation’, in Bellofiore, ed., Marxian Economics. A Reappraisal, vol. 1, Macmillan, London 1998; Roberto Finelli, ‘Abstraction versus Contradictions: Observations on Chris Arthur’s ‘The New Dialectic and Marx’s Capital’’, in Historical Materalism, vol. 15, 2  2007. 

14. Kark Marx, Capital, cit. pp. 873-940.

15. This is not to imply all ancient philosophy saw a process of cumulative growth; or that cumulative growth of knowledge originates only in the intellectual activity of the first philosophers. As stated above, the actual production of knowledge is across quadrants. Rather, the argument is that the Greek milieu saw for the first time conditions favourable for such a production; and that in some cases (we can think of Empedocle’s elements, Anaxagoras’ infinite parts and Democritus’ atoms) innovative theories were built one upon another in response to previous, sometimes incompatible views (e.g., Parmenides and Heraclitus on change and motion). Nor should it be implied that philosophy in general has a vocation for cumulative growth: the opposite can be argued, especially since the ‘repositioning’ of philosophy vis-à-vis the natural sciences starting from Kant. (The case of analytic philosophy, where theories do indeed seem to build one upon another, can be read as a confirmation a contrario, given the inspiration this tradition draws from scientific research). 

16. At least since the investigations carried out by Francis M. Cornford and his Cambridge colleagues at the beginning of the twentieth century (See Francis M. Cornford, From Religion to Philosophy. A Study in the Origins of Western Speculation, Edward Arnold, London 1912).

17. In his groundbreaking Intellectual and Manual Labour, cit., Sohn Rethel argues that the shaping of abstract philosophical thought was influenced, since its origins, by the commodity form and money, as well as by a rigid division between intellectual and manual labour. The outlook proposed here goes in a different direction. Greater emphasis is placed on the cumulative mode of knowledge production and, as we shall now see, on the role of alphabetical writing (a Greek innovation) in getting it started. Also, for the interpretation of Marx proposed here (see note 13 above) it is problematic to extend (as Sohn Rethel does) categories such as the commodity form to pre-capitalist social formations. Finally, the author's founding distinction between intellectual and manual labour seems to lose much of its plausibility in the present era, when mental labour is subsumed under capital no less than manual labour.

18. See  Geoffrey E.R. Lloyd, Magic, Reason and Experience. Studies in the Origin and Development of Greek Science, Cambridge Univ. Press, Cambridge 1979; Jean Pierre Vernant, The Origins of Greek Thought, Cornell Univ. Press, New York 1984. 

19. Homeric epics, for instance, can be interpreted as an oral culture’s encyclopaedia. See Milman Parry, The Making of Homeric Verse, Claredon Press, Oxford 1971. 

20. See Walter Ong, Orality and Literacy: The Technologizing of the Word, Methuen, London 1982. 

21. For a profound reflection on the significance of alphabetic writing with respect to previous writing systems, see Clarisse Herrenschmidt, ‘Writing between Visible and Invisible Worlds in Iran, Israel, and Greece’, in Bottero et al., Ancestor of the West. Writing, Reasoning and Religion in Mesopotamia, Elam, and Greece, The University of Chicago Press, Chicago 2000. 

22. The concept of ‘double articulation’ was first introduced in linguistics by André Martinet in Syntaxe générale, Armand Colin Editeur, Paris 1985. 

23. Eric Havelock wrote memorable pages on the influence of the written word on the origins of philosophical thought (Eric Havelock, Preface to Plato, Harvard Univ. Press, Cambridge MA 1963). In his view, the profound reason for the condemnation of poetry in Republic despite, paradoxically, Socrates’s objections to writing in Phaedrus, and despite the Platonic style, dialogic and therefore close to the spoken word, is Plato’s de facto endorsement of literacy. 

24. See Edmund Husserl, The Crisis of European Sciences and Transcendental Phenomenology, Northwestern Univ. Press, Evanston 1959. 

25. See Alexandre Koiré, Etudes d’histoire da la pensée Philosophique, Max Leclerc et C., Paris 1961. 

26. English translation from Galileo Galilei, ‘Il Saggiatore’, Opere, Vol. 1, Utet, Torino 1964, p. 632.

27. See M. Nortion Wise, ed., The values of precision, Princeton Univ. Press, Princeton 1995. 

28. Karl Marx, Capital, cit., pp. 1019-38.

29. See Thomas Kuhn, The Structure of Scientific Revolutions, University of Chicago Press, Chicago 1962. 

30. The capture and transfer of surplus through the use of extra-economic and political means and the depredation of resources from less capitalistically developed countries - 'Accumulation by dispossession', in David Harvey's terms (The New Imperialism, Oxford Univ. Press, Oxford 2003) - is a constant feature of capitalist development, and is hugely relevant to this day.

31. For an invaluable reconstruction of the roots of the algorithmic thinking leading to AI in the surveillance and control of labour processes, see Matteo Pasquinelli, The Eye of the Master. A Social History of Artificial Intelligence, Verso, London 2023. 

32. The theoretical basis for the software/hardware distinction can be traced back to Alan Turing's work in the 1930s (see his ‘On computable numbers, with an application to the Entsheidungsproblem’, in Proceedings of the London Mathematical Society, 42 1936, pp. 230-265). But it is only in the 1970s, thanks to the microprocessor, that this distinction started to impact economic production concretely.

33. Even data (which, despite their structured form, have indexical components pointing to times, places, persons and objects) are not commonly sold. It is often easier to trade entire companies and start-ups, with knowledge valorisation processes already in place, than the data they produce.

34. Karl Marx, Capital: A critique of Political Economy, Volume 3, translated by David Fernbach, Penguin, London 1981, p. 523.

35. See, for instance: Shahriar Akter et al., ‘Algorithmic bias in data-driven innovation in the age of AI’, International Journal of Information Management, Oct. 2021; Ting-An Lin and Po-Hsuan Cameron Chen, ‘Artificial Intelligence in a Structurally Unjust Society’, Feminist Philosophy Quarterly, Vol. 8, 3/4 2022.

36. This tendency is accentuated by the short-termism (not too distant, it could be argued, from the tendency in financial markets to favour immediate profits over medium-term returns) that the use of automated algorithms accentuates when applied to the quantitative logic of citations. In Giuseppe Longo’s words (‘Letter to Alan Turing’, Theory, Culture & Society, Special issue on Transversal Posthumanities 2018) "Inventions like yours [Turing’s] took ten, twenty, thirty years to be appreciated: the impact factor of journals is now computed by machines based on the number of citations of articles following the two years following publication. In mathematics and in physics…it takes ten years just to understand a difficult result to problems that have been open for decades. Networked machines that compute the number of citations kill from the outset any attempt to venture as you [Turing] did, onto paths that are entirely new”. See also Giuseppe Longo, ‘Science, problem solving and bibliometrics’, Invited Lecture, Academia Europaea Conference on ‘Use and abuse of bibliometrics’, Stockholm, May 2013. Also on this topic: Yves Gendron, Jane Andrew, Christine Cooper, ‘The perils of artificial intelligence in academic publishing’, Critical Perspectives on Accounting 2021.

37. See Frederic Kaplan, ‘Linguistic Capitalism and Algorithmic Mediation’, Representations, vol. 127, 1 2014. Taking the idea of a wholly automatised production of knowledge to the extreme, we would live in a completely synchronised world where the concept of new knowledge would no longer make sense: every predicate would already be contained in its subject; every contingent action would already have been predicted by algorithms. For an investigation into the consequences on the ‘public sphere’ that go with algorithmic reason, see, in this direction of thought, Anish Mohammed and Shaj Mohan, ‘Principle of Sufficient Reason 2.0: On Information Metaphysics’, in Divya Dwivedi, Sanil V (edd.), The Public Sphere from Outside the West, Bloomsbury Academic, London 2015, pp. 240-256. 

38. As is clealy explained in Ilia Shumailov et al., ‘The Curse of Recursion: Training on Generated Data Makes Models Forget’, arXiv, 27 May 2023 (, we are facing the very concrete risk of ‘a degenerative process affecting generations of learned generative models, where generated data end up polluting the training set of the next generation of models’. The authors conclude ‘it may become increasingly difficult to train newer versions of LLMs without access to data that was crawled from the Internet prior to the mass adoption of the technology, or direct access to data generated by humans at scale.’

39. In this sense, AI fits into the general tendency of rich countries to morph into ‘rentier economies’ benefiting from surplus value produced in the developing world. See David Harvey, The New Imperialism, cit.; Ugo Pagano, ‘The Crisis of Intellectual Monopoly Capitalism’, Cambridge Journal of Economics, vol. 38 2014; Brett Christophers, Rentier Capitalism, Who Owns the Economy and Who Pays for It?, London, Verso 2020. So deep is the transformation that the global economy is undergoing as a consequence of a very limited number of platforms monopolising the knowledge necessary to artificially accumulate more knowledge that some Marxist authors question whether we are still, strictly speaking, in a capitalist - and not a (techno) feudal - regime of accumulation. See in particular Cédric Durand (Technoféodalisme. Critique de l’économie numérique, La Découverte, Paris 2020) and Yanis Varoufakis (Technofeudalism, What Killed Capitalism, Bodley Head, London 2023); for a general overview of the debate, see Evgeny Morozov, ‘Critique of Techno-feudal reason’, New Left Review, 133/134, 2022. 

40. Karl Marx, Capital Volume 3, cit. p. 516.

41. See Suzanne de Brunhoff, Marx on Money, translated by Maurice J. Goldbloom, Urizen Books, New York 1976; David Harvey, The Limits to Capital, Basil Blackwell, Oxford 1982; Cédric Durand, Fictitious Capital: How Finance is Appropriating Our Future, translated by David Broder, Verso, London 2017.

Related Articles

The Political Uses of Literature


Philosophie quantique : Le monde est-il extérieur ?


bottom of page