Dyer-Witheford: AI Capitalism

Interview with Nick Dyer-Witheford on book Inhuman Power: Artificial Intelligence and the Future of Capitalism

DIGILABOUR: In your book, you criticize both the liberal view and the “autonomist” perspective on artificial intelligence. I found your point very interesting. Although you recognize the value of this latter approach, what are the misconceptions about autonomy in the question of artificial intelligence?

NICK DYER-WITHEFORD: Thank you for your interest in Inhuman Power. I am only one of the authors of this book, the other two being my brilliant colleagues, Atle Mikkola Kjøsen and James Steinhoff. The comments here are hence a personal interpretation of a synthetic collaboration; my co-authors might quite legitimately emphasize different aspects of our collective work. In regard to the liberal and autonomist accounts of artificial intelligence: the liberal view is of course that artificial intelligence (AI) is just one more stage in the technological progress achieved by the market economy, an advance attributable to the creative powers of capitalism, bringing greater material prosperity, consumer convenience and all-round improvement in the human condition. Whatever problems AI may bring in the realms of, say, employment or social surveillance, they can, in this view, be corrected by education for vocational training, and minor reforms to privacy protection. The fact that the development of AI is almost entirely in the hands of oligopolistic capital, along with its state-security military and police partners, is ignored, obfuscated or just accepted as the way of the world. What is more surprising is the relative complacency with which this situation is looked on by left, or what remains of it. This complacency appears to have two sources. One the one hand, there are those who doubt the actuality of recent advances in AI, suggesting that these are heavily oversold by corporate promoters, and amount to more hype than reality. This is a view that for the moment, undoubtedly contains a large part of truth, but may well be inadequate to the long-term trajectory of capitalist techno-development.  On the other hand stands the perspective of “left accelerationists”, who are themselves AI enthusiasts, and see this technology a harbinger of a postcapitalist world, in which automation dissolves the wage-work nexus. This optic can quote in its support some impressive passages from Marx in his most technologically optimistic forecasts about the progressive force of ever-expanding powers of production. But it extinguishes the more somber side of his analysis of capitalist machinery as domination, and his flashes of nightmare critique of capital as an alien and inhuman force the drives towards the status of an “automatic subject.” It is here that the issue of autonomism comes in. The tradition of operaismo (“workerism”), from which today’s autonomist Marxism derives, was remarkable for its scathing refusal of capital’s doctrine of technological progress and its insight into of how machinery is used as a managerial weapon against the working class. However, in the transformation of operaismo to contemporary post-operaismo thought, this heterodox subversion was lost, and replaced by an upbeat insistence on possibilities of cyborg re-appropriations of computing and network technologies. Now, this is a view to which I myself contributed, so a certain amount of auto-critique is in order! But we must recognize that since the 1990s the condition of then-nascent digital capital has changed significantly, notably through the successful consolidation of the giant platform enterprises, such as Google, Amazon and Facebook, and their systematic appropriations of big data and   mobile communication, all of which set the scene for recent breakthroughs in AI technology, especially in the field of machine learning. Given this situation it seemed necessary to the three of us who authored Inhuman Power to restore a critical perspective on the machinic trajectory of capital that is now yielding new forms of AI. In the short term such a critique addresses both the intensifications in labour exploitation and in command over the entre social factory currently enabled by AI. In the mid-term, it takes seriously the likely assaults on employees in industries from transportation to call centers now being prepared by AI developers. And in the longer term, it considers the implications of a capitalist directed “singularity” that, in the name of increased efficiency and productivity, aims at the creation of nothing less than a successor-species rendering humankind obsolete. For those who hope that AI will permit a society in which humans are free from capital, it is important to remember that the obverse of this arrangement may be that capital becomes free from the human. We could be looking at a scenario, not of the growing autonomy of workers, but of the deepening autonomization of capital.

DIGILABOUR: What does it mean to consider IA as a general condition of production?

DYER-WITHEFORD: Marx’s concept of general conditions of production refers to the technologies, institutions and practices which form the environment for capitalist production in a given place and time. We emphasize that at the moment AI deployment is limited (though still wider in scope than many people imagine). Various types of “narrow” domain specific AI are, and have been used in industrial robots, search engines and social media and military and police systems. So called “general AI”—roughly speaking, AI with human, equivalent intelligence, and beyond that, super-intelligence—remains very much the stuff if science fiction. However, the commercial uses of AI are multiplying across homes and workplaces. In describing AI as a “general condition of production” we are suggesting that it may become a type of infrastructure that provides the taken-for-granted prerequisites for a new phase of capitalist development. Roads and canals and sailing ships were general conditions of production for mercantile capital; steam powered machinery, railways, steam ships and later electrical power, telegraphs, telephones, radio and television were general conditions of production of industrial capital. To say that these conditions of production are general is not to suggest that they are available free; the great railway tycoons of the nineteenth century made their millions building one of the century’s key general conditions of production. But it is to say that they become facilities foundational for all types of competitive capitalist enterprise, and hence also propel profound transformative of social life. As Andrew Ng, Stanford professor, entrepreneur, former Chief Scientist at Baidu, and former head of Google Brain pronounced in 2016, the objective of his corporate sponsors is to make AI “the new electricity.” The ambition of the great oligopolists AI developers—in the US, Google, Amazon, Microsoft, Facebook and IBM, in China, Baidu and Alibaba –is not simply to use AI to increase the efficiency of their search engines, product recommendations, warehouse operations etc. It is rather to become the vendors, largely through cloud based services, of AI capacities that other businesses, and indeed individuals, cannot do without on a day to day basis. They aim for a new instantiation of capital in which AI applications such as autonomous vehicles, chatbot personal assistants and social media agents, and an Internet of Things connecting robotic applications in industry, logistics and households, saturate everyday life. If this is achieved, it will also mark a new phase in the subsumption or envelopment of human life by capitalist techno-structures—an AI-capitalism.

DIGILABOUR: How to think about the struggles involving AI-capital? How can we reposition the scenario on artificial intelligence from “other possible worlds”?

DYER-WITHEFORD: A great deal of current discussion about AI focuses on labour market issues, in effect addressing the question, “will a robot take my job?”  Around this there has been a protracted debate between “AI apocalyptics” (mostly computer scientists) who foresee an imminent general crisis of employment caused by AI automation, with abrupt job losses across many types of work, and “business as usual” theorists (mostly mainstream economists), who insist that technological change, while destroying jobs in some sectors, always creates compensatory job opportunities in other areas of the economy. This argument is now very choreographed and predictable, though also highly speculative. We think it is quite possible that AI will increase surplus populations; render employment increasingly precarious and polarized between high waged techno-elites and low waged menial jobs; and may eventually precipitate a general crisis of employment—though this process, interacting in contradictory ways with the regular business cycle and capital’s patterns of recurrent crisis and recovery, well-mapped by Marx, may well take the protracted form of a “slow tsunami” rather than the sudden onset anticipated by AI apocalyptics. What we would emphasize, however, is that right now, in the present, there are emerging a series of conflicts over the negative effects of AI capitalism. In this regard we delineate a “heptagon of struggles”. This involves: i) the struggles of workers who are already subject to the surveillance, work intensification and wage pressures of machine-learning driven algorithmic management: these include not only the well-known examples of Amazon fulfillment-centre workers and gig-economy food-couriers or on-demand-drivers, but also the multitude of online “click-workers” involved in the actual production of machine learning systems, either as data cleaners or content moderators, ii) the protests of high-tech workers in Silicon Valley and elsewhere over their employers’ involvement in AI-production for the US military, immigration authorities or border police,  iii) the anti-surveillance movement, which has been growing since the Snowden revelations, and now confronts increased state and corporate powers provided by technologies such as machine-learning driven facial recognition systems, iv) activism against algorithmic discrimination along the lines of gender and race that have repeatedly appeared in AI systems for job hiring, pre-emptive policing, welfare monitoring and many other social activities, v) the movements contesting large AI firms plans for control of information generated by “smart cities”, which would make them major corporate arbiters over urban design and planning, vi) network defection following from this the wave of revulsion over AI engineered techniques of viral propaganda and dis and mis0information revealed in the Cambridge Analytica scandals– a scandal which, while focused on a specific nefarious electoral manipulation ultimately raises large questions about the techniques of advertisement driving capital’s entire communication systems,  vii) the generalized “techlash” against the oligopolistic powers of large information firms, who are now also the main controllers and determiners of AI development, which is now bringing to the fore issues of regulation, anti-trust legislation and even alternative forms of ownership.  None of these movements necessarily has objections to capital driven AI and machine learning as its central demand. But corporate-directed AI now functions as assort of “invisible attractor” around which these antagonisms are now forming antagonisms that may be intensified ifs sectoral or general employment crises following from intensified AI deployment appear.

DIGILABOUR: What are the blindspots in research on  capitalism and artificial intelligence?

DYER-WITHEFORD: There is no shortage of topics on which further research into AI-capitalism is required. The immediate consequences of machine-learning in workplaces and the gig-economy, is the proper subject of a new round of workers’ inquiry, now indeed under way in many quarters;  investigation into the construction of the corporate algorithms used by social media to shape the conditions of the wider social factory, research which also involves campaigns for access to information and public control of research agendas, is clearly vital; so too is closer examination of the corporate partnerships with the military and police, already a hot topic in Silicon Valley and, given the deteriorating nature of international relations and intensifying border regulation, likely to become even more prominent. However, the issue that over which I am left most curious after helping write this book is a less empirical one. It the issue of finding communist or socialist agendas for addressing AI other than of the now widely-popular left accelerationist approval for expanding the means of production devised by capital. For reasons I hope this interview has already made clear, we are skeptical about the idea that capital-developed AI can be used as a lever to bring a post-capitalist order into being –as in the now ubiquitous post-work formula for AI plus UBI (universal basic incomes). To us, this seems like a recipe for leaving an entirely disempowered proletariat still resident within a system of general commodification, and at the mercy of a capitalism now endowed with god-like powers. At the end of the book, we sketch some alternatives to this option, the absence of which might be considered a “blindspot” for contemporary radicalism. Our perspective does not close the door on possible emancipatory applications of AI, were such systems to be trained, developed and delimited within what we can shorthand as a communist order. Some of my co-authors are therefore interested in the possibilities of specifically socialist or communist transhumanism. Others of us incline more to a perspective that deviates from Marx’s allegiance to techno-modernist Prometheanism. For if the horizons of socialism or communism remain fixed on prospects of unlimited economic expansion, it becomes, I think, hard to avoid accelerationist logic. Such growth will tend towards intensified use of AI, not least to provide eco-modernist patches and fixes the problems of industrial and informational capital, such as global climate catastrophe. However, such route paradoxically leads to prospects of human self-obsolescence. The question then arises as to whether, as a counter-move, some articulation between Marxism and radical political ecology, such as that incipient in today’s “de-growth” movements, might be envisaged, something beyond the now widely-discussed (and certainly important) New Green Deal proposals. This would require a movement aimed at a global levelling of wealth, a massive program of social equalization, in combination with a regionally focused powering-down and de-celeration of the continuous economic growth indispensable to capitalism, accompanied by a deep democratization of both work institutions and scientific and technological research agendas. Such a path—and here I elaborate a personal opinion, rather than one fully worked out with and endorsed by my co-authors–might open a way to diminishing reliance on inhuman, or, more precisely, a-human, AI systems, or at least opens a space for some genuine social deliberation on the conditions of their adoption, rather than the de facto submission to the competitive automatism dictated by high-tech capital. The impulse to such a new social and ecological levelling, an articulation of equality and sufficiency, would demand an innovative social insurgency: to use some very controversial, and admittedly rather Eurocentric, examples, it would call for something like a rapprochement between Extinction Rebellion, the Gillets Jaunes and the Gillets Noirs! However, such a project, involving as it does a radical re-fit of the much of the philosophic equipment of the left is clearly one calling for further theoretical conversation and political experimentation.