Will the revolution be brought to us by means of statistical patterning backed by massive computational power, and fuelled by the accelerated enclosure of energy, minerals and water commons? Can it ride the coattails of a tech ideology of measurable ‘intelligence’ that comes straight out of the eugenics movement? Can leftist AI innovations circumvent the apparent lethargy of human analysis and turbo-charge liberation by means of the machine process? Recent experiments in machine learning on the left seem to be based on these hopes.

The Tricontinental has recently unveiled a tool which promises to resolve a problem they have apparently been dwelling on: the problem that the “Marxist corpus resists generic AI” because of its size, complexity and need for synthetic reading across texts. The prescribed response to this diagnosed problem is the development of an AI tool which can absorb the entire canon of Marxist thought and read across this dialectically so that we no longer have to do so. Added to myriad other experiments to date in developing ‘feminist’ and ‘decolonial’ AI tools, this is the latest foray of the left into short-circuiting the practice of collective learning and collaborative interpretation that has made Marx’s work relevant to a vast range of social movements and political projects. Can we really forsake the friction of dialectical reading and hope instead that a machine-led interpretation of liberation texts will deepen collective revolutionary consciousness?

In an essay for this publication, William Foley argues that we can – to him, the issue with AI is the question of who owns it. Writing in the vein of what Alex Hanna and Emily Bender might call ‘AI Hype’, Foley marvels that “AI is solving decades-old maths problems; it’s winning prestigious literary awards; it’s producing poetry that is rated more beautiful and more authentically human than actual human productions.” Is it though? 

Anthropic – the makers of Foley’s large language model (LLM) of choice – announced on 23rd September 2026 that they had made what they called the first AI-driven advance in genetics. Within days, it emerged that it was not Anthropic’s AI that had made this advance but a Copenhagen scientist, Mario Rodríguez Mestre, who had been using Anthropic’s Claude to analyse the genetic data he had been studying for years. Anthropic seemingly cut out the process of human judgement that has kept Mestre and his lab from publishing their discovery without further study, in order to claim the breakthrough for AI. 

The dream of a total archive

“Move fast and break things” is really a recipe for the theft of labour, rather than meaningful science or art. There is little evidence that so-called AI is capable of the kind of miracles that cause its proponents to swoon. As scholars like Timnit Gebru have long argued, even the hysteria that holds that the so-called ‘good guys’ must get hold of AI, because it can break into all our systems and kill us all, mistakes poor coding for extraordinary power.

This theft has not incidentally come along with the resurgence of race science and eugenics. The technologies we now call AI emerged out of the statistical projects that, in the 19th and early 20th centuries, fed scientific racism. ‘Intelligence’, to this logic, is singular, measurable and comparable. Indeed, the fantasy of knowledge and progress that underlies AI is the dream of a total archive: the idea that if we feed all human knowledge into a machine, it can simply pattern it all and make sense of it.

A whole world, mapped, known, mastered. That this is a colonial fantasy is obvious, but bears repeating. In it, there is no embodied knowing, no intuition, no sensorium, no relationality. The art and poetry that large language models generate are statistical normals rendered in literature and art: not communicative projects that feel or seek to remake the world. To stare in awe at this is to mistake, as Olivia Guest argues, correlation for cognition.

The global meme machine

Whilst experiments in AI are always framed and justified in terms of the urgency of the liberation needs of some collectives, they are never placed in the broader context of the capitalist and colonial relations which shape AI’s role in accelerating accumulation, expropriation and contamination in the world system. 

Absent, in other words, is any reflection on the material conditions out of which so-called ‘artificial intelligence’ has emerged, and which it sustains and reproduces, regardless of whether its content is Marx’s corpus or not, and regardless of who owns it. The phenomenon branded by venture capitalists as ‘artificial intelligence’ should really be understood as material statistical patterning, dependent on an intensely resource-hungry machine process, and on expropriated human labour. The data centres powering leftist forays into AI require the expansion of mineral extraction, a massive increase in energy demand which is amplifying fossil fuel burning and carbon emissions, as well as freshwater consumption to cool servers, all of which is leaving communities and ecologies in conditions of scarcity. The build out of these material-hungry data centres may well be the most rapid industrial expansion in history, with socioecologies in their surrounds paying the price in terms of outflows of biophysical value to the global meme machines.

Capitalism is a system of accumulation which determines biophysical flows of water, energy, minerals and so on. Under the capitalist mode of production throughout history, this material and energy throughput has been accelerated by particular labour and technological ‘innovations’ – from the intensification of workflow and labour exploitation under Taylorism, to the hydrocarbon-fuelled machine processes which have turbo-charged capitalist metabolism. The machine process enables not only accelerated material and energy throughput, but also intensified production of wastes, the release of toxins and contamination of the socioecologies at the heart of planetary life. Pro-AI screeds on the left ring hollow because they fail to recognise the inbuilt condition of planetary entropy that data centres are accelerating.

As Marxists, we instead ask why one community’s AI liberation should come at the expense of the depletion of another’s resources, and the expropriation of their land and labour? Why should we accede to a paradigm that mistakes statistical normalisation for a praxis of political reading and collective deliberation and calls it ‘intelligence’?

Finally, returning to the Tricontinental’s RedMarx, to read Marx through the machine is particularly ironic, in part because it accepts alienation from the text itself and cedes to the mystification of AI whilst ignoring the material conditions of large language models. It also disavows a history of dialectical reading, not as an isolated and individualised project, but as collective praxis.

We read Marx’s work together in classrooms, in the unions and in our movements because the friction of working through it allows us to collectively analyse the actual material conditions of the present and develop a revolutionary consciousness for our ecologically degraded times.


Sneha Krishnan is Associate Professor in Human Geography at the University of Oxford and a Tutorial Fellow at Brasenose College. She researches how geographies of caste and colonialism shape experiences of girlhood, and is an editor of Gender, Place & Culture.

Lisa Tilley teaches Political Ecology at SOAS and researches capitalist metabolism and anticapitalist organising in the world system. Her book 'Liberation Ecology' is forthcoming with Pluto.

The link has been copied!