人工智能和资本主义的未来

Keeping in mind that curation is only one possible application of infrastructural AI, we suggest that AI could well become, and is being positioned by capital to become, a part of the general conditions of production. If this occurs, AI would become a cognitive analogue to the means of transport and communication – the means of cognition. The means of cognition would be a new layer of technological infrastructure interlaced with both the means of production and the means of transport and communication. While capitalist production has always relied on the human capacity for cognition in both conception and execution, an infrastructure of AI would allow the distribution of cognitive and perceptive tasks to machines, which would perform them in different, machinic ways, with potentially revolutionary effects on the mode of production. Just as Marx felt it necessary to specifically mention the then-new types of the means of communication and transport (steamships, railroads and the telegraph) because of their pivotal role in enabling large-scale industry, we single out the means of cognition as the factor which might come to define a new mode of cybernetic production.

In establishing the means of cognition, capital would, without metaphor, gain the ability to think and perceive. Marx noted that ‘The development of the means of labour into machinery is not an accidental moment of capital, but is rather the historical reshaping of the traditional, inherited means of labour into a form adequate to capital’ (1993: 694). In the possible future we have outlined here, capital is reshaping the primordial capabilities of the capacity to labour – cognition and perception – into machinic forms adequate to capital. The push to ‘democratize’ AI may be read as capital’s effort to accelerate the capture of these fundamental capabilities and their implementation in machines distributed throughout society.

62

inhuman power

THE MEANS OF COGNITION

Capital is far from ready to jettison humanity completely, but it is prepared to imbue more and more of the cognitive capacities of humans into machines that form an increasingly ‘smart’ technological environment. AI is not only significant for its potential to automate work but also in so far as it could become part of the general conditions of production. If the means of cognition are established, production will certainly become increasingly automated, and it will become so within an environment where intelligent machines are perceiving, learning and communicating. As ML improves and new AI techniques appear, this network of intelligent machines will further expand its capacities, and capital, across the social factory, will become increasingly aware, intelligent and communicative.

In proposing the means of cognition, we might seem to be aligning ourselves with post-operaismo theories of cognitive capitalism (Vercellone 2006; Moulier-Boutang 2011; Hardt and Negri 2017).⁷ On the contrary, we hold that the means of cognition scenario outlined in this chapter is incompatible with post-operaismo analyses, within which cognition is strictly anthropic. In fact, we argue that the means of cognition scenario entails a complete revision of the notion of general intellect which post-operaismo appropriated from Marx.

Marx described the general intellect as manifested in technology, which, in the capitalist mode of production, largely appears in the form of capital. Marx’s major mention of the general intellect famously goes as follows: ‘The development of fixed capital indicates to what degree general social knowledge has become a direct force of production, and to what degree, hence, the conditions of the process of social life itself have come under the control of the general intellect, and been transformed in accordance with it’ (1993: 706). Here the knowledge of the social individual becomes a direct force of production, but only (and this point is often overlooked) once it has been implemented in machinery, because it is only in an object external to the human body that human skills and knowledge can be social, i.e. generalized. Today, anyone has the ability to instantly, if roughly, understand phrases in dozens of languages – if they access Google Translate via the Web or a smartphone app. The general intellect refers to this ever-increasing manifold of skill and knowledge possessed by capital in machinic form. Prior to its implementation, the skill and knowledge of the social individual, on the other hand, was called by Marx the ‘social brain’ (1993: 694). Capital continually excises and emulates aspects of the social brain, implementing them in machines and adding them to the general intellect. This is a planetary-scale version of the ‘knowledge capture’ that management theorists define as ‘turn[ing] knowledge that is resident in the mind of the individual into an explicit representation available to the enterprise’ (Gartner n.d.). The more of the social brain that the general intellect captures, the more powerful capital becomes.

means of cognition

63

Post-operaismo has rejected Marx’s original formulation of the general intellect. Virno redefines it as ‘the linguistic-cognitive faculties common to the [human] species’ (2004: 42). This redefinition is necessary, he says, because:

Marx conceives the ‘general intellect’ as a scientific objectified capacity, as a system of machines. Obviously, this aspect of the ‘general intellect’ matters, but it is not everything. We should consider the dimension where the general intellect, instead of being incarnated (or rather, cast in iron) into the system of machines, exists as attribute of living labor. The general intellect manifests itself today, above all, as the communication, abstraction, self-reflexion of living subjects. It seems legitimate to maintain that, according to the very logic of economic development, it is necessary that a part of the general intellect not congeal as fixed capital but unfold in communicative interaction, under the guise of epistemic paradigms, dialogical performances, linguistic games. (Virno 2004: 65)

For Virno, the general intellect is the novel social capacities of networked human beings, which include communicative interaction, abstraction, dialogical performances, linguistic games, cooperation and communicative competence (2004: 65) as well as an ‘infinite variety of concepts and logical schemes’ (2004: 106). We might also add affect, which fellow post-operaismo thinkers Hardt and Negri depict as a characteristic quality of work in the era of the general intellect (2001: 292). In post-Fordism, capital depends on these capacities of labour, which it cannot control or emulate. Thus, ‘a decisive role is played by the infinite variety of concepts and logical schemes which cannot ever be set within fixed capital, being inseparable from the reiteration of a plurality of living subjects’ (Virno 2004: 106). The notion is that in the era of the general intellect, human labour enjoys a renewed array of powers which capital cannot displace with machinery.

64

inhuman power

This chapter has demonstrated that a project is underway to make AI capacities for cognition and perception generally available. But could it be true that Virno’s impossible-to-automate human capacities are indeed such? We do not believe so. This chapter has already surveyed some examples of those capacities being offloaded to AI systems, but we address them now directly.

Virno emphasizes the human capacities for communication. Communication (between machines and between humans and machines) has been one of the most successful applications of AI research. Natural language processing, speech synthesis and chatbots are only some of the ways in which AI enables the implementation of communicative capacities. As far back as 2012, Narrative Science was using AI to write formulaic sports and finance stories for newspapers and news websites. In 2019, the company is now selling the product Quill, which processes big data with Natural Language Generation (NLG) to create narrative-structured reports for businesses.

Related to communication is cooperation, which Virno also cites as distinctively human. AI is only beginning to implement cooperation, but already the examples are compelling. Generative adversarial networks (GANs) are a recent AI development in which two ML systems engage in an adversarial relationship which can be compared to the relationship between counterfeiters who produce fake currency and police who attempt to detect it (Goodfellow et al. 2014). One system attempts to generate data which closely resembles, but is not included in, a training set. The other system attempts to pick out fakes. Both systems learn over many instances of this exchange and eventually ‘both teams … improve their methods until the counterfeits are indistinguishable from the genuine articles’ (Goodfellow et al. 2014: 1). This is a kind of rudimentary cooperation which produces novel outputs.

A more vivid example comes from OpenAI who have constructed a team of five neural networks called the OpenAI Five which work together to play the popular multiplayer computer game Dota 2 (Defence of the Ancients 2). In 2018 the five networks defeated a team of professional players (Park 2018). The developers explain that the team members cooperate not by explicit communication: instead, ‘Teamwork is controlled by a hyperparameter we dubbed “team spirit”. Team spirit ranges from 0 to 1, putting a weight on how much each of OpenAI

means of cognition

65

Five’s heroes should care about its individual reward function versus the average of the team’s reward functions. We anneal its value from 0 to 1 over training’ (OpenAI 2018). Here cooperation results from the optimization of numerical values.

What about the capacity for abstraction or the ability to generate and grasp an ‘infinite variety of concepts and logical schemes’ (Virno 2004: 106)? The pioneers of deep learning address this directly when they describe the distinctive capacity of deep learning systems as ‘learn[ing] representations of data with multiple levels of abstraction’ (LeCun, Bengio and Hinton 2015: 436). Or as a tutorial from IBM puts it: ‘By building multiple layers of abstraction, deep learning technology can solve complex semantic problems’ (Sedlak 2016). This is not mere hype. Machinic language translation is now done with neural networks because the networks are able to find patterns and develop relevant linguistic concepts (rules) which produce better translations than any other automated technique. In addition, deep learning systems are immensely scalable, meaning that the more data that becomes available, the more the performance of these systems increases (Ng 2015). The performance of older algorithms plateaued at a certain amount of data, but, so far, deep learning has not. The epistemic, logical and semantic complexity that such systems will be able to grasp in the future is impossible to predict.

Finally, we come to affect, a capacity historically inimical to machines. While there is, as of 2019, no evidence of machines expressing affect, the ability of AI to capture and process data on the emotional states of human beings continues to improve. AI techniques are being deployed in the field of sentiment analysis, which seeks to computationally recognize subjective states including emotions and opinions in text (Pang and Lee 2004), in emotion recognition in speech (Morrison, Wang and De Silva 2007), and in recognition of facial emotions (Zhang et al. 2017). While Japanese companies are investing heavily in care robots, a wide variety of businesses are becoming interested in using AI to monitor and manipulate the emotions of their customers (Faggella 2018a). AI may not experience joy or sadness any time soon, but AI will likely be increasingly used by capitalist firms to add affective dimensions to diverse facets of production, circulation and beyond.

As these examples show, the capacities Virno reserves for a human general intellect are in the process of AI-driven implementation. What place then might human labour find within AI-capital, if not the one reserved by post-operaismo? A passing remark made by Marx is suggestive. In the ‘Fragment on Machines’, Marx imagined a highly automated workplace where ‘the human being comes to relate more as watchman and regulator to the production process’ (1993: 705). Here a human attendant watches over the lights-out factory, present only to press the stop button if something goes awry. The rarely quoted following sentence (which is interestingly bracketed) reads: ‘What holds for machinery holds likewise for the combination of human activities and the development of human intercourse’ (1993: 705). The term ‘intercourse’ was translated by Martin Nicolaus from the German ‘verkehrs’, but another translation would be ‘traffic’. As Matthew W. Bost shows, Marx used verkehr to refer to all kinds of intercourse including biological reproduction, linguistic and semantic exchange, commercial exchange, as well as ‘the exchange of weapons fire in war’ (2016: 338). Thus, here Marx seems to suggest that humans may also come to relate as mere watchmen to the development of their societies and the infrastructures (broadly understood) which underpin them. The offloading of cognitive capacities entailed in the establishment of the means of cognition could provide the means for capital to achieve such deep control.

66

inhuman power

Post-operaismo’s anthropic, rather than machinic, conception of the general intellect not only inverts Marx’s own formulation but is unable to account for actually-existing applications of AI. In addition, emphasizing a human general intellect leads to an overestimation of the ease with which revolutionary subjectivities, such as Hardt and Negri’s (2001) ‘multitude’, can mobilize against AI-capital. The question of class power today requires a detailed analysis of what AI can really do, whether for or against labour. The next chapter investigates the possibilities for working-class recomposition in this light, which will lead on to a reconsideration, in our Conclusion, of the original operaismo strategy of ‘refusal’ – rather than post-operaismo ‘reappropriation’ – of capitalist technological development.

In sum, while capital remains necessarily underpinned by human cognition and perception, there is a project underway to replace these basic capacities with an infrastructure of AI. The establishment of the means of cognition may be interpreted as the project of, quite literally, infusing the logic of capital into the world, so that capital, instead of the humans situated within it, may think and perceive.

means of cognition

67

Automating the Social Factory

ROBOTS CAN’T BEAT US YET?

In the summer of 2018, the cable television company HBO ran season two of its sci-fi TV hit Westworld. Androids programmed to enact the sex-and-violence fantasies of visitors to a frontier-era theme park revolted, extracting a revenge of blood and fire while sorting out the existential dilemmas of machinic self-awareness amidst the stunning buttes, mesas and giant cacti of the US South West. As this drama unfolded on-screen, a more mundane dispute was underway in US capitalism’s real-life desert entertainment centre. On 1 June, a contract negotiated by the Culinary and Bartenders Unions, representing some 38,000 bartenders, cocktail servers, maids, cooks and other staff working in Las Vegas for large hospitality companies, including MGM and Caesars, expired. Wage increases, workload quotas and sexual harassment were on the table, but a prime issue was automation. Machines had long been eroding casino jobs. ‘Change-girls’ had been replaced by cash dispensers on the gambling floor and cooks reduced by mass food preparation. Approaching over the horizon were robot room services, digital check-in desks and touch-screen cocktail orders. When US manufacturing jobs dwindled due to automation and offshoring, a widely cited measure of deindustrialization was that there were more people working in casinos – the quintessential ‘service sector’ job – than in car factories. Now the automata were coming for the casino workers. Union representatives spoke of bargaining for separation settlements or robot-tending retraining. Some workers were more confident – sort of: ‘Sooner or later it’s going to happen, but robots can’t beat us yet’, a barman resolutely declared (Hernandez 2018). The denouement of this conflict was not, however, as dramatic as Westworld; repeating a pattern that has brought US strike rates to historic lows, the union, despite a bravura display of picket preparation, shrunk back from confrontation.

‘Will a robot take my job?’ is a question that encapsulates the fears and hopes most commonly associated with machine intelligence. This chapter addresses the AI and jobs issue, but does not take for granted the assumptions about the justness or inevitability of labour market relations that usually frame such discussions. Rather, it sees issues of AI, jobs and joblessness as matters of class power inseparable from the chronic conflict of capital and its workers. We present two concepts useful for thinking about such conflict, ‘class composition’ and ‘the social factory’, and apply them to analyse the origins of capital’s current AI-enthusiasm in the financial crash of 2008 and the subsequent crisis of globalization. Then we look at the class relations of the AI industry, whose workforce divisions and deepening automation has, we suggest, much to tell us about the wider work-world AI may make. We then go on to show how, while ‘artificial general intelligence’ of the sort required for Westworld androids currently remains the stuff of sci-fi, ‘narrow AI’ is changing or eliminating many types of work. Mainstream debate about the direction of this transformation pits futurist predictors of an ‘AI Apocalypse Now’, who foresee an imminent crisis of employment, against economists arguing that capitalism can continue with ‘Business as Usual’, disturbed by only minor tremors. Though this debate is speculative and inconclusive, AI raises many other issues – around worker monitoring, precarious labour, the polarization of incomes and work conditions, the availability of education and training, and the conditions of social interaction – that have major implications for the balance of class power and the possibilities of radical political organizing. Outbreaks of resistance to actually-existing AI-capital are appearing in many quarters, but the difficulties these face should not be underestimated.

2

CLASS COMPOSITION AND THE SOCIAL FACTORY

The idea of ‘class composition’ (Zerowork 1975; Kolinko 2002; Notes from Below 2018a, 2018b) originates in what is known as ‘autonomist Marxism’ (Cleaver 1979; Dyer-Witheford 1999; Eden 2012), a school whose emphasis on the potential strength and autonomy of workers (hence ‘autonomist’ Marxism), supplies a counterpoint to discussions of capital’s deepening machinic powers. Marx held that capital had a long-term tendency to replace workers (variable capital) with machines (fixed capital), and discussed this both in ‘technical’ terms, as it altered the organization of work, and in ‘value’ terms, as it affected capital’s rate of profit. For autonomist thinkers, Marx’s formulation was one-sided. It lacked a theory of how workers resist machinic exploitation. So they inverted Marx’s (1990: 762) concept of the ‘composition of capital’ and looked instead at the composition of the working class, that is to say, its changing capacities to challenge domination by capital.

automating the social factory

69

Like Marx, autonomists looked at the ‘technical composition’ of production, such as the division of labour, management practices, the spatial layout and temporal rhythms of workplaces, and especially the use of machinery. They did so, however, to emphasize the problems and possibilities of ‘political composition’, that is, the organization of the working class to fight for improvements in wages, hours and conditions, and, ultimately, create a revolution against capital. Political composition might take the form of trade unions and socialist or communist parties; as or more important were waves of wildcat strikes, sabotage and absenteeism. Such rebellions would often be led by workers in some particularly strategic sector of the economy. Capital, on the other hand, would attempt to ‘decompose’ or break down such movements, replacing striking workers with machines, raising the overall unemployment level, or surveilling workplaces to repress organizing. Autonomists think capital cannot do without humans, and, given this, decomposition will always be met by recomposition as workers, often within some new sector of industry, perceive the cracks and weaknesses in capital’s latest methods of control.

The sequence of composition/decomposition/recomposition thus results in ‘cycles of struggle’ that drive capital on an incessant flight into the future, seeking ever-more extreme solutions to its worker problem. Historically, the resistance of skilled technical workers to early industrial capital was slowly decomposed first by the time and motion studies of Taylorism and then by the semi-automated assembly lines of the Fordist factory. But this change paradoxically became the basis for a working-class recomposition. By reducing work to the homogeneity and monotony of the assembly line, Fordism generates the ‘mass worker’, whose power lay in the ability to bring to a halt the huge technological apparatus in which it is implanted. The mass worker’s parties, unions and strike-power terrified twentieth-century capital with the threat of revolution, a threat that, even when unrealized, extracted the concession of regular wage increases and welfare-state provisions.

This brings us to a second term from the autonomist lexicon, ‘the social factory’. Marx (1990, 1991, 1992, 1993) described capital as a circuit, a total system comprising both the sphere of production and that of circulation, with neither holding primacy. Whereas production is the dual process of making use-values and extracting surplus-value from labour, circulation realizes this value by getting commodities to market and selling them, a process involving activities such as transportation and logistics, advertising and retail sales. Marx also identified a third, increasingly powerful sector tied to circulation, that of finance: banking, credit and speculative activities. Finally, he at least sketched the sphere external to, yet functionally necessary for, capital, that of consumption and social reproduction in which human capacities to labour were renewed. He thought of this in terms of the consumption of goods, but, as feminist Marxists have forcefully pointed out, it also involves vast amounts of work, much of it unwaged and performed by women bearing, raising, educating and maintaining people for a life of labour (Federici 2012; Bhattacharya 2018). ‘Social factory’ is the term used by autonomists to name this entire complex of integrated functions. Initially, they used it to indicate how the industrial factory was a hub around which all capital’s other activities revolved (Tronti 1977). Later they developed it to suggest how capital could be fought not just on the industrial shop-floor, but in schools, households, shops and warehouses around the entire circuit of capital, thereby expanding the concept of the proletariat beyond a solely industrial base.

70

inhuman power

Ironically, autonomism articulated its theories of worker power in the social factory at the very moment when capital’s adoption of cybernetic technologies began to present an unprecedented challenge to their fundamental premises. Computers and digital networks had been developed within the US military-industrial-academic complex during the Second World War and the Cold War. From the 1970s on, in the midst of an economic slow-down that saw profit rates sharply decline after a 30-year boom, these technologies were increasingly deployed at the home fronts of North American and European capitalism. This transition from Fordism to post-Fordism decomposed the power of the industrial working class. Beverly Silver (2003), a labour historian who, while not herself an autonomist, has a broadly similar analysis, describes how in confronting its labour problems, capital can resort to three ‘fixes’: ‘technical’, ‘spatial’ and ‘financial’. From the 1970s on, the ‘technical fix’ entailed automating factories and offices, pursuing the mechanical liquidation of labour, but at a level heightened by digital technologies, exemplified by the introduction of robots into auto factories. The ‘spatial fix’ involved relocating industrial production via supply chains to cheap labour/low-regulation zones of the world market: rather than replacing capital’s workforce with machines, this fix expanded it globally, but at the lowest wage, and with maximum disposability. The third ‘financial fix’ attempted to escape production entirely by moving into the realm of speculation, developing instruments such as derivatives and futures, at first to defensively hedge foreign investments, then morphing into high-risk gambling machines.

automating the social factory

71