7332639 Inhuman Power Artificial Intelligence and the Future of Capitalism Nick Dyer Witheford, Atle Mikkola Kjøsen, James Steinhoff

Nick Dyer-Witheford, Atle Mikkola Kjosen and James jpeg) Artificial Intelligence and the Future of Capitalism Inhuman Power Digital Barricades: Interventions in Digital Culture and Politics Series editors: Professor Jodi Dean, Hobart and William Smith Colleges Dr Joss Hands, Newcastle University Professor Tim Jordan, University of Sussex Also available: Shooting a Revolution: Visual Media and Warfare in Syria Donatella Della Ratta Cyber-Proletariat: Global Labour in the Digital Vortex Nick Dyer-Witheford The Digital Party: Political Organisation and Online Democracy Paolo Gerbaudo Gadget Consciousness: Collective Thought, Will and Action in the Age of Social Media Joss Hands Information Politics: Liberation and Exploitation in the Digital Society Tim Jordan Sad by Design: On Platform Nihilism Geert Lovink Unreal Objects: Digital Materialities, Technoscientific Projects and Political Realities Kate O’Riordan

Artificial Intelligence and the Future of Capitalism Nick Dyer-Witheford, Atle Mikkola Kjøsen and James Steinhoff PLUTO PRESS First published 2019 by Pluto Press 345 Archway Road, London N6 5AA com Copyright © Nick Dyer-Witheford, Atle Mikkola Kjøsen and James Steinhoff 2019 The right of Nick Dyer-Witheford, Atle Mikkola Kjøsen and James Steinhoff to be identified as the authors of this work has been asserted by them in accordance with the Copyright, Designs and Patents Act 1988. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN 978 0 7453 3861 3 Hardback ISBN 978 0 7453 3860 6 Paperback ISBN 978 1 7868 0395 5 PDF eBook ISBN 978 1 7868 0397 9 Kindle eBook ISBN 978 1 7868 0396 2 EPUB eBook

This book is printed on paper suitable for recycling and made from fully managed and sustained forest sources. Logging, pulping and manufacturing processes are expected to conform to the environmental standards of the country of origin. Typeset by Stanford DTP Services, Northampton, England Series Preface vi Acknowledgements vii Introduction: AI-Capital 1 1 Means of Cognition 30 2 Automating the Social Factory 68 3 Perfect Machines, Inhuman Labour 110 Conclusion: Communist AI 145 Notes 163 Bibliography 168 Index 200 Series Preface Crisis and conflict open up opportunities for liberation. In the early twenty-first century, these moments are marked by struggles enacted over and across the boundaries of the virtual, the digital, the actual and the real.

Digital cultures and politics connect people even as they simultaneously place them under surveillance and allow their lives to be mined for advertising. This series aims to intervene in such cultural and political conjunctures. It features critical explorations of the new terrains and practices of resistance, producing critical and informed explorations of the possibilities for revolt and liberation. Emerging research on digital cultures and politics investigates the effects of the widespread digitisation of increasing numbers of cultural objects, the new channels of communication swirling around us and the changing means of producing, remixing and distributing digital objects.

This research tends to oscillate between agendas of hope, that make remarkable claims for increased participation, and agendas of fear, that assume expanded repression and commodification. To avoid the opposites of hope and fear, the books in this series aggregate around the idea of the barricade. As sources of enclosure as well as defences for liberated space, barricades are erected where struggles are fierce and the stakes are high. They are necessarily partisan divides, different politici- zations and deployments of a common surface. In this sense, new media objects, their networked circuits and settings, as well as their material, informational and biological carriers all act as digital barricades.

Jodi Dean, Joss Hands and Tim Jordan Acknowledgements The authors thank each other for a collegial and comradely collab- oration. We thank the organizers of the ‘Das Kapital at 150: Marx’s Critique of Political Economy and the Global Crisis Today’ conference at Hofstra University in April 2017, where a panel presentation gave us our first opportunity to articulate some of these ideas. We also thank the organizers of the conference ‘Log Out! Worker Resistance Within and Against the Platform Economy’, held on 6 March 2018 at the University of Toronto, which inspired much of the ‘Heptagon of Struggles’ section of Chapter 2.

We thank David Castle for so immediately, and then patiently, supporting publication of this book, and the staff at Pluto for their work on it. Nick Dyer-Witheford thanks his dear wife Anne for countenanc- ing and enlivening the writing of another book. Atle Mikkola Kjøsen thanks his lovely partner Siobhan for all her support and companion- ship throughout the writing process. James Steinhoff thanks Marcello Guarini, Stephen Pender, Jeff Noonan, Deborah Cook, Philip Rose and Chris Tindale for helping him grow a brain.. Introduction: AI-Capital THE MOST VALUABLE THING Capitalism is today possessed by the Artificial Intelligence (AI) question.

Consider the Vancouver start-up Sanctuary Cognitive Systems Corpora- tion, which aims to develop ‘humanoid robots that can move, speak and think for themselves and interact – as intellectual peers – with real people’. Its owner, Geordie Rose, a quantum computing pioneer, concedes to an interviewer that, while building blocks towards this goal are already in common use in multiplying types of ‘narrow AI’ and specialized robotics applications, none are remotely close to an ‘Artificial General Intelli- gence’ (AGI) capable of full human emulation. His company’s mission ‘to unlock how human intelligence works and to replicate it on a mass scale’ therefore ‘sounds like a mind-boggling moonshot’.

However, Rose is undeterred, for if he succeeds, it will be ‘the most valuable thing ever created. What we’re talking about is fundamentally altering the basis of capitalism itself ’ (Silicoff 2018). It would be easy to dismiss this quest, were Sanctuary not competing against some of the most powerful capitalists in the world, also striving to produce AGI and related technologies: Elon Musk and his non-profit OpenAI; Vicarious FPC,, backed by Samsung and technology bil- lionaires Mark Zuckerberg and Jeff Bezos; and DeepMind Technologies, acquired by Google/Alphabet, whose owners, Sergei Brin and Larry Page, are patrons of the transhumanist Raymond Kurzweil (2005a), the most famous prophet of a ‘technological singularity’ in which computers attain human-equivalent intelligence.

Sanctuary Cognitive Systems Corp may or may not survive (it is clearly cash-strapped and looking for angel investors). But its story of soaring technological ambition, mission-driven digital entrepreneurship and creepy androids ‘like underworld creatures from a Hieronymus Bosch painting’ (Silicoff 2018) is symptomatic of the AI-fever sweeping the world market, a fever that also manifests in a burgeoning business literature on AI applications, torrents of con- flicting predictions about AI’s consequences for employment, utopian speculation on the creation of ‘Life 3.0’ (Tegmark 2017), and fictions 2. inhuman power ranging from pulp robo-apocalypses (Wilson 2012, 2015) to complex literary explorations on the new techno-existential horizon posited by AI (Mason 2017).

Defining AI is difficult. Nonetheless, Rose is correct that what is often termed ‘narrow AI’ is already present in the algorithmic processes that now inform much of everyday life. For warehouse workers or military personnel, such AIs may incarnate in the chassis of a robot delivery vehicle or semi-autonomous killer drone. Most AIs, however, act invisibly in the background of activities conducted on smartphones and computers; in search engine results, social media feeds, video games and targeted advertisements; in the acceptance or rejection of applications for bank loans or welfare assistance; in a call centre inquiry or summons to an on-demand cab; or in encounters with police or border guards, scanning their shadowed screens.

In these ways, AI has been with us for years. Once upon a time, people on the left referred to the regimes of the USSR and Eastern Europe as ‘actually-existing socialism’ (Bahro 1978), indicating an incipient but imperfect realization of hopes for a new social order. We propose an analogous formulation: ‘actually-existing AI-capitalism’, designating a phase of experimental and uneven adoption of the technologies in which so many hopes are invested. This phase may be protracted far longer than AI enthusiasts anticipate. It may stagnate, stall out and implode (as ‘actually-existing socialism’ did). But it could also intensify or expand in a transition either to a significantly trans- formed capitalism, or to a radically different social formation.

This reference to the fate of socialism brings us to the vantage point from which we critique AI-capitalism. Prophets of a technological sin- gularity expect its arrival around 2045 (Kurzweil 2005a). This would put it 201 years after a critical observer of early industrial capitalism penned his own prediction as to its eventual outcome: ‘finally … an inhuman power rules over everything’ (Marx 1975 [1844]: 366). The young Marx was not writing about artificial intelligence or robots. He was describing the ‘alienation’ of workers dispossessed by capital of control over what they made, how they made it, their relations with fellow human beings and of their very ‘species-being’.

The argument we make in this book is that AI should be seen as the culmination of this process, a moment where the market-system assumes a life of its own. AI, we posit, is ‘alien power’ (Marx 1990: 716) – the power of autonomous capital. We read AI and Marxism through one another: AI through Marx, because Marx’s introduction: ai-capital. 3 analysis of capitalism is the most comprehensive critical account of the fusion of commodification and technology driving AI forward today; Marx in the light of AI, because AI problematizes human exception- alism, agency and labour in ways that profoundly challenge Marxist assumptions, and hence requires careful examination by those who share Marx’s aspiration for revolution against and beyond capital.

THREE POLEMICS The argument of this book interweaves three polemic critiques. The first is a critique of AI as an instrument of capital, with all this entails in terms of both the exploitation in and ejection from waged work of human labour, and the concentration of wealth and social power in the hands of the corporate owners of high technology. Depictions of AI as the outcome of a disinterested process of scientific research are naive. Machine intelligence is the product not just of a technological logic, but simultaneously of a social logic, the logic of producing surplus-value.

Capitalism is the fusion of these technological and social logics and AI is the most recent manifestation of its chimerical merging of computa- tion with commodification. Jump-started by the digital experiments of the US military-industrial complex, AI emerged and developed within a socio-economic order that rewards those who own the means for automating human labour, accelerating sales, elaborating financial spec- ulation and intensifying military-police control over potential restive populations. Whether or not AI may be put to different uses – ‘reconfigured’ (Bernes 2013; Toscano 2014; Steinhoff 2017) to contribute to or create a different social order – is a question we discuss later.

What is apparent is that the owners of the great digital corporations regard AI as their technology – and with good reason, for it is they who possess the intellectual property rights, the vast research budgets, the labour-time of AI scientists, the data and the centres that store it, telecommunications networks, and the ties to an enabling state apparatus that are the preconditions for the creation of AI. It is they and their high-ranking managerial cadres who are in a position to implant their goals and priorities within AI software and hardware, ‘baking-in’ their values – in practice, the one prime directive, to expand surplus-value – to its design.

There may appear to be surprising diversity of opinion about AI amongst corporate leaders, ranging from ecstatic embrace to apocalyptic 4. inhuman power warning. What is shared, however, is the tacit agreement that it is they who are to dictate the direction of AI, to determine in their high-level conclaves and privileged conversations with government how (not if) it is to be adopted, and with what balance between wild gladiatorial free-enterprise competition and cautionary ethical regulation and policy safety-nets to prevent unwelcome tumults. Whether it is Sergey Brin endorsing the idea of the singularity while consolidating his company’s monopolistic powers to direct it, or Elon Musk warning of AI catastro- phe while building (or attempting to build) fully automated factories for the production of self-driving vehicles, or Bill Gates offering feeble robot tax plans (and thereby drawing the instantaneous ridicule of peers), the colourful clashes of corporate personalities cover the more sober reality that these great AI moguls are no more, or less, than the personifications of abstract forces of market calculation that drive towards the maximi- zation of profit.

They also obscure the massive hubris of the capitalist class that believes it can control the forces it has unleashed. For we did not quite complete our titular quote from the young Marx: ‘finally – and this goes for the capitalists too – an inhuman power rules over everything (1975 [1844]: 366, emphasis added). In making this critique of capitalism’s encounter with AI we also, however, take issue with leftist theorists who share such concerns, some of whom, like us, quote Marx in their evaluations of AI. So here we quarrel with interlocutors we respect and have learned from, but with whom we differ.

There are two specific left perspectives against which we argue, perspectives that we dub ‘minimizing’ and ‘maximizing’ views on AI. The left ‘minimalist’ position dismisses current discourses on AI as hype and hucksterism. A more moderate version grants them a limited credibility but insists this is not sufficient to seriously change previous analyses of capital and class (Huws 2014; Moody 2018a). For a strong statement of this minimalist position, we can take Astra Taylor’s ‘The Automation Charade’ (2018), an essay whose basic thesis is that ‘the rise of the robots has been greatly exaggerated’ (like many authors, Taylor sees AI and robotics as pretty much synonymous, an unfortunate gloss we discuss later).

Taylor agrees with our point that technological change, and automation in particular, is not a neutral process, but rather wielded from a position of class power. However, her main argument is that it is not just the actuality of automation, but more its possibility, that is weaponized to intimidate workers. She cites the threats of robotized introduction: ai-capital. 5 burger-flippers and touch-screen self-service kiosks by fast food corpo- rations trying quell the Fight for 15 minimum wage movement. Some of those threats proved hollow, and those that have been realized, Taylor points out, still leave lots of workers toiling in McDonald’s.

In light of this, she proposes ‘making our idea of automation itself obsolescent. A new term, “fauxtomation”, seems far more fitting’ (2018). Socialist feminists, she suggests, have, through their close engagement with domestic toil as unwaged work, a special insight into capitalism’s ineradicable dependence on human labour, even where that labour is unacknowl- edged, unrewarded and conducted by women and racialized minorities. She goes on to stress the way the introduction of machines has intensi- fied, rather than eliminated, work, emphasizing the behind-the-scenes dependence of Silicon Valley’s digital platforms on the invisible work of figures such as content moderators.

Against this background, Taylor takes as a clarion-call moment of insight a response she reports from the famous Marxist feminist theorist, Silvia Federici, to a conference question about capital’s tendency to generate ‘surplus populations’: ‘Don’t let them make you think that you are ’ Many of Taylor’s points are excellent; we expand on some of them later, especially in Chapter 2, where we discuss the labour conditions of AI automation. We concur that ‘automation has an ideological function as well as a technological dimension’, but we disagree with her overall emphasis. While the aggregate employment effects of AI and robotics are uncertain and hotly debated, dismissal of automation as a ‘charade’ is deeply ahistorical.

Generations of workers, from hand-loom weavers to assembly line auto-workers and cold metal print-setters would testify that there is nothing ‘faux’ about capital’s tendency to replace humans with machines. The millions of people migrating from planetary zones bypassed by analog and digital supply chains and automated factories testify to the reality of surplus populations. While Federici may have been quite rightly suggesting that we should rethink who or what should be considered socially disposable, there is no doubt that capital always has made people and indeed entire populations ‘disposable’ (which, of course, is why it has to be resisted).

Shrinking from that reality at the moment when a new instalment of corporate machinic power raises such disposability to a new level, and writing it all off as bluff and hype, may be reassuring, but it is unwise, sentimental and dangerously complacent. Probably recognizing this, at the conclusion of her essay, Taylor abruptly changes course, and concedes: ‘There is no denying that technologi- 6. inhuman power cal possibilities that could hardly be imagined a generation ago now exist, and that artificial intelligence and advances in machine learning and vision put a whole new range of jobs at risk.

Entire industries have already been automated into ’ And she rightly remarks that the ‘emphasis on technological factors alone, as though “disruptive innovation” comes from nowhere or is as natural as a cool breeze, casts an air of blameless inevitability over something that has deep roots in class conflict’ (2018). To which we say d’accord. But confronting these issues demands understanding AI and its automating capacities, accom- panied though they are with abundant mystification and fetishization, as something more than just a ‘charade’. Our third object of critique, the left ‘maximalist’ position, is the diametric opposite of the ‘minimalist’ approach.

Not only does it hold that AI and associated technologies, such as robotics, are ‘for real’, and have the capacity to drastically transform the conditions of production and work, it also sees these capacities as stepping stones to socialism. Proponents of this view look optimistically at the automating capacities of AI as an opportunity to ameliorate, perhaps eventually abolish, the exploitation of wage-labour, opening up prospects for a society in which people enjoy more free time, for pleasure, personal development and/ or political engagement. This seems to offer a path for socialists that is more achievable than the daunting prospect of a full-scale revolution against capital.

Instead, it can be attained by a social democratic government prepared both to foster the technologies of the fourth industrial revolution and to introduce a ‘universal basic income’ (UBI) or ‘citizens’ income’ – a guaranteed payment to all citizens independent of any waged job. Lenin famously wrote that communism equals ‘Soviets plus electrification’. It is fair to say that ‘AI plus UBI’ has become the formula for techno-progressive social democratic thought. A constella- tion of thinkers has formed around this attractor, articulated in works such as Nick Srnicek and Alex Williams’s Inventing the Future: Postcap- italism and a World Without Work (2015); Paul Mason’s Postcapitalism (2015) and Aaron Bastani’s (2014, 2019) arguments for ‘fully automated luxury communism’; the xenofeminist (Hester 2018) line of post-gender futurism; and a cluster of autonomist or post-operaismo theorists.

Again, we sympathize with and in many respects share the aspirations of this group; indeed, one of us has written about digital technologies in a very similar vein (Dyer-Witheford 2014), while another has suggested that Marxism might usefully incorporate similarly maximalist elements introduction: ai-capital. 7 of transhumanist thought (Steinhoff 2014). However, we have written this book in part to directly challenge some premises of such AI-optimism. In particular, we want to contest the idea that AI can easily be detached, disentangled and re-appropriated from capitalism. Here it is useful to think about the sources on which the maximalist position draws – in part from Marx’s own sometimes enthusiastic embrace of the modern- izing powers of the forces of production to catalyse the emergence of socialism or communism, and also the uptake and reinterpretation of this position by poststructuralist theorists such as Gilles Deleuze and Félix Guattari.

Perhaps the most important source, however, is the ‘accel- erationist’ thinking of the anti-Marxist philosopher Nick Land (2012), who uncompromisingly argues for and celebrates what he sees as the unstoppable and species transforming (or terminating) power of com- putation. Indeed, the inaugurating document for the maximalist line of thought we have mapped here is Williams and Srnicek’s ‘Accelerationist Manifesto’ (2013), which attempted a leftist re-do of Land’s thought: the most accurate shorthand for the group of ‘maximalist’ theorists we have described is ‘left accelerationists’. As we will argue at length later, this appropriation of Landian thought dodges some of its originator’s key arguments.

For one, Land (2014) held AI to be the consummatory technology of capitalism, one that implanted the logic of capital at its very core. AI, in Land’s view, is not merely appro- priated by capital, but constituted by it: it is a technology made from and for its processes of labour automation, commodity acceleration and financial speculation. A second, yet more disquieting Landian point, is that this mutual embedment of capital and AI leads not to human eman- cipation from capitalism, but, on the contrary, to capital’s emancipation from the human: a capital that no longer needs homo sapiens; human

1 These are not comfortable thoughts. And they are made even less comfortable by the fact that Land in his recent writings has emerged as a reactionary champion of racist and misogynist ‘dark enlightenment’ ideas that have in complex ways infiltrated the culture of Silicon Valley where much AI production takes 2 While we emphatically disas- sociate ourselves from this aspect of Landian thought, we nonetheless believe that any communist position on AI has to take his original accel- erationist proposition – that AI has an elective affinity with capitalism and is fundamentally inhuman – seriously.

Given our anti-capitalist critique of both left minimalist denial and maximalist celebration of AI, it might be expected we go on to enunciate 8. inhuman power some middle-ground, moderate position. This is not the case, or true only in the sense that we want to remove the floor beneath both minimalist complacency and maximalist optimism. Our critique of AI can best be characterized as ‘abyssal’, and this in two senses. First, we confess, as we think other AI thinkers should, that there are vast indeterminacies about the directions and destinations of AI-infused capitalism. Peering into the conflicting estimates of AI’s near and far future capacities and deploy- ments can, and should, instil political vertigo.

The everyday uses of AI now commonplace in advanced capitalism give some indicators of its future trajectory, but no certainties. This may seem an odd assertion for a Marxist theorization of AI, given that Marxism has in many incarna- tions asserted bold teleological certainties; however, as we argue later, Marx’s work itself contains divergent accounts of the outcome of capital’s technological compulsions. We read it as a matrix of possibilities, rather than a promissory note. In and of itself, this approach undercuts compla- cencies both that social struggles persist unchanged, regardless of new technologies, or, conversely that, because of the same new technologies, capital’s self-destruction is imminent.

That said, however, the second ‘abyssal’ aspect of our AI analysis is that amongst the maze of future possibilities, some potential outcomes can be discerned that are far more deeply disturbing than is allowed by either the maximalist or minimalist positions, with their respective confidence about the continuation or the end of capitalism. These outcomes throw into question assumptions about the labour theory of value, the continued centrality of struggles at the point of production, or even the confidence that capitalism cannot survive the abolition of its human waged workforce. These points demand consideration, not to justify defeatism, but as a component of a revival of revolutionary communist thought.

This is what we mean when we say that, at the same time as making a Marxist critique of AI, we make an AI-informed critique of Marxism. What then is AI? ‘A MACHINE CAN BE MADE TO SIMULATE IT’ To understand the effervescence surrounding AI we need to define what AI is and how it functions. We are, emphatically, not AI experts; we will make errors deriving from our lack of technical knowledge as well as from the rapidly evolving nature of the field. Despite such difficulties, we believe grappling with basic AI concepts and how AI actually works

introduction: ai-capital. 9 is important. Too many accounts of AI, celebratory or dismissive, skip this effort. But it is only through some familiarity with the science and technology of AI that an effective critique can be mounted. The workshop at Dartmouth College in Hanover, New Hampshire in 1956 is usually taken as the start of the field and study of ‘artificial intel- ligence’. The organizers described their goal as follows: The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.

An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves. (McCarthy et al. 1955) Since then, definitions of AI have been many and vague. AI experts show no consensus (Faggella 2018c). Compounding this definitional problem is the ‘AI Effect’, whereby as soon as AI can do something, it is no longer considered to require intelligence. Pamela McCorduck noted that in the history of AI ‘every time somebody figured out how to make a computer do something – play good checkers, solve simple but relatively informal problems – there was chorus of critics to say, “that’s not thinking”’ (2004: 204).

One recent AI textbook quotes Elaine Rich’s pithy definition of AI from 1983: ‘the study of how to make computers do things at which, at the moment, people are better’ (quoted in Ertel 2018: 2). A more formal definition of AI we find useful is: The essence of AI – indeed, the essence of intelligence – is the ability to make appropriate generalizations in a timely fashion based on limited data. The broader the domain of application, the quicker con- clusions are drawn with minimal information, the more intelligent the behaviour. (Kaplan 2016: 5–6) This definition distinguishes AI from mere computation and allow us to differentiate between different types of existing and hypothetical AIs by considering their speed, quantity of information required, and generality of application.

Kaplan’s definition, however, says nothing about what AI looks like out in the world. AI does not mean robot, a confusion that can be blamed on pop culture. The roboticist Alan Winfield offers three complementary definitions of a robot: 10. inhuman power 1. an artificial device that can sense its environment and purposefully act on or in that environment; 2. an embodied artificial intelligence; or 3. a machine that can autonomously carry out useful work (2012: 8) The most important aspect of Winfield’s definitions is that, despite differing morphologies, all robots have bodies. AI, however, is software and, therefore, need not be embodied, though it requires computing hardware to run on.

Advanced robots employ AI for functions including perception, planning actions, and learning, but a robot body does not necessarily entail AI, nor does an AI system necessarily entail a robot body. To distinguish actually-existing AI from its speculative future incar- nations, it is helpful to employ the following three categories: narrow AI, artificial general intelligence (AGI), and artificial superintelligence (ASI). Actually-existing AI is narrow: ‘the vast majority of current AI approaches … are primarily designed to address narrow tasks’ (Johnson et al. 2016: 4246). Most AI research, all commercial applications of AI, and the AI that consumers use daily, are such task-based tools.

They are functionally more akin to microscopes than the anthropomorphic and politically active droid L3–37 in Solo. These systems have none or very little ability to do anything beyond their particular domain of function- ality. An AI system that recognizes faces in photographs is not going to be able to process recordings of speech, play Go, or compose emails, and it is definitely not going to be able to speak Farsi. We will discuss dozens of existing narrow AI systems over the course of this book. On the basis of generality, narrow AI is contrasted with artificial general intelligence (AGI), which refers to AI with ‘the capacity for efficient cross-domain optimization’ or ‘the ability to transfer learning from one domain to other domains’ (Muehlhauser 2013).

AGI refers to an AI with the capacity to engage and behave intelligently in a wide variety of contexts and to apply knowledge learned in one context to novel situations, meaning it would be ‘capable of reasoning across many intellectual domains’ (Baum 2018a: 3). As of 2019, AGI remains a spec- ulative technology, although serious research is now being conducted on it in both public and private institutions. We discuss AGI in Chapter 3. Artificial superintelligence (ASI) is yet more speculative. While an ASI ‘is likely to have general intelligence’ (Baum 2018a: 3), it specifically refers to an AI ‘that greatly outperform[s] the best current human minds

introduction: ai-capital. 11 across many very general cognitive domains’ (Bostrom 2014: 63). ASI is a science fiction staple, but serious discussion of it also occurs in academic circles where it is often seen as swiftly following the creation of AGI (see g. Bostrom 2014; Torres 2018; Baum 2018a; 2018b). Most commonly, the scenario imagined is that an AGI gains the ability to self-modify and evolves into a god-like ASI with unpredictable powers. The consequences of such an event are impossible to predict with certainty, but the mere possibility of it occurring compels thinkers and institutions – including Nick Bostrom and the Future of Humanity Institute at Oxford, Seth D.

Baum and the Global Catastrophic Risk Institute, as well as Eliezer Yudkowsky and the Machine Intelligence Research Institute (MIRI) – to argue that we must seriously research the possibility now. Another important distinction is that between ‘strong’ and ‘weak’ AI. While sometimes the term strong AI is used to refer to AGI (Kurzweil 2005a: 260), the term originally derives from the work of the philoso- pher John Searle (1980), who used it to describe the position of those who believe that an advanced AI would be conscious. Searle critiques this view from the sceptical position of weak AI, which holds that machines can never be conscious.

Searle’s famous Chinese Room thought experiment, which hypothesizes a human (or machine) equipped with an encyclopaedic set of rules for translating Chinese into English without being able to speak or understand the language, attempted to prove this position. As Kaplan puts it, ‘strong AI posits that machines do or ultimately will have minds, while weak AI asserts that they merely simulate, rather than duplicate, real intelligence’ (2016: 68). We do not take a definite stance on the question of machine consciousness. The arguments in this book do not depend on machine consciousness being physically or even logically possible, nor on the impossibility of such.

On this topic we are functionally agnostic, and, as we argue in Chapter 3, so too is capital. Actually-existing narrow AI is typically divided into three schools of thought: Good Ol’ Fashioned AI (GOFAI), machine learning (ML), and the situated, embodied and dynamical framework (SED). GOFAI, also known as symbolic AI, was the first approach to AI and remained dominant until the 1980s (Boden 2014: 89). It is an approach that aims to implement high-level cognitive functions, such as logical reasoning, in machines through the manipulation of information encoded in a symbolic language. Such a system creates internal representations of its world or a problem domain in a symbolic language and performs logical

12. inhuman power manipulations on this representation to think or act. These systems are often constructed out of sets of clearly defined rules. The best examples of GOFAI are so-called ‘expert systems’ or ‘knowledge systems’, which emerged and proliferated in the 1980s. These were intended to capture the knowledge of human experts and make it available to less skilled workers or ignorant managers. Expert systems were used for medical diagnosis, credit scoring and analysis, and business management, but the most famous is IBM’s chess-playing system Deep Blue, which in 1997 defeated the reigning world champion Garry Kasparov.

However, GOFAI required vast sets of rules with myriad possible interactions. Solving complex problems in this way necessitates tremendous com- putational power; for these and other reasons, approaches other than GOFAI were pursued. One reaction to the problems of GOFAI is the ‘situated, embodied, dynamical (SED) framework’ (Beer 2014: 128); Rodney Brooks, a pioneer in the field, called his approach ‘nouvelle AI’ to emphasize its quali- tative break with GOFAI (Copeland 2000). SED researchers are often motivated by ‘Moravec’s paradox’ – the observation by roboticist Hans Moravec that ‘it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility’ (1988: 15).

SED approaches to AI emphasize the irreducible importance of the body – with its perceptual apparatuses and morphology – to cognition: for this school, it is through solving material problems that machines can evolve intelligent behaviours. Such approaches are therefore often concerned with robotics and artificial life in addition to AI. SED practitioners initially focused on very simple, insect-like robots, but in the 2010s more complex, partially humanoid robots became possible and have been introduced into industrial settings. It is possible that some variety of the SED framework could be the next dominant AI paradigm.

Another reaction to GOFAI was machine learning (ML). The ML school, formerly called connectionism, existed as early as the Dartmouth workshop, gained some traction in the 1980s with advances in learning algorithms, but did not explode until the 2010s when big data and cheap computing power proliferated. As of 2019, ML is the dominant approach to 3 It is a statistical pattern-recognition approach. One NVidia researcher has described ML as a process comprised of three steps: ‘(1) take some data, (2) train a model on that data, and (3) use the trained introduction: ai-capital. 13 model to make predictions on new data’ (Dettmers 2015).

In other words, ML systems can be understood as creating their own models of inference. While ML may operate on a variety of architectures, the cutting edge of AI in the 2010s has largely run on artificial neural networks (ANNs) – computer programs that are inspired by, albeit quite different from, the human brain. ANNs roughly mimic the electrical operations of the brain’s neuronal connections rather than emulate high-level logic like GOFAI does. ANNs are ‘based on the assumption that cognition emerges through the interactions of a large number of simple processing elements or units

, ‘neurons’)’ (Sun 2014: 109). Artificial neurons are organized into a series of layers and each layer is connected to the layers above and below. The lowest level receives inputs – g. images, text or speech – in the form of data that has been vectorised (converted into long strings of numbers). Higher levels – called hidden layers – process data that is sent up from the layers below them. Early networks had only one hidden layer, but today’s networks have many more. In general, the more layers the underlying ANN architecture of the ML system has, the more complex patterns it can find.

The most advanced ML in the 2010s is called ‘deep learning’ because these networks are many layers deep (LeCun, Bengio and Hinton 2015: 436–44), with some networks possessing as many as 1,000 (He et al. 2016). While previous ML networks were constructed by hand, a ‘key aspect of deep learning is that these layers of features are not designed by human engineers: they are learned from data using a general-purpose learning procedure’ (LeCun, Bengio and Hinton 2015: 436). The artificial synapses which connect the layers of artificial neurons are ‘weighted’ with numeric values representing the strength of the con- nection.

The network ‘learns’ through adjusting the weights of these connections. We can thus think of an ML system as a fixed template with changeable parameters; ‘by assigning different values to these parame- ters, the program can do different things’ (Alpaydin 2016: 24). ANNs are exposed to a data set, which might be images of faces or audio clips of people saying hello. In a process called ‘training’, the network is exposed to many instances of the chosen object(s) and the weights of the synapses are adjusted by a learning algorithm until the network learns to output the correct response, recognizing faces or the word hello, as the case may be (Kaplan 2016: 30).

14. inhuman power It is useful to distinguish between the three broad types of machine learning – supervised, unsupervised and reinforcement learning. Supervised learning has been the most successful so far. In this approach, input data is labelled by human teachers, usually in terms of categories, and the system learns those categories by discerning patterns across the supplied examples. Given enough photos of red hexagonal signs with the word STOP on them, in various visibility conditions, and from various angles, a supervised learning system can learn the concept of a stop sign. However, because of the necessity of labelling, supervised learning entails a lot of human labour.

This has driven companies to develop techniques of unsupervised learning to enable a network to generate categories and labels on its own (Alpaydin 2016: 117). The idea is that with exposure to enough data, the system will identify ‘incredibly sophisticated and complex correlations’ across the data set (Kaplan 2016: 30). In so doing, it may be said to generate its own concept of a stop sign. Some deep learning pioneers argue this will eventually become the central ML approach because it mimics how humans and animals evolved to learn; not by being told what everything in the world is called, but through observing it (LeCun, Bengio and Hinton 2015: 442).

Reinforcement learning lies somewhere in between supervised and unsupervised learning. The pioneers of this approach describe it as ‘learning what to do … so as to maximize a numerical reward signal. The learner is not told which actions to take … but instead must discover which actions yield the most reward by trying them’ (Sutton and Barto 1998: 127). Reinforcement learning was thought to be limited to simple domains, until 2013 when the UK firm DeepMind combined it with unsupervised learning to teach a system to play Atari games with superhuman skill, without programming any knowledge about the games into the system, and giving it access only to the score of the games and the pixel information displayed on the screen (Knight 2017).

The same combination enabled AlphaGo’s win over Go master and world champion Lee Sedol. As Kaplan emphasizes, the learning in ML should be understood as ‘extract[ing] patterns from data’ (2016: 4 Instead of being built top-down as a set of rules for handling data, ML systems go bottom-up: ‘learning algorithms … are algorithms that make other algorithms … computers [that] write their own programs, so we don’t have to … [it is] the inverse of programming’ (Domingos 2015: 6–7). This is why ML advocates see it as a Copernican revolution in programming. If a introduction: ai-capital.

15 system can learn, the designer does not need to anticipate and program a solution for all potential situations the system is exposed to (Alpaydin 2016: 17). Ideally, the system will develop its own solutions and ‘the data itself … defines what to do next’ (Alpaydin 2016: 11). This would not, however, be a very powerful attribute of ML if these systems could not generalize what they have learned to data not included in the training data set. Thus they are assessed according to their ‘gen- eralization ability’ (Alpaydin 2016: 40). While the ML systems of 2019 are at best able to generalize to new data of a similar kind to that which they were trained on, increasingly sophisticated systems may approach AGI’s hypothetical capacities for generalization across domains.

This book is about the implications of AI, from machine learning to AGI, for the future of capitalism. MACHINE MARX Machinery is crucial in Marx’s analysis of capital, so much so that almost any serious study of his work engages the topic in some way. Here we rapidly map those aspects of Marx’s thought that are most important to the discussion of AI, highlighting points relevant for our arguments later in this book. There are, we suggest, three main strands in Marx’s thoughts about machines. In the first, major line, the machine is a supplement to the human labour that is the crucial creator of value within the capitalist system.

While production becomes increasingly machinic and inten- sifies the exploitation of workers, machinery ultimately contributes to the system’s terminal crisis. The other two, minor lines, logically emerge from this major current but also depart from it. Both of them posit a moment at which the machine becomes autonomous from labour. However, one sees the consequences as liberatory, the other as night- marish. These three strands of Marx’s machine-thinking can, with care, be conjugated together, but this depends on glossing tensions between them that widen and deepen as we confront the conundrums of AI. What we term the major line of Marx’s machine analysis unfolds in the first volume Capital.

In this account, machines, along with other equipment, buildings and raw materials, are ‘constant, fixed’ capital. 5 This is contrasted with the ‘variable’ capital of human labour (Marx 1990: 508–9; 1992: 237–48). This distinction between ‘fixed’ and ‘variable’ capital rests on the basic proposition that it is only human labour that creates value within capitalism: the machine, however gargantuan its 16. inhuman power powers seem relative to those of humans, can only act as a supplement or force-amplifier to the essential, human activity, increasing its efficiency, albeit by manifold times. Machinery, which has itself been built by humans, is ‘dead labour’.

Rather than generating new value, the machine already has value which it transfers to the product: it ‘yield[s] up its own value to the product it serves to beget’ (Marx 1990: 509). The social function of fixed capital is to produce relative surplus-value, which it does by reducing necessary labour-time and, hence, increasing surplus labour-time, e. by ‘shortening the part of the working day in which the worker works for himself, to lengthen the other part … he gives to the capitalist for nothing’ (Marx 1990: 492). Increasing the productivity of labour means that the worker’s output is increased: more commodities are produced in less time, and consequently these commodities are cheapened because less value is objectified in each individual commodity.

Marx’s detailed explanation of machinery occurs in Chapter 15 of Capital, Volume 1, which concerns ‘Machinery and Large-Scale Industry’ (1990: 492–639). Here, Marx was concerned ‘only with broad and general characteristics’ of machinery (1990: 492). The genealogy of machinery is found in tools; the instruments of handicraft labour are turned into machinery, automation technology being ‘fully developed machinery’, which has three different parts: (1) the motor mechanism that is the driving force or motive power; (2) the transmitting mechanism that regulates, changes and distributes motion; and (3) the tool or working machine that uses this motion to modify the object of labour (1990: 494).

A key component of any machine is its ‘emancipation from the restraints of human strength’ which occurs when it obtains a regular and controllable motive power; Watt’s double-acting steam engine being the first major ‘self-acting prime mover’ (1990: 502). Whereas with mere tools the process of production is adapted to the worker, the system of machinery is a ‘vast automaton’ which confronts the worker as a ‘pre-existing material condition of production’ (1990: 508). The section on ‘The Struggle Between Worker and Machine’ in Chapter 15 emphasizes how capital’s technological dynamic is insepa- rable from class conflict.

Marx examined the paradox by which, under capital, labour-reducing machinery creates a hell for labourers: the mechanical lightening of demands for physical strength catalyses the large-scale induction of women and children into factories; the capacity of machines to run indefinitely, and the need to pay for their purchase, introduction: ai-capital. 17 leads to a prolongation of the working day; the ability to accelerate and multiply machine operations results in the intensification of work. Marx also addressed technological unemployment, describing how ‘[t]he instrument of labor strikes down the laborer’ in a process where machinery ‘act[s] as a superior competitor to the worker, always on the point of making him superfluous, and capital proclaims this fact loudly and deliberately, as well as making use of it’ (1990: 562).

Indeed, Marx suggested ‘it would be possible to write a whole history of the innova- tions made since 1830 for the sole purpose of providing capital with weapons against working class revolt’ (1990: 563). Mechanization is also propelled by competition between rival cap- italists. The value of a commodity depends on the amount of socially necessary labour expended in its production. If a capitalist can, by intro- ducing technology, reduce the labour for which she pays, while still selling the product at the prevailing price, she will enjoy greater profits than her competitors. This advantage will eventually be neutralized as use of the labour-saving innovation becomes generalized, but this just sets the scene for the next wave of automation.

Thus both class conflict and competition between enterprises give capital an intrinsic drive to replace humans with machines. Marx described this as the tendency for capital to increase its ‘organic composition’, that is to say, the proportion of constant (machines, buildings and raw materials) compared to variable capital (labour) (1990: 762). Machinery also, however, throws capital into crisis. In Capital, Marx alternated between two explanations as to why this is. One, perhaps the most readily understandable, is that because machinery enables capi- talists to increase production while (other things being equal) reducing their wage bill, it fosters gigantic imbalances between the increasing volumes of commodities produced and the purchasing power available to buy them.

This brings on economic stagnation and paralysis, with factories closing and unemployment lines growing, until enough firms go out of business to eliminate the glut of overproduction and get the system moving again – or a social revolution breaks out. Marx’s other explanation, less intuitively obvious, but arguably more profound, has to do with the tendency to a falling rate of profit (FROP) (1991: 317–38). Because the value of a commodity ultimately depends on the amount of socially necessary labour required for its production, replacing humans with machines lowers the value of the commodity – and hence, eventually, the price it commands, and the profit per item

18. inhuman power capital can command. Because automation cheapens goods, it makes each single one of these goods less profitable to the capitalist. In the face of this tendency of the rate of profit to fall – a direct result of automation – capital can, at least for a time, maintain the mass of profits by increasing the sheer volume of production, but it is running against its own value-decreasing machinic momentum. Marx (1991: 339–48) listed a number of ways capital can hold this process at bay or even temporar- ily reverse it, but the tendency of increasingly machinic production is, again, towards spasmodic crisis, this time caused by the flagging prof- itability of business, and moments of unemployment, immiseration and social tumult.

These two versions of crisis theory, and their degree of compatibil- ity, have been intensely debated. But in both versions, capital’s recurrent crises arise from its inherent drive to substitute constant (machinery) for variable capital (labour). This either reduces consumption power (by cutting wages) or lowers the profitability of production (by cheapening goods), or both. The outcome is repeated throughout deepening cycles of economic breakdown, each of which offers the possibility of a revolution by a working class suffering from downward pressures on wages and the threat of unemployment, but still central enough to production to halt or take control over it.

This, then, with its internal bifurcations and attendant controversies, is what we call the major line of Marxist thinking about capitalism and machines. An analysis of AI undertaken within this current will analyse it in terms of labour exploitation, inter-capitalist competition and capitalism’s techno-induced crisis tendencies. What we dub the two minor tendencies in Marx’s thought are extensions of, but also deviations from, this major line. Latent in Marx’s account of capital’s increasing mechanization is the idea that the positions of worker (initially the main, value creating actor) and the machine (at first the worker’s power-amplifying supplement) invert.

The worker, who at the handicraft stage was the subject of the labour process, becomes an automaton of repetitive, repeated motions, responding to automatic machinery rather than using it; the automatic machinery has become the subject of the labour process. Two subsidiary, perhaps maverick, lines of Marx’s thought develop this idea to its logical conclusion – but in two different directions. The first, and by far most famous, comes from the ‘Fragment on Machines’ in Marx’s Grundrisse, which from the 1970s on has been seen as an extraordinary anticipation of high-technology capital. In introduction: ai-capital.

19 the ‘Fragment’, Marx envisaged capital making vast techno-scientific achievements by mobilizing the ‘general intellect’. This enables it to reach a level of automation that, while not eliminating human labour entirely, reduces and relegates it to the peripheral position of supervising a mainly machinic process. This might seem the final triumph of capital over its troublesome working class, but Marx in the ‘Fragment’ presented it as a pyrrhic victory. By removing the necessity to base production on wage-labour (and hence liquidating the possibility of basing consump- tion on waged income), it undermines value, e. the whole basis of capital’s social organization.

Automation inadvertently subverts capital by abolishing work. This is consonant with other celebrated passages of Marx’s concerning the liberatory nature of technological development, most notably the account of how developing ‘forces of production’ burst apart fossilized ‘relations of production’, making way for the appearance of a whole new ‘mode of production’ (Marx 1973 [1859]). At first glance, the ‘Fragment’ might seem just to restate in especially emphatic form the predictions about the mounting organic composi- tion of capital that inform Marx’s thinking about the FROP. As George Caffentzis (2013) points out, however, there is a divergence between the two theories, for while the FROP depends for its operation on the validity of the labour theory of value, the ‘Fragment’, in contrast, posits a dynamic of capitalist collapse arising from the liquidation of labour – and value.

The Grundrisse was translated into English, French and Italian at the very time when computers were first beginning to enter the workplace, and from the moment of its appearance it was taken as harbinger of a high-technology, ‘cyborg’ communism which would overcome all the dif- ficulties a drab, industrial, actually-existing socialism was experiencing in organizing work on a non-capitalist basis by simply eliminating the need to work at all. It is therefore no surprise that the ‘Fragment on Machines’ is a foundational document for theorists of left accelerationism, postcapital- ism and fully automated luxury communism, so much so that Frederick Harry Pitts (2017) has dubbed these all instances of ‘Fragment Thinking’.

A second minor strain in Marx’s thought, less discussed than the ‘Fragment’, is the nightmare vision of capital in the ‘Results of the Imme- diate Process of Production’ (appearing in some editions of Capital as an Appendix to Volume 1) (1990: 949–1084). In this text, Marx gives an account of the process of capitalist ‘subsumption’ – roughly translated as domination, envelopment or take-over – of labour. He details two moments: formal and real. In the first moment, capital organizes labour 20. inhuman power as wage-labour, thus merely changing its social form, while leaving the content of labour,

e. how it is carried out, the same as in pre-capitalist artisanal handicraft. In the second moment, however, the content of labour changes – it is really subsumed – to better meet the dictate and demands of the capitalist production of surplus-value. Initially, real subsumption occurs through introducing a division of labour into the handicrafts, but subsequently by the automation of labour by machinery, which requires capital to absorb socially produced scientific knowledge to develop technologies adequate to and commensurate with its own pri- orities – notably, the automation of production and the acceleration of the circulation of commodities.

In the transition from formal to real sub- sumption, ‘absolute’ exploitation of labour – the extension of the working day – is displaced by the extraction of ‘relative surplus-value’, which is based on increasing the productivity of labour by intensifying the labour process through the division of labour and machinery. As this process builds, Marx argues that a situation emerges in which, in the industrial factory (only nascently visible in his time), the worker confronts a fully ‘alien power’ that appears endowed with a ‘colossal independence’ from human agency, rendered ‘autonomous’ by techno-science. This account might seem just another description of the mounting organic composition of capital, or indeed of the semi-automated ‘animated monster’ of capitalist machinery featured in Grundrisse (1993: 470).

There are, however, differences in inflection. For one thing, since the subsumption argument is that capital actually adopts machinery that it designs to its systemic requirements (the valorization of value through elimination of human labour and acceleration of commodity circula- tion), it becomes more difficult to envisage how the ‘forces of production’ conflict with ‘relations of production’ – if anything, the former would seem to reinforce the latter. There is certainly no hint in the ‘Results’ of either the crisis-inducing falling rate of profit or of the self-destructive labour-abolishing logic of the ‘Fragment’. There is just the towering presence of an all-but-incomprehensible production apparatus that looms over and surpasses the worker it once depended on.

This is all the more apparent because, while the first half of the appendix includes one of the most complete discussions in Marx of the powers of the ‘collective worker’ engaged in the cooperative fusion of various types of work, from manual labour to engineering, by the end the collective worker is dwarfed and seems virtually obliterated by the machine apparatus to which its powers have been transferred. introduction: ai-capital. 21 It can fairly be argued that, read as an appendix to Capital, this document should be understood only in the light of what precedes it, so that capitalist breakdown can be assumed.

And it can further be asserted that the machinic autonomy of capital is only an ‘appearance’ – a mystifi- cation that has to be seen through to detect the continuing, if baroquely veiled, importance of labour-power situated down remote production chains. Maybe so. But as even chronic optimists like Antonio Negri (2017) note, ‘appearance’ in Marx doesn’t mean ‘shadowy, superficial or insubstantial’: on the contrary, it means a concrete social reality created on the basis of a mystified and disguised process, namely the incremen- tal subsumption of the power of workers into machinery. While Marx developed this sombre vision through analysis of the industrial factory, it invites thought about a further stage of ‘hyper-subsumption’ in which capital’s autonomizing force manifests as AI: later in this book we explain how this new stage of subsumption is unfolding, and may culminate in AGI.

Both the major and minor lines of Marx’s machine analysis describe dynamics in play in the actually-existing AI-capitalism of 2019. We are not at the heights or depths of machinic capitalism foreseen in either the ‘Fragment’ or the ‘Results’. Workers on Foxconn assembly lines, in Amazon warehouses or on Facebook content moderation sites all attest to the continued use of machines to intensify and speed up human labour. Crises such as the 2008 Wall Street crash reveal the deep insta- bilities which machinically depressed wages, digitally organized cheap labour and high-speed financial trading are bringing to capitalism.

So much of the major line of Marx’s machine thought holds true, perhaps truer than ever, as the 2020s approach. But ML-driven AI, developed in part in response to the crisis of capitalist globalization, is placing on the horizon possibilities that resemble those in Marx’s visions of capitalism’s machinic extremes. The ‘Fragment on Machines’ and the ‘Results of the Immediate Process of Production’ are recto and verso of one page, a page that speaks both of machine power liberating humanity from capital, and of a capital rendered autonomous from humanity. Capitalists and communists alike, be careful what you wish for!

IN THE AGE OF SELF-REPLICATING AUTOMATA Marx’s discussion of machinery stops with the steam-powered factory. Only towards the end of his life were electricity and electromagnetism 22. inhuman power harnessed, and although early computer technology like Jacquard’s Loom and Babbage’s failed Difference and Analytical Engines existed when he wrote, he did not discuss them. So to bring Marxism to bear on AI, Marx’s account must be amended. There is a formidable Marxian literature on ‘cybernetic capitalism’ and ‘digital labour’, but Marxian analysis specifi- cally devoted to AI, analysing the specificity of its technology, political economy and class implications, is relatively rare.

There are, however, three treatments that have influenced our thinking. The first is Tessa Morris-Suzuki’s examination of Japan’s high- technology capitalism in the 1980s, the period which saw the introduction of robots in auto manufacture, the explosive growth of the video game industry, and projects such as Japan’s ultimately doomed ‘Fifth Generation Computer Systems’, an early AI initiative supported by Japan’s Ministry of International Trade and Industry. Writing in this context, Morris-Suzuki suggested – much as we have here – that some of the disarray of the contemporary left stems from its reluctance to confront the possibility of a highly automated capitalism, instead taking ‘one of two contrasting positions’ – either denial ‘that the contemporary “information revolution” represents any fundamental change in the nature of capitalism’ or the assertion ‘that it spells the death agony of the capitalist system’ (Morris- Suzuki 1986: 81).

Taking issue with the claim by the famous Marxist scholar Ernest Mandel (1975: 207) that large-scale automation of production constitutes the ‘absolute inner limit’ of capitalism, Morris- Suzuki argued it was time to consider ways capital might perpetuate itself under such conditions. These, she said, would include the transfer of labour from production to ‘perpetual innovation’, a proletarianization of technical jobs, the corporatization of an education system geared to the production of elite research scientists, and the creation of a workforce ‘easily taken up and easily discarded’ (Morris-Suzuki 1984: 120). This now reads as a prescient description of the present.

The new wave of AI poses the same problem Morris-Suzuki articulated, but at a higher level. ML and other new AI techniques are beginning to encroach on the activities she saw as the only available refuge for labour chased out of industrial production by machines. The second is Ramin Ramtin’s remarkable Capitalism and Automation: Revolution in Technology and Capitalist Breakdown (1991). In this book, Ramtin made the first systematic attempt to rethink Marx’s theory of machinery in the light of the cybernetic technology that was driving the digital automation of the 1980s. He proposed that to Marx’s three-part

introduction: ai-capital. 23 anatomization of industrial machinery, comprising motor power, trans- mission mechanism and tool head, had to be added the guiding or control function – a function once considered dependent on human intelli- gence and senses but now increasingly automated with information, including sensor, technology. By proposing this revision Ramtin offered a way towards a theorization of computers that, without endorsing the post-industrial euphoria about the information revolution, also recognized the qualitative change digital technologies brought. Ramtin’s work was, however, also notable for the unflinching eye it cast on the possible consequences of this development for a Marxist analysis of class struggle.

He suggested that the full-scale cybernetic onslaught of capital against its working class would bring to the fore issues of unemploy- ment that had receded into the background during the postwar boom. His insistence that Marx’s notion of proletarianization be recognized as a concept not just of workplace exploitation but also of the liability of ejection from work in many ways anticipates the discussions of ‘surplus populations’ that would emerge in the wake of the 2008 crash, which we take up later in this book. Morris-Suzuki and Ramtin not only pointed to important changes in work and labour conditions associated with early AI.

Between them, they also provided important revisions of Marx’s basic conceptualiza- tion of machinery. Ramtin, drawing on cybernetics theory, pointed out that Marx’s account of machines omits the key function of the control, presumably assuming it is ultimately directed by a human agent. This assumption is, however, Ramtin pointed out, precisely what was challenged by cyberneticists as they introduced the theory of feedback into machine operation. It is, he suggested, precisely the control function that distinguishes automation from mechanization and makes it qual- itatively new. Morris-Suzuki emphasized the vastly increased scope and flexibility of machine application that comes with the separation of ‘hardware’ and ‘software’: with machinery whose operations can be changed by the switching of instructional programs, so that machines start to attain some of the variability that had previously been seen as unique to human labour.

While Ramtin and Morris-Suzuki provide analytic anticipations of AI drawn from early moments of digital automation, our third exemplar of Marxist AI analysis, George Caffentzis, provides a crucial theori- zation as to why many of the predictions of jobless futures of that era have not come true. In a series of essays written from 1980 to 2008, 24. inhuman power Caffentzis argued that the apparent job-destroying powers of AI had to be considered in light of its antithesis, the expansion of the service sector and global sweatshops; one had to think ‘Africa’ and ‘automata’ together.

He draws on the ninth chapter of Volume 3 of Capital, ‘Formation of a General Rate of Profit (Average Rate of Profit) and Transformation of Commodity Values into Prices of Production’ (1991: 254–73). Here Marx suggested that while the profit extracted by capital as a whole depends on the overall amount of surplus-value extracted within its entire system, there is no direct correspondence between any individual capitalist’s profit and the amount of socially necessary labour they employ. Value is a social phenomenon and any and all value produced goes into a social pool after the commodities have been exchanged.

But capitalists also appropriate surplus-value from this pool in the form of profit; an individual capitalist will appropriate more profit if their capital, relative to other capitals, is of a larger size, has a higher organic composition, and a higher average profit rate (Marx 1991: 241–73; Caffentzis 2013: 132–4). Thus highly automated businesses syphon-off the surplus-value generated by labour-intensive capital. Caffentzis called this ‘the law of the increasing dispersion of organic composition’, by which ‘every increase in the introduction of science and technology... in one branch of industry... will lead to an equivalent increase in the introduction of low organic composition production in [an]other’ (2008: 65).

Caffentzis’s account of capital’s contradictory movements towards high-technology (automata) and low-wage-labour (Africa) suggests reflection on how its current AI-fever is induced not only by technological breakthroughs, but by increasing frustrations in finding cheap labour. AI NOVUM Before and during the writing of this book, in addition to assimilat- ing materials about Marxist theory and the science, economics and sociology of AI, we read and watched a great deal of AI science fiction (henceforward, AI-SF), including not just the classic films – 2001, Blade Runner, Terminator – that are inevitable points of reference for all AI discussion, but a wave of more recent writings and productions that accompanied the emergence of ML.

Some will regard this as evidence of impaired judgement. Harry Collins (2018: 5–13) renames AI as ‘Artific- tional Intelligence’ because, he argues, SF has encouraged a widespread overestimate of AI capabilities: depictions of the purported superhuman introduction: ai-capital. 25 singularity encourage ‘the surrender’ – human abdication to stupid computer programs. Similarly, Mike Cook (2018), an AI researcher, designates AI-SF, alongside excessive respect for scientists and the over-selling of AI, as one of a set of cultural factors contributing to a ‘basic lack of understanding’ about what ML can and cannot do. We agree, but think that this is not all there is to AI-SF.

Darko Suvin (1979) famously proposed that SF cognitively explores potentiali- ties incipient in a society at a given moment by focusing on a ‘novum’ (Latin for ‘new thing’) – a term he borrows from the Marxist theorist Ernst Bloch (1986 [1955]) to denote some new force appearing on the ‘front line of historical process’ (Suvin 1979: 63–84). Centring itself on the novum of AI, and thereby estranging and de-familiarizing our current reality, AI-SF conducts thought experiments about the possible directions – dizzyingly utopian, terrifyingly dystopian or depressingly mundane – of actually-existing AI-capitalism. In the case of AI-SF, the importance of these thought experiments is accentuated because of the now well-documented feedback loop between computer science and science fiction, in which scientists inspired by SF work towards the actu- alization of its imagined worlds.

It is, of course, quite true that there is not much thought behind many AI-SF thought experiments. Collins is properly scathing about Hollywood representations of AI as a superhuman James Bond villain or sexual manipulator. Such anthropocentric depictions obscure the profoundly ‘inhuman’ nature of AI. Techno-amplifying already familiar cultural industry tropes, they are complicit with corporate promises that AI offers us a future the same as the present, only bigger and better. However, there are also other kinds of AI-SF that are far more critical in their perspective. While composing this book we have sometimes tried to categorize such works according to which of the contending left perspec- tives – sceptical, accelerationist and abyssal – they might correspond to.

Cyberpunk AI-SF has affinities with the sceptical Marxist perspective on AI, even while it extrapolates technological capacities well beyond what the sceptics would consider plausible. This is because cyberpunk fictions give an unsparing anatomization of capital in an era of intel- ligent machines. They defamiliarize the squalors of actually-existing AI-capital by projection into a future where truly shining AI fuses with grimy proletarianization to yield a glistening noir. Superbly realized in the original Blade Runner (though not so much in its visually impressive yet oppressively patriarchal Blade Runner 2049 remake), this genre 26. inhuman power continues to be embellished in recent AI-SF.

Judd Trichter’s novel Love in the Age of Mechanical Reproduction (2015) sends its hero on a grotesquely picaresque journey across a decaying Los Angeles where androids and humans coexist in a state of mounting antagonism and illicit liaison, seeking to recover the parts of his disassembled android lover: capital deepens its organic composition while relentlessly main- taining its domination over its fixed and variable components alike. Different in tone and pace, and original in its feminist thematics and domestic settings, yet sharing a similar problematic, is Andromeda Romano-Lax’s Plum Rains (2018). Set in 2029 Japan it deals with a con- frontation between a migrant Filipina worker and an intelligent machine over the care of elderly women, opening onto a complex and melancholy meditation on precarity, colonization and slavery.

Although works of this sort may end on hopeful prospects of combined android-human revolt or escape, their overall tenor is to emphasize the subjugation of AI to the brutal equation by which capital owns machines, exploits humans, and substitutes one for the other as profit dictates. A whole other species of AI-SF is, however, a haven of socialist utopianism. The possibilities of AI-based social and ecological planning underpin many of Kim Stanley Robinson’s explorations of postcapitalist futures, most explicitly in his 2312, where one of the forces enabling a break with an old earth ‘decisively under the thumb of late capitalism’ by other planetary settlements is the possibility that ‘the total annual economy of the solar system could be called out on a quantum computer in less than a second’, so that ‘supercomputers and artificial intelligences … make it possible to fully compute a non-market society’ (Robinson 2013: 125).

However, the most striking example of this genre is the late Ian Banks’s great series of ‘Culture’ novels, from Consider Phlebas (1988) to The Hydrogen Sonata (2013). In the Culture hyper-intelligent and benevolent ‘Minds’ – evolved AIs – preside over a galactic common- wealth, coexisting with a humanity that, relieved of material need and even mortality by machinic tutelage, continues an adventurous, individ- ualized and complex unfolding of its species-being. Here, in post-scarcity society, planning has been rendered redundant by plenitude – a fictional rendition of ‘fully automated luxury communism’ (Merchant 2015). Diametrically opposite to such optimistic visions, and closer to our own abyssal perspective, is a series of depictions of AI as a ‘bad novum’ – not because of any anthropocentric enmity, but rather as the systemic culmination of runaway capitalism.

The classic example is Charles introduction: ai-capital. 27 Stross’s Accelerando (2005), in which what initially seems a light-hearted tale of digital entrepreneurialism turns into a harrowing story of how AIs proceed to dismantle the solar system as raw material for their ever expanding computational, marketized network, eating the universe in a competitive race that casually discards hominids as sub-optimal, under-performing agents. What Accelerando proposes is that, contra the ecstatic visions of Kurzweil and other utopian prophets of the Singular- ity, the real meaning of such an event is likely to be that ‘The destiny of intelligent tool using life [i]s to be a stepping stone in the evolution of corporate instruments’ (Stross 2005: 240; see Shaviro 2009).

A somewhat similar recent vision comes from computer scientist and AI researcher Zachary Mason (who has worked on Amazon’s ML-powered recom- mendation systems) (Locke 2017). In his Void Star (2017) protagonists in a San Francisco beset by inequality and climate change find their personal lives effectively subsumed and assimilated by the activities of indecipherable superintelligent AIs. These AIs ‘engage with humanity only as byproducts affected by their actions, while they compute other- worldly questions of symbol manipulation’ (Locke 2017). Massive and eerie derangements of personal identity result as a billionaire plutocrat attempts to mobilize such AI instrumentally.

The truly great AI-SF is, however, perhaps that which slips across, plays with and permutates the possibilities we have schematically reviewed. Still unsurpassed in this regard (and of special interest because of the author’s evident deep engagement with Trotskyism) is Ken MacLeod’s extraordinary Fall Revolution quartet (2008; 2009), which, by adopting a ‘branching futures’ conceit, comprehends a series of AI social outcomes ranging from human-controlled AI planned-economies to hive-mind digital absorption to a de-growth abjuration of AI. A rather similar effect is achieved by the sequencing of Peter Watts’s recent Sunflowers short story and novella (2018) cycle.

In this sequence, the protagonist’s early ecstatic connection to AI yields to cynicism and then horror as she is recruited in an endless interstellar exploration journey under the direction of a narrow AI – ‘the chimp’ (Watts 2018) – because the corporate directors of this colonizing journey dare not instantiate AGI that might escape their control. For this reason, the cargo includes a cry- ogenically frozen crew summoned out of sleep at century-long intervals whenever an emergency requiring lateral thinking occurs. Yet despite this function, the humans are, it transpires, also expendable on a strict cost-benefit basis.

The most recent story in this series ends with the 28. inhuman power defeat by AI surveillance powers of an attempted human mutiny. At the close, the heroine, who has made herself complicit with the suppression to forestall the oxygen-starvation death of all her co-workers, reflects that the emergence of an autonomous AI might offer the best possibility for the crew’s escape from enslavement. Yet the overall sombre tone of Watts’s universe raises the obvious question of whether such an intelli- gence would be an ally or an enemy to its proletarian fabricators. In such instances – that is to say, at its best, rather than its worst – SF is a machine for thinking, and in the case of AI-SF, a machine for thinking about machine thinking and capitalism.

For that reason, in the following chapters we occasionally weave references to AI-SF into our analysis. CHAPTER SUMMARY In what follows we draw on the work of Marx and Marxist scholars to make our own assessment of the present state and future prospects of capital’s rendezvous with AI. Chapter 1 presents a political economic account of the current state of what we term the AI industry. It describes the main protagonists – the giant tech companies in the US, China and elsewhere, and their interaction with both state research programmes and communities of open-source AI developers.

It then goes on from current analysis to near future prognostication, suggesting that the ambitions of the great AI cor- porations point towards the establishment of AI as a new component of ‘the general conditions of production’, as a ubiquitous infrastructure, akin to the railways of the first industrial revolution or the electrical utilities of the second, on which all other forms of commodity production and circulation will come to rely. Chapter 2 takes up the issue of AI and employment from the per- spective of autonomist Marxism’s class composition theory. It argues that AI should be seen as a second wave of a cybernetic offensive waged by digital capital against its working class, a new onslaught occasioned by the 2008 economic crisis.

After reviewing some applications of AI within the social factory of advanced capitalism, we review the debate amongst futurists and economists as to whether AI will generate an imminent employment apocalypse or just a continuation of capital’s processes of job destruction and creation. Whether or not AI brings about an immediate jobs crisis, many other aspects of its deployment are likely to exert downward pressure on wages and working conditions, and it introduction: ai-capital. 29 is already precipitating an array of social struggles in and beyond the workplace. Chapter 3 challenges the (to some) reassuring assumption that capital could not survive omnipresent AI automation.

Taking its orientation from value theory, and assessing the long-term possibilities of AGI, it proposes that the humanist assumptions underpinning this belief no longer hold; Homo sapiens is not necessarily the only possible subject of capitalist proletarianization. If AI approaches or attains the horizon of singularity, the vistas that open up are not therefore those of inevitable capitalist collapse, but rather of the elevation of machine capital as a literally automatic subject autonomous from human beings. Social democratic programmes of ‘full automation now’ may therefore merely be positioning themselves as benign accomplices to this trajectory.

Our Conclusion draws out some political assumptions of the preceding analysis. While uncertainty is inescapable in thinking about AI, socialist strategies for reforming AI-capital by introducing a universal basic income and eco-modern techno-planning fail to confront the depth of the problem AI presents to projects of human emancipation. A communist orientation to AI focuses on transforming the ownership of the means of production so that real choices can be made about the adoption or abandonment of such technologies. The emergence of a new mode of production is, moreover, likely to occur under conditions of extreme social conflict and ecological disaster.

In this context, it is not only capitalism that will be inhuman, for the form of the human that emerges, if any does, from the struggle against AI-capital will not be the same as that which entered into it. 1 Means of Cognition [T]he Microsoft view is that AI needs to be included – or in Microsoft speak, ‘infused’ – in everything, from a simple word processor to a quantum computer. James Thompson (2018) THE NEW ELECTRICITY In 2016, Andrew Ng, Stanford professor, entrepreneur, former Chief Scientist at Baidu and former head of Google Brain, pronounced AI ‘the new electricity’ and argued: ‘Just as electricity transformed almost everything 100 years ago, today I actually have a hard time thinking of an industry that I don’t think AI will transform in the next several years’ (Lynch 2017).

Ng is not the only one to espouse the notion of AI as a basic utility leading to a new industrial revolution – the idea is implicit in proclamations of a ‘fourth industrial revolution’ issued by capitalist insti- tutions such as the World Economic Forum (Schwab 2017). It has also been explicitly advanced by tech guru Kevin Kelly (2014), who predicts that in the near future we will have a ‘common utility’ of ‘cheap, reliable, industrial-grade digital smartness running behind everything … Like all utilities, AI will be supremely boring, even as it transforms the Internet, the global economy, and civilization.

It will enliven inert objects, much as electricity did more than a century ago. Everything that we formerly electrified we will now ’ Companies such as Viv d.) deploy this idea in their business plans, asserting that with their AI platform ‘intelligence becomes a utility’. Predictions such as those of Ng and Kelly suggest that AI could become part of what Marx referred to as the ‘general conditions of production’ (Marx 1990: 506; 1993: 530), e. the technologies, institu- tions and practices which form the environment for capitalist production in a given place and time.

Marx spoke of infrastructure, which includes the means of communication and transport, as a significant component means of cognition. 31 of the general conditions of production. If AI becomes the new elec- tricity, it will be applied not only as an intensified form of workplace automation, but also as a basis for a deep and extensive infrastructural reorganization of the capitalist economy as such. This ubiquity of AI would mean that it would not take the form of particular tools deployed by individual capitalists, but, like electricity and telecommunications are today, it would be infrastructure – the means of cognition – presup- posed by the production processes of any and all capitalist enterprises.

As such, it would be a general condition of production. We propose the term ‘means of cognition’ – the AI-equivalent to Marx’s means of com- munication and transport – but insist that it not be conflated with the post-operaismo notion of ‘cognitive capitalism’ (Moulier-Boutang 2011), for reasons we discuss in the conclusion of this chapter. To make this argument first requires a review of the history of capital- ism’s adoption of AI, a survey of some existing and anticipated commercial applications of AI founded on the ML approach, and analysis of the con- temporary AI industry.

While the basis for accumulation in this industry is a highly advanced techno-scientific commodity, it is, like all capitalist enterprises, governed by compulsions to produce surplus-value, e. seek profit, compete, attract investment, control markets and defeat rivals through the formation of oligopolies and monopolies. We draw attention, however, to two structural features of this industry that could contribute to AI becoming a part of the general conditions of production: its lavish support and subsidization by neoliberal nation states eager to foster AI development for economic, administrative and military purposes; and the seemingly anomalous presence of a large and vigorous open-source component to AI research, in which tools and templates are distributed for free and worked on cooperatively, but are nevertheless channelled towards the platforms and priorities of AI oligopolists.

We speculate on how, in the near future, ML-enabled functions of cognition and perception could become ubiquitous via applica- tions ranging from simple chatbots up to smart cities and the Internet of Things (IoT). These examples demonstrate some ways AI could be positioned as a general condition of production. This analysis paints a picture which runs counter to post-operaismo’s humanist reconfigura- tion of the notion of the ‘general intellect’ (Marx 1993: 706) as referring to the novel capacities of a networked multitude. Contrarily, the possible future of AI as part of the general conditions of production supports Marx’s original formulation of the general intellect as capital’s accumu-

32. inhuman power lated machinic capacities, excised from social human labour. While AI development does, for the moment, depend largely on the mining and processing of data drawn from a networked multitude, the aim of such development is to attain a whole new level of automation giving capital unprecedented independence from labour. THE AI INDUSTRY AND THE OLIGOPOLISTS OF MACHINE INTELLIGENCE While corporate interest in the actual and potential uses of the new AI are manifold, ranging from retail sales to entertainment and industrial production, the actual production of AI systems is a central concern for a more limited circle of high-tech companies.

We refer to this complex as ‘the AI industry’, distinct from the broader field of commercial AI applications. While business-oriented publications continually remind us that AI will ‘revolutionize’ capitalist production (Columbus 2016), our analysis suggests that such a transformation, if it occurs, is still in its earliest phases. Instead, we see AI as one emerging industry whose influence is tied up with that of other emerging technologies and is as yet difficult to ascertain with certainty. Although business interest in AI is high, outside the AI industry this does not entail high levels of actual investment in the technology.

A 2017 survey of attendees at an applied artificial intelligence conference concluded that ‘AI adoption … remains low with the majority of major success stories coming only from the largest tech players in the industry’ (Rayo 2018). AI development first appeared as a distinct industrial sector in the 1980s. This first era of the AI industry was based around GOFAI expert systems. During this era, the AI industry consisted of a few small companies which produced systems as means of production for, and typically in cooperation with, their corporate customers. In some cases, large firms established internal AI departments to develop pro- prietary expert systems.

Such systems required a considerable degree of specialization, had extremely narrow fields of application and required a lot of labour to produce and update. While attempts were made to develop ‘generic’ expert systems which could be applied to any field, they ultimately failed (Roland and Shiman 2002: 205). The commercial craze for these systems subsided in the 1990s, but around the same time the ML approach gained traction in academia and, during the 2010s, returned AI to the commercial realm, propelled by advances in computing power means of cognition. 33 and improved learning algorithms. By 2017, The Economist (2017a) was proposing a shortlist of domains in which ML’s power to ‘sift through data to recognize patterns and make predictions without being explicitly programmed to do so’ was becoming commercially important.

It is worth surveying a few. The ML-based AI industry is much more diverse than the first era of expert systems; this is one reason why advanced capitalism has recently contracted a serious bout of AI fever. The Economist (2017a) has been enthusiastic about the prospects of targeting online advertisements and product recommendations; the creation of virtual personal assistants and of augmented reality systems; and autonomous vehicles. As of early 2019, some of these were already highly advanced, while others only incipient. Algorithmic targeting of advertisements and recommendations has been a foundation of digital Web 2.0 enterprises for over a decade.

Digital personal assistants, such as Apple’s Siri, Amazon’s Alexa or Microsoft’s Cortana, are gradually becoming commonplace. Augmented reality (AR) products, overlaying physical reality with a mesh of virtual images and information, are only beginning to be sold as commodities or dis- tributed as free vehicles for in-app purchases and data mining. Games such as Pokémon Go and other mobile apps are testing the AR waters, while further frontiers, such as medical applications, are being actively researched by companies such as Google, Apple and Microsoft. Perhaps the biggest prize for the commercial use of ML, but also its most daunting challenge, is the creation of self-driving cars and trucks, a ‘moonshot’ that has attracted leading information companies such as Google and Baidu, established auto-industry giants such as Ford, General Motors and Daimler, and upstart entrants such as Uber and Tesla, all racing to transform capitalism’s entire transportation

1 AI industry enterprises build ML technologies, often initially for use in their own business operations, but also as commodities for sale or rent, or as a ‘free’ service. They produce commodities for both of the major ‘departments’ into which Marx divided society’s total product and its total production process: (Department 1) means of production, e. commodities intended for productive consumption; (Department 2) means of subsistence, e. commodities destined for individual con- sumption (1992: 471). Some commentators on ML have suggested that, neatly corresponding with these two departments, there will be ‘two AIs’: one for business applications, the other for consumer devices (Economist 2017a).

In Department 1 we find examples like SAP’s 34. inhuman power HANA, a ML-powered cloud database platform that enables behemoths like Walmart to monitor their entire organization’s functioning in fine-grained, real-time detail (Ruth 2017), and Andrew Ng’s start-up ai (founded in 2017) which aims to totally overhaul industrial manufacturing by providing ‘AI-powered adaptive manufacturing, automated quality control, predictive maintenance, and more’ (Landing). In Department 2, examples include various consumer commodities like Amazon Home and similar devices. Marketed as a ‘smart speaker’, Home is a user voice interface to the Alexa digital personal assistant that enables a variety of home automation and organizational tasks.

AI is also found in other smart devices like phones and TVs and is also ‘given away’ as a component of free product-services such as Facebook, Twitter or YouTube where ML-based recommender systems curate timelines and give users suggestions on what to watch or listen to next. In turn, these systems gather customer data to fuel advertising revenues. However, as we will see, production of both Department 1 and Department 2 AI is often dominated by the same oligopolistic corporations, and may also be interconnected in a variety of ways, including the use of shared cloud computing facilities.

From 2015 on there has been a rapid escalation of corporate investment in AI research, venture funding of ML start-ups, and competitive hiring of AI talent as well as lots of acquisition and merger activity. Measuring the scale of this activity is difficult. According to one analysis, the AI industry had a revenue of $126 billion in 2015 and is projected to grow to $3,061 billion by 2024 (Statista 2016: 9), but another reckons worldwide spending on AI stood at only $19.1 billion in 2018, an increase of 54.2 per cent over 2017, and predicts it will reach a mere $52.2 billion by 2021 (International Data Corporation 2018).

The Economist (2017a) calculates that in 2017 companies globally spent around $21.3 billion in mergers and acquisitions related to AI – 26 times more than in 2015. While such conflicting estimates (often manifestly driven by the self-interest of AI vendors and business consultancies) are confusing, it is clear that AI has seized the imagination of advanced capital’s representatives (see also Press 2018). As The Economist (2017a) puts it, ‘Fueled by rivalry, high hopes and hype, the AI boom can feel like the first California gold ’ Corporate competition for ML experts is ferocious. One study, based on LinkedIn profile data, puts the number of PhD educated people ‘capable of working in AI research and applications’ at 22,000, with only 3,074 currently looking for work (Gagné 2018).

Demand far exceeds means of cognition. 35 supply (Economist 2017b). US information capitalists are in competi- tion both with new contenders – such as major auto companies with autonomous vehicle projects – but also now with China’s tech companies, some of which have set up subsidiaries in Silicon Valley. As hiring top talent is seen as crucial for the success of AI-capital, this competition has ‘set off a trend of firms plundering academic departments to hire professors and graduate students before they finish their degrees’ and created an atmosphere in which job fairs resemble frantic ‘Thanksgiving Black Friday sales at Walmart’ (Economist 2017b).

This competition for ML talent also means that wages are high. A recent New York Times article reports that ‘Typical I. special- ists, including both Ds fresh out of school and people with less education and just a few years of experience, can be paid from $300,000 to $500,000 a year or more in salary and company stock, according to nine people who work for major tech companies or have entertained job offers from them’ (Metz 2017b). When Google acquired DeepMind in 2014, it paid $650 million for a company of 50 employees; in 2016, the lab’s ‘staff costs’ alone, as it expanded to 400 employees, totalled $138 million, an average of $345,000 an employee.

In the light of such figures, it has been suggested that ML experts are ‘the new investment bankers’ (Shead 2017). The rewards are even higher, of course, for executives with experience managing AI projects. In a court case against Uber over ownership of autonomous car technologies, Google revealed that one of the leaders of its self-driving car division took home over $120 million in incentives before jumping ship to join their competitor. However, even fresh graduates with skills in ML may make ‘in excess of £100,000 and sometimes up to £1 million’ while still in their mid-twenties (Shead 2017).

The AI industry is international in scope. Between 2016 and 2018 it became widely recognized as a critical axis of technological competition between the United States and China, particularly given its potential for military application in an era of growing tensions. Important Chinese AI developers include its largest search engine corporation, Baidu, and ecommerce giant Alibaba (K-F. Lee 2018). Other important national sites for the AI industry include Canada, Israel and the United Kingdom. However, nearly all assessments suggest that the United States is the leading location (Jang 2017; Rapp and O’Keefe 2018; Fabian 2018).

By one estimate, which surveyed over 3,000 companies around the world involved in aspects of AI development, 40 per cent are in the 36. inhuman power US (Fabian 2018). Six, however, are preeminent: Alphabet (Google’s parent company), Amazon, Apple, Facebook, IBM and Microsoft. These companies all exemplify what Tarleton Gillespie (2010) and Nick Srnicek (2016) respectively describe as ‘platforms’ or ‘platform capital- ism’, a key feature of which is the digital gathering of big data generated by customers, be they users of search-engines, social media networks and video or music streaming services, or computer software or retail consumers.

Access to such troves of data makes platform firms favoura- ble sites for training ML systems. IBM Amongst these, IBM is in many ways an outlier, even though ‘Big Blue’ has a long record of interest in AI, stretching from its researchers’ involvement in the famous 1956 Dartmouth workshop to the triumph of its chess-playing Deep Blue over world champion Garry Kasparov in 1997 and its AI Watson, whose 2011 victory over human competitors in the television quiz show Jeopardy made it briefly the public face of the new generation of AI. Yet despite IBM’s $15 million investment in the system, Watson has subsequently had only limited commercial success.

While it has been described as ‘one of the most complete cognitive platforms available’ (Kisner, Wishnow and Ivannikov 2017: 1), and has been applied eclectically to commercial ventures in fields from fashion to telecommunications, IBM’s major emphasis was on potential uses in the highly profitable medical and health insurance sectors. In 2018, however, the company laid off many of the staff in this key division and announced it would be seeking new areas of focus. It is uncertain how far this setback was the result of technological failures and how much was due to the rigidities of IBM’s organizational practices (Strickland 2018).

IBM is likely hampered in its AI efforts due to not possessing the large proprietary pools of big data necessary for training ML systems; instead IBM has to acquire it, expensively, by buying up smaller firms engaged in medical research and data collection (Kisner, Wishnow and Ivannikov 2017: 19–20). The other major US AI producers, however, do not suffer from this problem. Alphabet (Google) Alphabet has been harvesting user data and applying it to advance their AI projects for years, first by algorithmically improving search patterns means of cognition. 37 and matching them with ad placements, and then using similar methods for categorizing, filtering and recommending video content on YouTube or predicting which apps users of its Android mobile phone operating system would purchase.

Alphabet’s Google Brain unit is widely seen as the leading corporate ML research group. Between 2014 and 2018 Google bought up no less than 12 AI-related companies (Patrizio 2018), the most notable being DeepMind, which made the ML system AlphaGo that in 2016 scored an uncanny victory over the reigning human Go world champion, thus supplanting Watson as the poster-boy for AI. Such research connects not only with Google’s algorithmic online services and its Google Home devices but also with its Waymo autonomous vehicles unit and suite of robotics-related companies it acquired in the early 2000s.

The development of AI is an endeavour fervently advocated by Google’s owners Sergey Brin and Larry Page as well as the transhumanist thinker Ray Kurzweil who is their ‘director of engineering’ (Simonite 2017); the combination of vast funds, deep expertise and ideological commitment places Google in an exceptional position in commercial AI research. Other US platform capitalists are, however, following similar paths. Facebook and Amazon Algorithmic analysis and prediction have been central to the success of Facebook in plotting the ‘social graph’ of users’ interests and interrela- tions which drives its massive online advertising revenues.

Facebook AI Research (FAIR) has four AI laboratories around the world, is active in conducting cutting-edge AI research, and has made several AI-related acquisitions, such as the company Ozlo, which builds virtual assistants (Patrizio 2018). Amazon’s development of ‘recommendations’ for customers across its escalating retail and logistics operations relies on the algorithmic analysis of vast volumes of consumer data, now increas- ingly integrated with the operations of its huge and partially robotized warehousing and order fulfilment systems. Amazon has a number of specific AI-based products, including the Echo ‘smart home’ system; its digital personal assistant, Alexa; Lex, a business version of Alexa; Polly, which turns text into speech, and Rekognition, an image recognition service.

In addition, ML permeates the suite of services in the Fulfilment by Amazon programme the retailer offers to third-party sellers. 38. inhuman power Microsoft and Apple Although of an older generation of IT companies than Google, Facebook and Amazon, Microsoft draws not only on decades of software development experience, but also on the records of interaction with millions of computer users for projects such as Cortana (the digital assistant bundled with its software), and more and less successful chatbot ventures, ranging from the catastrophic Tay (which machine-learned from online conversations to be a racist, sexist Nazi that had to be terminated with extreme prejudice) to the more innocuous teenager-imitating Zo.

Finally, Microsoft’s long-time competitor, Apple, which initially seemed the least AI-avid of the major platform capitalists, has also entered the arena in 2016–17, with the ongoing development of its digital personal assistant Siri and the creation of FaceID, a facial recognition security system for iPhone users. In 2018, it dramatically poached Google’s AI chief (Patrizio 2 Beyond the Tech Giants Around the top-rank tech corporations are clustered many smaller start-up companies that are attempting to carve out niches in spe- cialized branches of the AI industry, ranging from biotechnology, to farming, education, merchandising and surveillance (Zilis and Cham 2016; Patrizio 2018).

This scene is ever changing. It is likely that, in a pattern familiar from previous cycles of IT development, many of these companies will flare up and burn out, with the successes likely to be acquired by AI industry giants: ‘115 of 120 AI companies that exited the market in 2017 did so by acquisition’ (Patrizio 2018). Companies producing specialized hardware for AI systems also continue to appear in the shadows of dominant hardware firms like Nvidia and Intel. Venture capitalists invested $1.5 billion in hardware start-ups in 2017, twice as much as two years previously; new companies such as Cerebras, Graphcore and Cambricon have each attracted over $100 million in speculative funding.

Such start-ups ‘are racing toward one of two goals: Find a profitable niche or get acquired. Fast’ (C. Metz 2018). STATE ACTORS: AI SUPERPOWERS Another crucial actor in AI development is the state. Marx never completed a systematic study of the capitalist state, and attempts by means of cognition. 39 his followers to do so have stirred some of the most complex debates in Marxist 3 However, one function Marx clearly did ascribe to the state in capitalism was the creation of certain general conditions of production (Marx 1990: 506; 1993: 530) such as infrastructures, which could then be transferred to private capital once they became profitable ventures.

Such privatized industries in turn provide capitalist states with technological powers exercised in the name of national security against rivals: this is the dynamic of military-industrial complexes. This path of state-capital interactions is classically demonstrated by the develop- ment of digital technologies in the US. Computers and networks were incubated in the military-corporate-academic wing of the Pentagon before passing into general commercial use (Edwards 1996; Mazzucato 2013). The most celebrated, but by no means singular, case is the funding of the internet’s creation by the US Advanced Research Project Agency (ARPA). Contrary to the libertarian self-presentation of digital capital, the creation of US high-tech industries depended on state-sponsored research, subsidization and contracts for the creation of the technolo- gies which played a major role in America’s Cold War victory over its state-socialist opponents.

Such state-capital interactions in the development of digital technol- ogies were widely adopted beyond the US in the era of globalization following the end of the Cold War. While many theorists, both bourgeois and Marxist, declared the decline of the nation state, in actuality, capitalist globalization unfolded through the mediation of nation states whose activities characteristically involved the support and subsidiza- tion of the digital industries and infrastructures on which competitive participation in the world-market depended (Schiller 1999; Powers and Jablonski 2015). The so-called ‘Washington consensus’ governing glo- balization masked mounting antagonisms, in particular between a Cold War-victorious United States and the defeated ‘post-socialist’ Russia and China, which were compelled to capitulate to or compromise with US-led capitalism.

In 2008, when the Wall Street crash manifestly weakened the US as imperial hegemon, these conflicts emerged more sharply in a flare up of economic and military rivalries to which contending nation’s plans for AI are now integral. US AI development was already built on the basis of state-initiated infrastructures and technologies; leaders of AI research, such as IBM and later Google (Nesbitt 2017), have been supported by US defence-related funding: according to one report no less than 16 US government agencies 40. inhuman power fund AI development (Fabian 4 However, it was only in 2016 that the US Federal Government began formulating an overall National Artificial Intelligence R&D Strategic Plan.

The Trump administration has named AI a national priority because of ‘its role in helping the S. lead in technological innovation as well as its role in information state- craft, weaponization, and surveillance’, while the President has made it clear that he will not ‘stand in the way’ of AI-capital by burdening it with regulations about the social or employment effects of AI (Future of Life Institute 2018). Amongst a variety of measures, the US Depart- ment of Defense’s announcement that it would invest up to $2 billion over five years towards the advancement of AI was prominent: use of ML in surveillance and for the control of ‘swarming’ drones and other semi-autonomous weapons appear to be a high priority (Scharre 2018).

High-tech workers at Google and some other Silicon Valley companies have revolted against participation in military contracts – an important instance of resistance to AI we discuss in Chapter 2. Despite these protests, however, the Pentagon made clear that this was only the first phase of an ‘AI surge’ (Seligman 2018) that would include other projects such as the $10 billion to be spent over the next ten years for a special military cloud computing Joint Enterprise Defense Infrastructure (JEDI). This sudden urgency about AI on the part of the US state is attributa- ble to the emergence of a serious digital rival: China.

In 2017, the People’s Republic announced its Next Generation Artificial Intelligence Develop- ment Plan, with initiatives and goals for R&D, industrialization, talent development, education and skills acquisition, standards, regulation, ethics and security. It envisages China becoming the ‘primary’ centre for AI innovation by 2030 (Dutton 2018). Given the immense gulf in techno- logical development that separated China and the US even in 2000, this ambition seems staggering. What gives it some plausibility is both the pace of China’s economic growth since 2000 and changes in the nature of AI development. Kai-Fu Lee (2018), former head of Google China and now a champion of China’s AI programme, argues that the most recent generation of AI – ML in particular – has passed from the moment of watershed breakthroughs to one of innovative application.

Lee holds that in this scenario, China has an AI advantage in so far as it possesses large numbers of software engineers (not necessarily of superstar quality, but highly proficient), a fiercely competitive digital capital sector, and huge quantities of data gathered virtually without restraint. He suggests that the United States and China will, by 2030, constitute a state level means of cognition. 41 ‘duopoly’ in terms of control of the AI industry. While his account ends with benign hopes for cooperation, the underlying vectors of intense economic and military competition are all too visible in his account of a ‘new world order’ dominated by ‘AI Superpowers’ (K-F.

Lee 2018). Other states are understandably reluctant to acquiesce to this vision of an America-China AI duopoly. In 2018, the European Union acknowledged that fierce international competition demanded coor- dinated action for it to remain at the forefront of AI development and announced a joint ‘public-private’ programme aimed to increase invest- ments in AI research and development by at least €20 billion by 2020 (Middleton 2018). A striking demonstration of the centrality of state policy for AI development, and a major reason for European corpora- tions’ anxieties on this score, is the potential impact of the EU’s recent General Data Protection Regulation (GDPR).

This legislation places limits on corporate data gathering that are stringent by comparison with the policies of the US and China. The GDPR is criticized by business lobbyists for cramping potential ML projects – a disturbing harbinger of the potential for AI-capital to normalize and require regimes of high-surveillance governance. In an attempt to staunch a ‘brain drain’ of AI talent to super-salaried positions with US and Chinese corpo- rations, the EU also launched a multinational European AI institute – the European Lab for Learning and Intelligent Systems (ELLIS) – with centres in a number of European countries (Rankin 2018).

Beneath this common European front for AI, however, national rivalries continue to roil, with the UK, France and Germany jockeying for position as the regional AI leader (Shead 2018). The US, China and the EU are merely the largest contenders in a worldwide rush by states to attract AI-capital. Japan and Russia are also substantially subsidizing AI development; by mid-2018 more than 20 nations, ranging from New Zealand to Poland, Kenya and Tunisia, had formulated ‘artificial intelligence strategies’ (Dutton 2018). Deep anxieties underlie such mobilization. As we will see, at the level of individual cor- porations, AI-capital may tend to favour winner-take-all concentration of ownership and the creation of monopolies.

A similar dynamic may emerge at the level of international state relations. One of the more alarming features of Lee’s predictions of a ‘bipolar’ US-China dominance of AI-capital is that ‘while AI-rich countries rake in astounding profits countries that haven’t crossed a certain technological threshold will find themselves slipping backward... [into] a state of near total dependence 42. inhuman power and subservience’, frantically ‘trading market and data access’ for the use of AI facilities beyond their control (K-F. Lee 2018: Kindle Loc. 2759). This prospect, along with the yet blunter threat of AI military power, ensures that no state wants to be excluded from developing the emerging general conditions of capitalist production – and destruction.

EDGE, CLOUD AND ECONOMIES OF SCALE Essential to the AI industry, whether at the level of firm or nation, is the expensive hardware on which AI runs. Today this predominantly takes the form of ‘the cloud’ – vast, energy guzzling data centres that users can pay to access over the internet (Mosco 2014). As the availability of bandwidth and processing power have increased, the cloud has become available not only for storage, database and computing functions, but also in the past few years for cloud AI or ‘AI as a service’ (MSV 2018). AI-infused consumer devices and services send data to the cloud where the actual AI processing is done.

The cloud also enables the insertion of ML techniques (like image and voice recognition) into websites or programs, as well as the online building of ML models. The intense computational requirements for training deep ML models mean that few small companies can afford to purchase the requisite hardware and instead buy computing time from cloud providers. Tractica (2018) reports that the AI industry has seen a 300,000 times increase in computing power requirements since 2012, making cloud AI a space of aggressive competition. Amazon Web Services and Microsoft Azure have been the top performers, with Google Cloud, IBM and Baidu Cloud attempting to gain market share.

Cloud AI thus promises to make ML available to companies well beyond the inner circles of the AI industry, but already ownership of the cloud is consolidated in the hands of a ‘very select few’, with the tech giants being the dominant providers (Miller 2018). The cloud is complemented by an emerging technique called edge computing. Edge computing is an approach in which some processing is done locally on devices rather than sent to the cloud. Tractica (2018) estimates that the amount of AI edge devices shipped will increase from ‘161.4 million units in 2018 to 2.6 billion units worldwide annually by 2025’.

Edge computing offers advantages over the cloud when it comes to bandwidth, network latency issues and security. In applications such as autonomous vehicles, a bit of lag could be disastrous. In addition, keeping data on board a device offers obvious security benefits against means of cognition. 43 hackers. As computational requirements and numbers of users grow, edge computing also becomes attractive to cloud AI providers in so far as it can reduce the load on their clouds (Miller 2018). No one, however, expects the edge to replace the cloud. The two work in tandem. Rather than a decentralized disruptor of the cloud, the edge is likely to become another axis of competition for the tech giants, who are already invested heavily in it.

In sum, the AI industry is generating many interconnected commercial ventures – an ‘ecosystem’, to use a term widely adopted by those who like to naturalize capitalist activity. AI is beginning to be generalized beyond the tech industry, propelled by wide applicability, financial incentives and technological advancements. The prospects for further generaliza- tion indicate to us the plausibility of a future where capital presupposes access to these means of cognition, or, in other words, where AI becomes part of the general conditions of production. While there may be dramatic changes to the AI industry landscape over the course of this generalization, for now the tech giants occupy an apex position.

Control of cloud computing facilities, ownership of large data sets, and the wealth to hire the best from a limited pool of AI talent are some of their many advantages. These factors suggest that the domination of the information technology sector by a handful of corporate behemoths will continue in the age of AI. One likely trajectory for the AI industry is thus that which Marx described as inherent to the general law of capitalist accumulation, namely the centralization and concentration of capitalist power (1990: 777). This is acknowledged even by unequivocally pro-market observers. As The Economist (2017a) argues: ‘It seems likely that the incumbent tech groups will capture many of AI’s gains, given their wealth of data, computing power, smart algorithms and human talent, not to mention a head start on

’ And a business-oriented study enthusiastic about the commercial prospects of ML unabashedly acknowledges that there may be a ‘tradeoff between innovation and competition’: Like most software-related technologies, AI has scale economies. Furthermore, AI tools are often characterized by some degree of increasing returns: better prediction accuracy leads to more users, more users generate more data, and more data leads to better prediction accuracy. Businesses have greater incentives to build [AI] if they have more control, but along with scale economies, this may 44. inhuman power lead to monopolization. Faster innovation may benefit society from a short-term perspective but may not be optimal in the longer-term.

(Agrawal, Gans and Goldfarb 2018: 23) AI BUBBLE? The latter half of the 2010s saw increasingly widespread speculation on the emergence of an ‘AI bubble’ (Press 2018). In a 2017 report, the business consultancy Gartner gave warning: ‘As AI accelerates up the Hype Cycle,... most vendors are focused on the goal of simply building and marketing an AI-based product rather than first identifying needs, potential uses and the business value to customers’ (Hernandez 2017). It cautioned its business readers not to fall for ‘AI washing’ or exaggerated claims about the capabilities of AI systems. More luridly, a 2018 business report predicted that while AI would eventually ‘mature’ as a commercial sector, ‘the field will be littered with corpses on the way’ (Riot Research 2018).

Might the AI industry descend into another ‘AI winter’ before the technology is sufficiently generalized so as to constitute a part of the general conditions of production? The bursting of an ‘AI bubble’ and the failure of many of the recent entrants into the AI industry would not necessarily spell the end of AI-capital. The collapse of overvalued start-ups, and even large, estab- lished firms, making room for fresh entrants learning from their predecessors’ failures, has for centuries been a normal cyclical feature of capital’s innovations, from the railway and telegraph onward. This is all part and parcel of capital accumulation.

The most germane recent example is the bursting of the US com’ bubble in 2000, which laid waste to many early e-commerce attempts and, by knock-on effect, threw the telecommunications industry into crisis. For a few years this disaster stalled investment in new digital business, until the emergence of Google, Facebook and other Web 2.0 companies, who, with revised business models, joined the big survivors of the bust, such as Amazon and Microsoft, in a new and even larger wave of digital commodifica- tion. A similar crisis could be a mere bump in the road along which AI advances.

There are, however, other possibilities that might more deeply disturb booming forecasts for the AI industry. One would be the discovery of serious, endemic problems in the technology itself. Roman V. Yampolskiy), a researcher associated with the AI Safety movement, has compiled means of cognition. 45 a short inventory of failures in AI. Perhaps the most serious of these dates from 2013: ‘Object recognition neural networks saw phantom objects in particular noise ’ This refers to a well-documented tendency of ML object recognition systems to confuse, but report with very high confidence, certain abstract patterns with real-life entities such as peacocks or leopards.

This bizarre quirk appears to derive from such systems’ interpretation of radically new objects as merely extreme examples of the data sets on which they have been trained. The most immediate concern arising from this is that image-recognition systems used for military and security purposes are susceptible to malign spoofing by ‘adversarial images’ (Nguyen, Yosinski and Clune 2015). Even more troubling is the implication that neural networks are opaque to their producers, and can suddenly throw up entirely unanticipated results (Scharre 2018). Such disquiet has yet to seriously dampen AI enthusiasm. But disastrous AI errors might dent business confidence.

A series of fatal autonomous vehicle accidents in 2018 cast a pall, however temporary, over self-driving car research by reminding corporations and publics of the complex and costly liabilities such technology involves. There is also the possibility that AI may not deliver the goods that capital expects. The entire digital ‘information revolution’ has been bedevilled by the ‘productivity paradox’, a problem summed up in the sardonic observation by Robert Solow (1987) that ‘You can see the computer age everywhere but in the productivity ’ Apart from a brief period at the end of the 1990s, widespread adoption of digital tech- nologies in advanced capitalist economies has not (as of 2019) generated observable year over year productivity increases comparable with those yielded by earlier cycles of innovation.

The reasons are hotly disputed. Some economists argue that this apparent anomaly reflects problems in measuring the real significance of digital activity, or that economic gains of digitization merely need more time to manifest (Brynjolfsson, Rock and Syverson 2017). Others, such as Robert Gordon (2016), insist that the economic consequences of information technology are simply much less than those of inventions such as the automobile, electricity, urban sanitation chemicals and pharmaceuticals that powered US economic growth from 1870 to 1970. In the uneasy aftermath of the Wall Street crash of 2008, US corporate spending on new means of production, such as equipment and buildings, has been low, especially relative to the amount spent on financial and speculative activities such as dividends, stock buybacks and takeovers

46. inhuman power (Henwood 2018a). Michael Roberts (2018) proposes that ‘[p]roduc- tivity growth in all the major capitalist economies has slowed because of the failure of capitalists in most economies to step up investment in new technologies’. This picture could be changed, either by dramatic falls in the cost of AI-related technologies, by massive state subsidiza- tion of AI infrastructures, or by the threat of wage demands emerging from gradually tightening post-recession labour markets, which would incentivize automation. But there is as yet no absolute guarantee that the current wave of new AI research will actually make it out of the laboratory and into a wide transformation of business practices: the AI revolution might subside with a digitally voiced whimper.

THE GENERAL CONDITIONS OF PRODUCTION Let us suppose that capital’s current love affair with AI is not broken up by performance failures and commitment nerves. What might be the eventual offspring of this union? To elaborate the possible long-term significance of the second era of the AI industry we turn to an often-overlooked Marxian concept: the general conditions of produc- tion. What does Marx mean by this category that is mentioned in passing in Capital (1990: 472–3, 474, 505–6, 579, 652), but discussed in some depth in Grundrisse (1993: 308, 524–33, 725)? To begin, it is helpful to understand what it means that these conditions are general.

In Marx’s system, ‘general’ is almost always opposed to ‘particular’, and so it is with the general conditions of production, which he distinguished from the ‘particular conditions of production for one capitalist or another’ that are bought or produced directly by individual capitals to keep production going, and include material inputs (raw materials and intermediate goods), means of production, and labour (Marx 1993: 5 Whereas particular conditions concern the production of this or that individual capital, the general conditions of production are common to all capitals. In Grundrisse, Marx spelled out the relationship between an individual capital and the general conditions as ‘a specific relation of capital to the communal, general conditions of social production, as distinct from the conditions of a particular capital and its particular production process’ (Marx 1993: 533).

The general conditions are, therefore, ‘something that benefits (or impedes) all particular capitalist production processes’ (Kjøsen 2016: 65). Infrastructure is illustrative of the nature of the means of cognition. 47 general conditions, because roads, canals or railways ‘benefit not just a single capital, but all individual capitals in a given area’ (Kjøsen 2016: 65). With the general conditions of production, Marx thus described the general milieu in which an individual capital finds itself at a particular historical moment; it is the terrain of both class struggle and capitalist competition. Importantly, these conditions are general because they are potentially available to all individual capitals; this does not mean they are free or practically obtainable by all.

The case is different for different conditions. Something might be a commodity as well as a general condition. For example, transportation and communication – g. container shipping or the hardware required to connect to the internet – are something that all individual capitals have to pay for and are necessary for contemporary capitalist production. As long as a mode of transportation is common carriage, it counts as a general condition of production. Other conditions are free or are paid for indirectly, through taxation. For instance, any and all capitals find themselves on the world market whether they like it or not.

On the other hand, the protection of shipping lanes and the mainte- nance of transportation infrastructure are paid for by taxes. Infrastructure is but one vital component to the general conditions of production. As one of us has pointed out, the general conditions include a bewildering array of things: the means of communication and transport; the general use of buildings for production and storage; the market, e. the sphere and process of circulation; the political world order; the general state of science, technology and engineering; and, confusingly, also specific kinds of production – such as the production of machinery by machinery and the degree of automation in production – as well as the mass and velocity at which production occurs (Kjøsen 2016: 64).

What is the specific function of the general conditions of produc- tion? Marx argued, while discussing infrastructure, that ‘All general conditions of production... facilitate circulation or... make it possible... or... increase the force of production’ (1993: 530–1). A well-designed system of highways, bridges and tunnels will, of course, facilitate circu- lation in the sense of making it faster or, if built where there previously were no roads, would make circulation there possible. Speeding circu- lation makes it possible for capital to accelerate the cycle of extracting surplus-value in production and realizing it in the market (Marx 1992: 203).

Similar considerations apply to non-infrastructural general con- ditions of production; for example, increasing the degree of automation 48. inhuman power in production intensifies surplus-value extraction because machinery allows the application of physical force far exceeding that possessed by human labour alone (Marx 1990: 509). To really appreciate Marx’s argument about the function of the general conditions of production and thus how AI could function therein, it is necessary to discuss how they develop in lock-step with the mode of production itself. That is, the general conditions always refer to a specific locale and time period, meaning that the general conditions for the period of manufacture were different from the period of large-scale industry, which in turn differed from that of Fordism and so on, all the way up to a possible future AI-capitalism (Marx 1990: 505–6; Kjøsen 2016: 65–6).

Importantly, the general conditions for one period may be ‘inadequate’ or ‘unbearable fetters’ to the following one, and will thus have to be adapted or updated so that they become ‘appropriate’ to the new period; when they are appropriate, ‘the mode of production e. period] acquires an elasticity, a capacity for sudden extension by leaps and bounds’ (Marx 1990: 579; Kjøsen 2016: 65–6). That is, the velocity and volume of production can now occur at higher levels than during the previous period. How are the general conditions adapted to emergent modes of production? To sketch the possible capitalist future of AI we need to elaborate this dynamic.

The connection between the general conditions and the mode of production starts at the level of branches of industrial capital (Kjøsen 2016: 66). A revolution in production, for example through the invention of a machine, may force transformations in other branches that ‘are connected together by being separate phases of a process, and yet isolated by the social division of labour, in such a way that each of them produces an independent commodity’ (Marx 1990: 505). Any changes that cause productivity increases in terms of volume and/or speed in one branch require connected branches to adapt in order for the original branch to maintain its new level of productivity.

To illustrate this dynamic, Marx referred to the mechanical revolution in cotton-spinning that ‘called forth the invention of the gin’ because without this technology, the supply of cotton would not be able to keep up with mechanized cotton-spinning (1990: 505). With the general conditions, this dynamic is writ large; paying particular attention to the means of communication and transport, Marx argues that it was specifically the generalization of production with machinery and its resultant increase in speed and output that forced a change in the general conditions to means of cognition. 49 become appropriate to large-scale industrial production.

Thus, when particular branches of production are closely connected, ‘a revolution in terms of knowledge, technology and organization in one branch propagates throughout related branches, leading not only to growth in productivity, but also increased output, which in turn leads to new chain reactions throughout related branches of production and eventually to a revolution in the mode of production’, which thus becomes elastic in its productive capacity (Kjøsen 2016: 66–7). In addition to large-scale industry requiring an ‘immense transforma- tion’ in transportation and communications networks, it also required that machines could produce large quantities of uniform, precisely tooled machine parts: ‘Large-scale industry therefore had to take over the machine itself, its own characteristic instrument of production, and to produce machines by means of machines.

It was not till it did this that it could create for itself an adequate technical foundation and stand on its own feet’ (Marx 1990: 506). Only when machines started producing parts for machines could machinery as such become a general condition of production, meaning it was available to all individual capitals, in adequate quantity and quality. This is not to say that every individual capital must adopt a particular new technology for it to be considered part of the general conditions; it is sufficient that individual capitals have access to it on a more or less equal footing.

Competition will compel adoption; it will become part of the general conditions once it is widely used by a critical mass of individual capitals. This may be encouraged or even enabled by states, as for example in the 1990s when the US government fostered business adoption of the internet as an ‘Informa- tion Highway’ for commodity circulation (Schiller 1999). The high level of governmental interest in boosting national AI capacities suggests that states are already pushing for another revolution in the general conditions of production. Yet before we can discuss what this push might be a response to, or what form it might take, we need to outline the general conditions of production as they stand today.

THE GENERAL CONDITIONS OF PRODUCTION FOR TWENTY-FIRST-CENTURY CAPITALISM Marx sketched out how the general conditions of production for the period of manufacture were transformed into those adequate for large-scale industrial production. Since Marx’s time, the capitalist mode 50. inhuman power of production has gone through at least two other notable periods: Fordism, defined by Taylorism and the assembly line, and the period that followed it, defined primarily by ICTs and logistics, which is most commonly described as post-Fordism (Hardt and Negri 2001). The lack of consensus around how to define post-Fordism is evident in the panoply of names that have been given to the same period: digital capitalism (Schiller 1999), logistical or supply-chain capitalism (Toscano 2011; Cowen 2014; Kjøsen 2016), and cognitive capitalism (Moulier-Boutang 2011).

We recognize the amorphous quality of the term post-Fordism, and agree with many critiques made of it (see Amin 1994 for an early overview); and we particularly note that many theorists of post-Fordism have underplayed the continuities and overplayed the qualitative differences that this period has with Fordism. That being said, we also hold that it is evident that substantial changes have occurred since Fordism, such that the designation of a new period is worthwhile. Further, we suggest that we are perhaps entering a new period of the capitalist mode of production, beyond post-Fordism, which we refer to as actually-existing AI-capitalism.

This, we suggest, may be seen as a middle phase of a larger mode of cybernetic capitalism (Robins and Webster 1988; Peters, Britiz and Bulut 2009; Tiqqun 2001) which tends towards fully developed AI-capitalism. The accompanying table lays out a schematic history of capitalist modes of production and their attendant general conditions of production. Like any chart which purports to grasp complex phenomena, it oversimpli- fies and runs the risk of appearing more rigid that it is intended to. The periods listed here overlap with one another and were and still are being passed through in different ways and at different speeds in different places around the world.

Following Marx’s theorization, we propose that a revolution in one branch of the economy could provoke the necessity of widespread AI adoption such that it becomes part of the general conditions. In Chapter 2, we suggest that a large part of the excitement about AI develop- ment responds to crises encountered by capital’s globalizing search for cheap labour, and in Chapter 3 we argue that the prospect that AI may overcome the fetters to an advanced form of capital left over from a previous period – the persistent fetter of human labour – needs to be taken seriously.

But here we ask how generally available AI might be incorporated into capital’s inherently revolutionary dynamics. It is not difficult to understand why capital must rely, and how it relies, The General Conditions of Production in Historical Perspective Time Period 17th C–late 18th C Late 18th C– mid 19th C 19th C Late 19th C– late 20th C 1970s–2010s 2010s–??? Epoch of Production Mercantile Capital Manufacture Industrial Capitalism Cybernetic Capitalism Period of Production Manufacture Large-scale industry Fordism Post-Fordism Actually-existing AI-capitalism AI-capitalism Production Technologies and Organization Handicrafts, cooperation, hand tools Division of labour, hand tools Industrial machinery, division of labour Taylorism, assembly line, mass production Flexible production (mass customization), supply chain Narrow AI Type of Subsumption 1 Formal Subsumption Real Subsumption ‘Hyper-Subsumption’ 2 General Conditions of Production Canals, sail shipping, colonial markets, roads, draft animals, stage coaches Division of labour, asphalt roads, canals, sail shipping, colonial markets and system Steam power, machine tooling, production of machines by machines, railways, prime movers, river steamers, steamships, world market, telegraph, imperialism Electrical power, telegraph, radio, television, automobile, Bretton Woods (GATT, WB, IMF), world market Global and regional trade agreements (NAFTA, WTO), ICTs, networks, electronic financial markets, logistics, global supply chains, software, information, bar codes, scanning technology, container shipping and intermodal transportation, process mapping The cloud, big data, platforms, sensors, smartphones and personal computers, GPS, narrow AI, broadband internet, web AI autonomous vehicles, smart cities, digital personal assistants, increasingly advanced AI, production of AI by AI, 3D printing Notes: 1.

There are different interpretations of subsumption as either a logical or historical category. While this table adopts the latter, we do not all agree on using subsumption as a way to periodize the capitalist mode of production. The difference in interpretation of subsumption is related to how Capital is interpreted: (1) as a work depicting the historical development of capitalism (see g. Ernest Mandel 1990); or (2) as a theoretical analysis of capitalism that examines the ‘essential determinants of capitalism, those elements which must remain the same regardless of all historical variations so that we may speak of “capitalism” as such’ (Heinrich 2012: 31).

For a discussion on the historical vs. logical interpretation of subsumption, see Endnotes (2010). 2. The concept of ‘hyper-subsumption’ is similar to Stiegler’s concept of grammatization. 52. inhuman power on physical infrastructures, such as roads and ports, to overcome the fetter of space. But what fetters to capital’s valorization might generally available AI overcome? And how might AI become generally available? INFRASTRUCTURAL AI Our approach to the question of how AI might become part of the general conditions of production is informed by the ‘infrastructural turn’ in the humanities and political economy (Rossiter 2016; Cowen 2014; Steinhoff 2019a), as well as Marxist assessments of logistical and energy infrastructure (Toscano 2011; 2014; Bernes 2013; Kjøsen 2016).

Such critical approaches seek to counteract the frequent invisibility of infra- structure by showing how it is implicated in varieties of power relations. However, the particular notion of infrastructural AI comes to us directly from the representatives of AI-capital. As we have seen, commentators such as Andrew Ng and Kevin Kelly expect AI to become ubiquitous, distributed by a network infrastructure, just as electricity and internet access are distributed today. Is there a fetter to production or circula- tion, a supply bottleneck or something else that could be motivating this particular situating of AI? No less than Microsoft

d.) identifies one for us in what it terms the ‘fundamental constraint’ of human cognitive limi- tations. Framing modernity as an information explosion which escalates with the computer, Microsoft laments: ‘In the midst of this abundance of information, we’re still constrained by our human capacity to absorb ’ The notion is that the shift to a data-centric mode of production is underway and that, as Marx argued about large-scale industry, the tech industry will not be able to ‘stand on its own feet’ until it creates for itself an ‘adequate technical foundation’ (1990: 506). This foundation is infra- structural AI – the means of cognition.

The slogan under which major AI producers advance their creation of a generalized AI infrastructure for advanced capital is ‘the democra- tization of AI’ (Microsoft; Gosaduaff 2017; Gent 2018). Microsoft, Google and Amazon have all announced projects with this goal. Microsoft has announced a plan to ‘democratize Artificial Intelligence (AI), to take it from the ivory towers and make it accessible for all’. It lists the following four points, which deserve to be quoted in their entirety: We’re going to harness artificial intelligence to fundamentally change how we interact with the ambient computing, the agents, in our lives.

means of cognition. 53 We’re going to infuse every application that we interact with, on any device, at any point in time, with intelligence. We’ll make these same intelligent capabilities that are infused in our own apps – the cognitive capabilities – available to every application developer in the world. We’re building the world’s most powerful AI supercomputer and making it available to anyone, via the cloud, to enable all to harness its power and tackle AI challenges, large and small. (Microsoft d.) Microsoft imagines a world submerged in AI that is nothing short of techno-animistic: ‘As we infuse intelligence into everything, whether it’s your keyboard, your camera, or business applications, we are essen- tially teaching applications to see, to hear, to predict, to learn and take

’ Other tech giants have used the same terminology. Guy Ernest from Amazon describes Amazon Web Services as ‘democratizing AI’ by making their AI tools available for ‘any team size and skill, and for every use case’. The very same phrase has been uttered by Fei Fei Li, Google’s Chief Scientist of the Cloud and ML, and Michael Marin, a senior executive at IBM Internet of Things (Greene 2018; Simpson 2018). ‘Democratizing’ AI thus means generalizing both its deployment and the tools for creating it, making it increasingly available to end-users and allowing anyone, working in any field, even those without any AI training, to develop AI.

One of the most significant axes of the ‘democratization’ programme might seem to contradict the AI industry’s profit orientation. The AI industry is characterized by a large and vigorous open-source community, in which tools and templates for making AI are freely dis- tributed, projects are undertaken by cooperative online programming collectives, and products are released gratis for general use. Nearly all of the tech giants have open-sourced some of their AI-related materials (Simonite 2015; Crosby 2018). In 2015, Google released TensorFlow, a library of tools for deep learning programming, under the Apache 2.0 open-source licence, and it is now widely used.

Other tech giants, seeing Google’s success, followed suit. In 2017, Facebook opened several of its libraries, pre-trained models and data sets including its Caffe2 and PyTorch frameworks (Arakelyan 2017). Microsoft’s CNTK, Baidu’s Warp-CTC, and Amazon’s DSSTNE (the AI framework that powers its product recommendation system) are now all freely available and can 54. inhuman power be used to produce industry-grade AI (Stone 2016; Finley 2016). In addition, there are many other open-source AI projects not derived from the tech giants (Harvey 2017). Almost all AI projects today rely on such open-source toolkits. This is a significant milestone, we suggest, for AI becoming part of the general conditions of production.

Do such open-source projects mean AI will follow a path beyond the control of giant corporations? To answer this question, it is useful to remember the history of so-called free and open-source (FOSS) pro- gramming. Free software advocates such as Richard Stallman called for the non-commodification of software in the 1980s, but were largely defeated by the business-friendly open-source movement, championed not only by Linus Torvald, inventor of Linux, but also by the likes of Tim O’Reilly, CEO of O’Reilly Media (Liu 2018; Halliday 2018). As 2020 approaches, ‘open source’ is a buzzword for the business press and major IT corporations have shifted from seeing the open-source community as dangerously subversive to viewing it as a source of robust no-cost programming, a potential recruitment ground, and a strategic site for attracting users to their platforms (Weber 2004; Söderberg 2008; Tozzi 2017).

Indeed, some open-source projects are dominated by contribu- tions from employees of companies who use those projects. In the case of Linux, 2017 saw ‘well over 85 percent of all kernel development … done by developers who are being paid for their work’ (Corbet and Kroah-Hartman 2018: 15). The example of Google’s Android operating system illuminates open-source’s shift. Android was released in 2008 by Google on an ‘open’ basis to challenge Apple’s domination of the smartphone market. While Apple’s iOS remains exclusive to its iPhone, Android, in the hands of Samsung, has since 2017 become the globally dominant smartphone operating system.

But what has Google gained from this give-away? Initially, it was a defence to avoid Google’s search engine being cut out of an Apple-dominated mobile phone world. Subsequently, as Android itself rose to ascendancy, Google gained a widely-used platform. Google has now instituted ‘closed source creep’ in which ‘an open source base [is] paired with key proprietary apps and services’ (Amadeo 2018). Google has rendered more and more of the apps customers expect to find on an Android phone inaccessible not just to other operating services, but even to developers making versions of Android that Google does not control, such as the free Android clone Replicant.

Although ostensibly open-source, in practice Android largely operates as an annex means of cognition. 55 of Google’s larger data-harvesting operations, which in turn sustain its massive advertising revenues and training of its ML systems. In 2018, European Union antitrust regulators fined Google a record $5.1 billion for abusing its power in mobile phone markets, declaring that ‘Google has used Android as a vehicle to cement the dominance of its search engine’ (Satariano and Nicas 2018). Large corporate AI producers can thus not only coexist with, and indeed benefit from, open-source AI development, but can actually weaponize it against competitors (Vorhies 2016b).

Google’s TensorFlow can be run on competing (non-Google) clouds such as Amazon Web Services or Microsoft Azure, but this is likely due to Google being the latecomer to the cloud game. It is now vigorously attempting to take a slice of the market. It is possible that a Google dominant in AI could cajole or coerce developers onto the Google cloud, and TensorFlow could become ‘the Android of AI’ (Gershgorn 2015). Further, as the tech companies themselves admit, open sourcing software can lead to accel- erated development and improvements beyond what a single team at a company could accomplish.

As one commentator observes: Free software is good business for these companies, exactly because it allows more people to develop AI. Every big tech company is locked in a battle to gather as much AI talent as possible, and the more people flooding into the field the better. Plus, others make projects with the code that inspire new products, people outside the company find and fix bugs, and students are being taught on the software in undergrad and D. programs, creating a funnel for new talent that already know the company’s internal tools. (Gershgorn 2018) Open-source AI projects thus act as ‘on-ramps’ to the proprietorial infrastructures of large AI companies (Asay 2017).

Indeed, corporate encouragement of open-source AI is now so high as to compel critics of digital capitalism to worry not only that corporations will quash open-source AI, but also that their encouragement of it will produce malign results, such as the ‘deep fake’ pornography built with Google’s TensorFlow (Gershgorn 2018). So symbiotic have platform capitalism and open-source AI become since 2010 that both critiques may turn out to be true, with the commercially successful products of open-source developers gradually being consolidated under the aegis of large technology companies, while a shadow world of amateur dark-side AI

56. inhuman power is created on freely available corporate tools. These capital-open-source relationships typify what Paolo Virno terms ‘the communism of capitalism’ (2004: 110): corporations actively fostering a bottom-up, diversified and often free production of goods, and then harvesting this fecundity by commodifying its most successful fruits. Such a strategy is consistent with the AI industry dynamics we have already reviewed. Major AI developers, themselves the direct and indirect beneficiaries of government-supported AI research, are both supplying AI capacities to other businesses and fostering the growth of large open-source communities. For capital as a whole, this means that AI-driven business analytics, managerial tools and production automation will become increasingly available, supplied from a combi- nation of large cloud computing platforms and edge computing devices.

For end-users, apps and products will increasingly integrate AI functions, and tools for making AI will be increasingly available and easy to use. Such a ‘democratization’ of AI will be entirely consistent with the reaping of massive profits by the major oligopolists of the AI industry – just as the production of earlier generations of general conditions of production, such as railways and telecommunications, created the fortunes for the corporate producers of such infrastructural technologies. If AI becomes generally available, it will still remain under the control of these capitalist providers. THE SMART CITY, THE INTERNET OF THINGS, AMBIENT INTELLIGENCE ‘Democratization’ programmes are not the only way AI might be made generally available.

Three other topics favoured by high-tech capital illuminate this possibility from different angles: the Internet of Things (IoT), the smart city and ambient intelligence. The IoT can be simply defined as the ‘pervasive deployment of [networked] smart objects’ (Kopetz 2011: 307). While the internet is commonly understood as a technology for communication between humans, the IoT is a hypo- thetical or emerging internet comprised of machine-to-machine communications, empowered by technologies such as radio-frequency identification 6 The term IoT was coined by Kevin Ashton (2009), who believes ‘[w]e need to empower computers with their own means of gathering information, so they can see, hear and smell the world for themselves, in all its random glory.

RFID and sensor technology enable means of cognition. 57 computers to observe, identify and understand the world – without the limitations of human-entered ’ The goal of the IoT is for machines to dispense with human intermediaries and to communicate and act intelligently in the world on their own. Recapitulating Microsoft’s thesis about the fundamental constraint posed by human cognitive limitations, some analysts suggest that such a machinic population will require AI to be functionally realized: ‘the flood of data that comes from IoT devices … [has] limited value without AI technologies that are capable of finding valuable insights in the data’ (International Data Corporation 2016).

The smart city is a term lacking a consensus definition (Cocchia 2014), but visions of it provide another way of thinking about AI’s capitalist near future, focused on issues of urban development. Most concep- tions of the smart city agree on at least one criterion: the presence of ‘pervasive and ubiquitous computing and digitally instrumented devices built into the very fabric of urban environments’ (Kitchin 2014: 1). These may include sensors and cameras hooked up to various actuators and processors which, through machine-to-machine communication, auto- matically optimize traffic, maintenance, energy distribution or various other urban flows in such ways that social life is improved.

The smart city is thus a particularly urban manifestation of the IoT, but with a tacit political agenda of placing urban development increasingly in the hands of large AI-capitalists. As two critics put it, smart city discourse evinces a ‘free-floating utopianism about governance as a machine that would go of itself ’ (Sadowski and Pasquale 2015). Google is, perhaps unsur- prisingly, involved in smart city endeavours; as we discuss in Chapter 2, its project to establish a smart neighbourhood on Toronto’s waterfront was met with local resistance throughout 2018, although its fate remains uncertain. The so-called ‘democratization of AI’, the smart city and the IoT are all different ways of expressing the notion of ambient intelligence, or ‘electronic environments that are sensitive and responsive to the presence of people’ (Aarts and Encarnação 2006: 1).

The goal in the ambient intelligence paradigm is a situation in which ‘devices operate collec- tively using information and intelligence that is hidden in the network connecting the devices. Lighting, sound, vision, domestic appliances, and personal health care products all cooperate seamlessly with one another to improve the total user experience through the support of natural and intuitive user interfaces’ (Aarts and Encarnação 2006: 1). The goal is that rather than a user responding to their environment, user 58. inhuman power and environment are to be engaged in an ongoing bidirectional process of interaction (Aarts and Encarnação 2006: 11).

There will be a ‘new intelligent intermediary layer between people and systems’ (Panetta 2017) and the environment will become ‘proactive’ in its interactions with users (Aarts and Encarnação 2006: 11). Generalized ambient intelligence is one way AI could become part of the general conditions of production. If achieved, it would constitute a radical change to the technological milieu of capital, particularly if the AI which becomes ambient has ML capacities for perception and cognition. What would it be like if not only human knowledge and skills were transmuted into dead labour, but if dead labour gained the fundamental capacities for perceiving and cognizing that humans have historically monopolized?

Perception and cognition would, like electrifi- cation, become ubiquitous and mundane properties of things in general. Without any claims to predict the future, we can look at some existing AI applications that currently function as fixed capital and imagine a situation in which their use becomes part of the general conditions of production. One such application is the chatbot. Chatbots are ‘any software appli- cation that engages in a dialog with a human using natural language’ whether textual or auditory (Dale 2016: 813). As Helen Hester (2016) notes, these apps, intended to outsource to machines aspects of clerical, administrative and communications work, ‘represent, in many respects, the automation of what has been traditionally deemed to be women’s labour’ – a point manifest in the names and voices of Alexa, Cortana, Siri – technologies that ‘do gender’.

Today, chatbots are predominantly textual, but Google’s Duplex (under development as of October 2018) has demonstrated uncannily human-sounding verbal conversation skills in limited settings. Some observers have even objected to its all-too-human ‘“um” and “ah” sounds’ and demanded that it announce itself as an AI (R. Metz 2018). According to one study, 36 per cent of businesses already employ chatbots and an additional 44 per cent expect to by 2020 (Oracle 2016). Chatbots can be simple rule-based programs, but to achieve more robust functionality and ease of use, modern systems usually employ ML to learn from experience and natural language processing to converse more easily.

Google’s Duplex, for example, is an ML system built on a recurrent neural network (Leviathan and Matias 2018). Industry analysts expect that soon chatbots ‘will use AI to manage unstructured data and complex tasks’ (Panetta 2016). Chatbots are often embedded in means of cognition. 59 messenger programs such as Facebook Messenger and WeChat and are most often employed in customer service applications. In its first year of opening up chatbot functionality, Facebook had more than 33,000 active bots on Messenger (Vr 2016). The developers of Google’s Duplex hope that their creation will achieve the ‘long-standing goal of human-computer interaction’ of ‘making interaction with technology via natural conversation a reality’ (Leviathan and Matias 2018).

This neatly sums up the purpose of the chatbot: to act as a new interface that replaces previous customer-facing business elements, such as web stores and human technical support, with automated natural language conversation. The chatbot is intended to make online transactions as intuitive and simple as in-person trans- actions. Consultancy firm Gartner anticipates that ‘[i]nteracting with chatbots won’t require any particular set-up; the technology will simply understand and do as the human asks’ (Panetta 2016). Chatbots have not only found use in customer service, however. As Lebeuf, Storey and Zagalsky have shown, software developers have been creating their ‘own breed’ of chatbots to ‘reduce collaboration friction’ in the workplace (2017: 2).

As the authors demonstrate, developers use chatbots to promote the functionality of their group by coordinating schedules and tasks, promoting adherence to group norms and roles, demarcating roles, responsibilities and expertise, as well as monitoring and promoting cooperation and trust (2017: 3–4). They also use chatbots to curate large information flows and share knowledge and skills (2017: 5). In so doing, they are offloading chunks of their social activities to these bots which convert various tasks and processes into dialogic form. 2018 was a ‘milestone year’ for chatbots in terms of both technical advances and business applications (Seth 2018).

Yet, due to the domain-specific content required for each chatbot, these systems suffer from the same problem of heavy customization workload that faced GOFAI expert systems. However, research is underway on the possibil- ity of ‘bootstrapping’ or automating the production of chatbots in new fields of application using ML trained on data available in that domain (Babkin et al. 2017). If such work is successful, we can expect chatbots to proliferate further and approach the ubiquity we have posited for AI as a general condition of production. We can also consider another existing AI application which is expected to become widespread.

In 2016, Amazon opened its first branch of Amazon Go, an automated convenience store, in Seattle. As of early 60. inhuman power 2019, a tenth branch is under construction. Go, billed by Amazon as ‘just walk out shopping’, uses ‘computer vision, deep learning algorithms, and sensor fusion’ to dispense with cashiers and check-out lines (Amazon 2016). Essentially, Amazon Go relies on an array of devices for machine perception, the data from which is processed and synthesized by AI. Customers sign in to the store with the Amazon app and their actions within are tracked such that whichever items they pick up are automati- cally noted by the store, and upon walking out, their account is charged.

Despite Amazon’s automation rhetoric, as of 2019, Go stores still involve human fixers and overseers, much as autonomous vehicles do (Del Ray 2017). Melville (2017) suggests that rather than seriously delving into the low-margin grocery industry, Amazon is merely using it as a venue for testing out a new method of payment, e. a new interface for retail trans- actions. Through pervasive AI-powered machine perception, Amazon Go transforms the retail transaction from a human-to-human interac- tion, in the case of a cashier, or a human-to-machine interaction, in the case of self-checkout, into a commodity-to-machine interaction which remains invisible to the human customer.

Extrapolating from these examples, we can begin to imagine what it would be like if capacities for machine perception and cognition were generalized throughout society and thus became a general condition for production. Like electricity, AI perception and cognition could be put to many uses. In the cases discussed above, they are employed with the obvious goal of the replacement of human labour – we turn to the topic of AI effects on employment in the next chapter. But if we approach these cases from the perspective of limited human cognitive capacities as fetter posed by Microsoft, we can see how they might be more generally applied.

In the cases of AI-enabled chatbots and automated retail, AI cognition and perception are being deployed to streamline processes of social interaction by converting them into simplified, easily digestible forms administered by machines. Complex workplace social interactions among software developers and customer service transactions are thus reduced to dialogue with a chatbot which facilitates the process. In the case of Amazon Go, the retail transaction disappears altogether from human phenomenology, transformed by the integration of various data streams. In both cases AI acts as an interface which simplifies a complex situation. This is nothing new for the capitalist mode of production.

As Vincent Manzerolle and one of us (Manzerolle and Kjøsen 2015) show means of cognition. 61 in the context of near-field communication (NFC) devices, capital has enthusiastically embraced technologies which simplify and speed up transactions. Excited by AI’s capacities for further simplification and speed, the consulting firm Accenture (2017) has declared that AI will be the next user interface (UI), replacing the graphic user interface that is ubiquitous on our screens today. One of us has pointed out that analysts at Accenture deploy the unusual term ‘curation’ to describe the novelty of AI when it is considered as an interface (Steinhoff 2019a).

Accenture declares that ‘at the height of sophistication, AI orchestrates. It collab- orates across experiences and channels, often behind the scenes, to accomplish tasks. AI not only curates and acts based on its experiences, but also learns from interactions to help suggest and complete new tasks’ (Accenture 2017: 20). The vagueness of this description confirms the imagined widespread applicability of the function of curation – an intel- ligent, adaptive technique of cognitive automation. AI functions as an interface which not only represents some information to a user, but by perceiving and cognizing, actively gathers, processes, reveals and hides information before and while the user acts on it.

This is the imagined function of the ‘new intelligent layer’ posited by ambient intelligence advocates (Panetta 2017) and Kelly’s ‘cheap, reliable, industrial-grade digital smartness’ (2014). Recall that, for Marx, infrastructure functions to ‘facilitate circulation or even make it possible at all, or … increase the force of production’ (Marx 1993: 530–1), while the function of the general conditions as such is to give the mode of production an elasticity, e. a capacity for expanding the volume and velocity of production. Roads and digital networks facilitate circulation. Education systems increase the force of production by producing and distributing knowledge and ideology.