---
schema_version: "library-reader-v1"
source_type: "original_book"
title: "人工智能和资本主义的未来"
author: null
edition: null
publisher: null
year: null
isbn: "9780745338613"
library_document_id: "a2eab101-adfc-4a0d-8142-63a4d11cee02"
source_sha256: "15c6139a50a0b26b71d6ea595faa034f40b5f04c9236d1b2341bc033fc1d60fc"
pdf_physical_page_count: 225
pdf_physical_page_index_base: 0
range_start: 160
range_count: 10
html_url: "https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02?start=160&limit=10"
markdown_url: "https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02.md?start=160&limit=10"
previous_url: "https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02.md?start=150&limit=10"
next_url: "https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02.md?start=170&limit=10"
---
# 人工智能和资本主义的未来
> 页码规则：PDF_PHYSICAL_PAGE_INDEX 是从 0 开始的权威文件物理页索引；PRINTED_PAGE 是页面中识别出的印刷页码，可能为空，二者不得混用。
## PDF_PHYSICAL_PAGE_INDEX=160 | PRINTED_PAGE=152
The response from AI enthusiasts, capitalist and socialist, is that AI is precisely what is now needed to avert global warming and other environmental cataclysms. Not only does the very identification of global warming depend on advanced computing infrastructures (Edwards 2010), but the establishment of networks of sensors and monitors measuring and controlling energy, automatically orchestrating an array of clean energy sources, levelling out peak usage, coordinating surge pricing, etc., could, it is claimed, prevent, or at least moderate, climate crisis. AI will exercise a pastoral care over humanity. Such visions of high-technology environmental stewardship are part of the discourse of eco-modernity (Amblee 2018). The difficulty with this attractive vision is that AI, having apparently been converted from being part of the problem to being part of the solution, immediately threatens to turn back into the problem again because AI is a high-energy-use proposition (LePage 2018). Domestic use of electronic gadgets makes a contribution to global warming, but the great data centres crucial to AI are major heat-generating sources. Despite real advances in greening computer technology, some of the most publicized efforts, such as Google’s clean data centre policy, are based on offset credits, purchasing heat pollution rights from other companies (Geuss 2018b). As Benjamin Bratton (2016) has described, there is a real possibility that the energy expenditure required for the comprehensive machinic modelling and monitoring of global carbon emissions might actually contribute more to the heating of the planet than it would save. Underlying this problem is the paradox that if AI is being employed simultaneously to promote high economic growth societies, with their vigorous exploitation of the environment, and to mitigate that exploitation, it is running against itself. This, it seems to us, is what is envisaged by left accelerationism and fully automated luxury communism.
Wartime AI
Ecological crisis is not the only vector pushing towards a rethinking of rosy visions of high-tech socialism. In the ‘Fragment on Machines’, Marx speaks of high-technology, as it existed in his industrial age, exploding the foundations of capitalism. We take this rather literally. The cybernetic origins of AI lie in war, and so may its denouement. The rapidly expanding military application of ML and robotics in so-called New Cold Wars and wars on terror overlies the class-war dynamics we have charted in previous chapters. Discernible in what is widely termed an AI ‘arms race’ between the US and China are the not-so-faintly inscribed lines of badly fated collisions between declining and ascending great powers. This polarity, however, is only one term in a whole concatenation of shadow and hybrid conflicts waged in part with cybernetic and semi-autonomous weapons systems (Scharre 2018; Dyer-Witheford and Matviyenko 2019).
152
inhuman power
## PDF_PHYSICAL_PAGE_INDEX=161 | PRINTED_PAGE=153
The twentieth century demonstrated that the only force that can kill capital is capital itself: the proletariat is, sensu strictu, just the ‘gravedigger’. In 1917 in Russia and 1949 in China, revolution arose from inter-capitalist war. Anything that weak anti-capitalist forces might throw at capital in terms of sabotage, psychological warfare and mischief is dwarfed by what capital is launching against itself in the murky confluence of cyber-war and cyber-crime that render networks increasingly dysfunctional. Nick Land’s (2014) compelling vision of an unstoppable AI-capital ascendancy omits the possibility that the competitive dynamics of the world market result in the mutual destruction of contending cybernetic capitals. Given the tight interdependencies of cyber-war and nuclear war, this is a potentially species-fatal dynamic, but it is also potentially a revolution-generating one. The violent fragmentation of the turned-against-itself world market will produce revolts, of very varied political inflection. Communist versions will, as Bernes (2018) observes, inevitably be localized, and, we would add, likely be situated amid rapidly disintegrating networks and degrading infrastructures. This means that, if successful, they will quite possibly take control only of AI in ruins, and that they will do so only at the culmination of struggles in which AI has predominantly been in the hands of those attempting to suppress them.
Communist AI
A communist orientation to AI takes as its priority neither halting AI (Luddism) nor intensifying its development (accelerationism) but rather liquidating the structural dynamics of capital that have so far fostered its development – i.e. capital’s imperative to reduce the costs of labour-power as a factor of production and to speed up the circulation of other commodities. Whether or not there might be a ‘communist AI’ depends on whether ‘AI’ can exist outside of those conditions. From this point of view, the most promising parts of postcapitalist/left-accelerationist programmes are not those that advance the automation of work within capital, but rather those that point to the expropriation of AI-capital, the development of new forms of collective ownership of AI, and the application of AI to the collectivization of other sectors. At the moment, however, such possibilities appear only at the far end of a spectrum of discussion opened by events such as the Cambridge Analytica–Facebook scandal. Such discussions generally extend from the currently mainstreamed prospect of greater self-regulation by AI giants, to intensified government regulation (acceptable in Europe but not North America), to the remote possibility of serious trust-busting action against the concentration of ownership which is a feature of actually-existing AI-capital. On the far edges of this horizon, however, in contrast to capitalist discourse about AI as the new electricity, appear discussions of AI and its related infrastructures as a computational ‘public utility that might actually be subject to democratic control (Mosco 2017). This is the edge that should be pushed forward, towards the deeper concept of a post-revolutionary ‘communal utility’. Only where cuts in capital’s integument of material intellectual property appear are there prospects for working-class steering of AI development, for the involvement of workers and communities in determining what sorts of work should or should not be automated, and thus for a genuine determination by the ‘general intellect’ as to the design of AI – in other words, for the revival of aspirations for a digital-era equivalent to the Lucas Plan (the 1970s-era shop-stewards plan for the conversion and remaking of one of the UK’s major military-industrial corporations) (Holtwell 2018).
conclusion: communist ai
153
## PDF_PHYSICAL_PAGE_INDEX=162 | PRINTED_PAGE=154
We emphasize the importance of the ongoing conversations and arguments between libertarian commoners and state-oriented socialists in conceiving such a formation (Bauwens and Jose 2018; Murdock 2018). Within such a new form of ‘vast association’ – to cite Marx’s (1848) definition of communism – we can imagine the application of ML to a wide variety of public and communal projects in ways that erode capital’s social subsumption and replace it with new priorities. This is where there exist the possibilities of the wide (re)training of AI in logics other than that of the market, which otherwise points to a liquidation of the human as a frictional drag on profit accumulation. But we think that such a true democratization of AI should not be foreclosed by the assumption that its technologies would automatically be accelerated. Working hours might well be substantially reduced by the application of new technologies, with important gains in free time, but postcapitalist ‘work’ as a site of societal purpose and collective association would not necessarily be annihilated, cutting a swelling ‘useless class’ loose into an opioids-plus-Netflix wasteland. On the contrary, real collective decision-making about fourth industrial revolution technologies must include the possibility of roads not taken, and humans rejecting the liquidation of their activities by machinic intelligence. This would be a mode of production with something better to say than ‘the AI-generated UBI cheque is in the email’.
154
inhuman power
## PDF_PHYSICAL_PAGE_INDEX=163 | PRINTED_PAGE=155
The trajectory of AI-capital will not be diverted by the reformist measures of AI plus UBI and eco-modern climate planning, on a more-or-less painless path where it automates itself out of existence by increasing its organic composition and pulling the value rug it stands on from under its own feet. While left accelerationism and its allied schools of thought attempt a break with the rendezvous of capital and machine intelligence, they only abet that process.
To find a counter-proposition, we need something else, something akin, though not identical, to what Raniero Panzieri enunciated half a century ago, close to the start of the cybernetic era, when he articulated the doctrine of ‘refusal’ that lies at the root of what is known as autonomist Marxism: autonomist precisely in seeking, not the autonomy of capital from humans, but of humans from capital:
[T]he capitalist use of machinery is not, so to speak, a mere distortion of, or deviation from, some ‘objective’ development that is in itself rational, but that capital has determined technological development … that ‘the science, the gigantic natural forces, and the mass of social labour’ are embodied in the system of machinery, which, together with those three forces, constitutes the power of the ‘master’. Hence, vis-à-vis the ‘voided’ individual worker, technological development presents itself as a development of capitalism: as capital, and ‘because it is capital, the automatic mechanism is endowed, in the person of the capitalist, with consciousness and a will’. (Panzieri 2017)
Following from this, Panzieri postulated that ‘working-class overthrow of the system is a negation of the entire organization in which capitalist development is expressed – and first and foremost of technology insofar as it is linked to productivity’ (2017). That said, it is also true that there can be no straightforward recapitulation of operaismo’s original ‘refusal’ strategy, based as it was on the strength of the mass worker in the industrial factory; we are today in a fully social, if not global, factory and in one where the cybernetic processes that Panzieri and his comrades saw coming towards them from over an obscure horizon are now part of the relentless $2 4 / 7$ light of digital capitalism. In this context, it is quite true that the borders between refusal and reappropriation, between the saboteur, the hacker and the defector, are more fluid and complex than before. If we are sceptical of accelerationism, we also think that Walter Benjamin’s famous ‘pulling of the emergency brake’ is entering the realm of sepia-toned photo-images. The issue at this point is a series of swerves and transversal manoeuvres, as the locomotive of history goes completely off its imaginary tracks. But not all diagonal moves are equal: some lose you the game. All struggles against actually-existing AI-capital are within it, but this does not mean that the moment of negation, rather than acquiescence, can be relinquished: in the twinning of refusal and reappropriation, refusal comes first, if left adoptions of AI are not to be merely accessory to capital.2
conclusion: communist ai
155
## PDF_PHYSICAL_PAGE_INDEX=164 | PRINTED_PAGE=156
INHUMAN POWER
A fully developed AI-capitalism would realize the deepest shadows haunting Marxian thought about inhuman power. Chapter 3 explored how we may be progressing towards this future through the emergence of value-producing AGI. This can be understood as a value-theoretic or non-fetishistic analysis of what the original, and very anti-Marxist, accelerationist Nick Land (2014) meant by the ‘the teleological identity of capitalism and artificial intelligence’. Although, as we have already stressed, there are many aspects of Land’s work with which we are completely unsympathetic, his perspective on AI has the merit of being far franker than that of utopian theorists of the technological singularity such as Kurzweil. For Land (2017), AI is the culmination of a cybernetic process, not in the narrow sense of a particular doctrine of computer development, but in the sense that capital is itself a process of self-reinforcing technological advancement, a ‘positive feedback circuit’, within which
156
inhuman power
## PDF_PHYSICAL_PAGE_INDEX=165 | PRINTED_PAGE=157
commercialization and industrialization mutually excite each other in a runaway process, from which modernity draws its gradient … As the circuit is incrementally closed, or intensified, it exhibits ever greater autonomy, or automation. It becomes more tightly auto-productive (which is only what ‘positive feedback’ already says). Because it appeals to nothing beyond itself, it is inherently nihilistic. It has no conceivable meaning beside self-amplification. It grows in order to grow. Mankind is its temporary host, not its master. Its only purpose is itself. (Land 2017)
Elsewhere, Land more fully names what is at stake in the emergence of AI when he declares that if such a process is emancipatory, what it emancipates is not a ‘human species, who reaches species-being to emancipate human individuals’, but only the ‘means of production’ themselves:
so in using this word of emancipation, sure, I will totally nod along to it if what is meant by that is capital autonomization ... I’m no longer interested in ... pretending this is the same thing as what the left really means when they’re talking about emancipation. I don’t think it is. I think what the left means by emancipation is freedom from capital autonomization. (Vast Abrupt 2018)
Irrespective of the fetishistic nature of Land’s analysis, we agree with his conclusions: from our point of view, however, capitalist autonomization is what must be defeated and destroyed.
The great problem with Land’s perspective is that the advent of human-free AI-capital is not only hailed as inevitable but greeted with ‘adulation’ (Goldhill 2017) as a necessary evolutionary supersession. It is this celebratory fatalism about the emergence of inhuman superintelligence that connects Land’s writings on AI to his notorious involvement with racist and misogynist ‘neoreactionary’ strands of the alt-right. Neoreaction (or ‘NRx’) is an agenda for a futuristic and technological restoration of traditional political hierarchies of race, gender and class, and a resurgence of feudalism where corporate CEOs are the new monarchs. This political current, in which Land’s ideas mix with those of figures such as computer-scientist Curtis Yarvin (aka ‘Mencius Moldbug’), circulates widely through Silicon Valley and is supported by corporate moguls such as Peter Thiel, forming part of the cultural ambience of AI development (Burrows 2018; Sandifer and Graham 2018).
conclusion: communist ai
157
## PDF_PHYSICAL_PAGE_INDEX=166 | PRINTED_PAGE=158
The confluence between AGI research and Neoreaction is addressed by David Golumbia, whose critique of ‘computationalism’ (2009) and the attempt of much AI research to detach cognition from embodiment is influenced by and shares many of the same concerns as put forward by feminist and postcolonial theory. Golumbia (2019) argues that the flaws of AI systems in regard to race (and, we would add, gender and class) extend well beyond correctable instances of algorithmic bias. Rather, they lie in the very concept of an abstracted and technologically created ‘general intelligence’ that mirrors the mindset of a predominantly white (and male) AI research community. In this, he sees affinities between AGI research and the notoriously race-laden search for a measurable and objective ‘general IQ’. The quest for AGI is, Golumbia says, the search for a ‘Great White Robot God’.3 His analysis of the white supremacist bias of AI is, however, complicated by the emergence of China as an AI superpower. Nonetheless, Golumbia’s argument about the reactionary tendency of attempts to create a ‘singularity’ that potentially elevates the logic of the dominant social system – racist, sexist and, above all, capitalist – to a level of transcendental authority – is important; he is surely correct to identify a futurological fascist impulse in the affirmative adoption of Land’s vision of human-free capitalism by right-wing accelerationist AI developers and computer programmers. In the face of this material instantiation of neoreactionary ideas, it is important to recognize the possibilities Land names, not as a power to embrace, but as a force to oppose. Lands’ view of AI-capitalism is a history of Skynet written from the point of view of the Terminator; ours is from the perspective of Sarah Connor.
Capital is already an ‘automatic subject’, but with AGI it would also become autonomous from the labour of humans and, therefore, humanity. Capitalism could continue, but with inhuman general intelligences representing both sides of the struggle between capital and labour, one side accumulating wealth, while the other continues to work for a wage (whatever form it may take) in machinic misery. All the violent contradictions of capital could continue, but enacted by hyper-intelligent machines. Faced with an AGI that can think faster, do things faster, and which is not bound to a particular morphology with consequent biological needs, such as feeding, breathing and defecation, what would and could human workers do? And what would humans do when inhuman general intelligences started to objectify themselves in the world in a way that slowly or quickly makes the planet less and less habitable for human beings?
158
inhuman power
## PDF_PHYSICAL_PAGE_INDEX=167 | PRINTED_PAGE=159
Biological corporeality, inefficient and insufficient for valorization at machinic speed, would become an obstacle for capital to overcome: humans would have to discard it to survive. On this issue, AI-capital meets transhumanism, with its aspirations towards the transformation of human biology. Leading intellectuals of actually-existing AI-capital are explicit on this point. Kurzweil proposes that humans must ‘transcend [the] limitations of our biological bodies and brains’ (2005a: 9). Hans Moravec argues that what really matters in Homo sapiens is the mind, the ‘rest is merely jelly’, and advocates brain emulation – copying the ‘identity pattern’ and downloading it to hardware – because it enables a disembodied mind to be endowed with ‘all the advantages of machines’ (1988: 117).
Such speculative scientific proposals for transhumanizing the workforce have flowed into the corporate world. Between 2009 to 2017 Google funded the so-called Singularity University that served as a platform for the ideas of Kurzweil, who also worked for the company on ML and natural language processing (Simonite 2017). Although Google withdrew its support for the institution amidst a series of allegations of shady finances and sexual harassment (McBride 2018), ‘Singularity U’ has rapidly found other corporate sponsors, including US military-aerospace giant Boeing (Catalano 2018). Meanwhile, Elon Musk, while denouncing the dangers of runaway AI, suggests that the only answer to this threat is to implant computers into human brains so that hominids can cognitively keep pace with their machinic competitors. To this end he has invested in the company Neuralink, a neurotechnology company founded in 2017 to develop brain–computer interfaces: its aim, Musk says, is to achieve ‘symbiosis with artificial intelligence’ (Hamilton 2018).
To keep up with inhuman AGI labour, humans would have to become equally inhuman, mind and body, as immortal wage-labourers. Together with AGI workers, these no-longer-human beings would make obsolete those who decline transformation. Humans would face a choice: capitalist transhumanism or death. And this choice would generate the ultimate incarnation of Panzieri’s refusal to work, precisely because accepting individual death and species extinction would be the only alternative to working for a wage, 24/7, until the heat death of the universe.
Avoiding this choice requires a new mode of production; it therefore entails a communist revolution. But does repudiating the inhumanism of AI-led capitalist development amount then, to a last-ditch defence of classical humanism, a reaffirmation of human exceptionalism and species sovereignty? No. Communism too, should, indeed must, be inhuman. Critique is necessarily enunciated from a human perspective, but de-centring the human from humanity’s picture of the universe is both demanded by a scientific worldview and, paradoxically, is a requirement of human species survival. It is necessary to intellectually and viscerally understand humankind as bound-in with and indeed constituted in and by systems of other species and of non-living agencies, including (but not limited to) machines. All modes of production have their own anthropogenesis, thus producing different kinds of human (Read 2017). The ‘human’ that (possibly) emerges from a struggle against AI-capital will be different from the ‘human’ that went into it. We see at least two branching paths.
conclusion: communist ai
159
## PDF_PHYSICAL_PAGE_INDEX=168 | PRINTED_PAGE=160
The ‘human’ in a communist society might reappropriate transhumanism and technologically rework itself. A nascent communist society could either choose or be forced (by ecological collapse, perhaps, or the aftermath of war) to radically modify the physical form of the human, including its cognitive apparatus, metabolic system and body. Marx himself recognized that ‘man [sic] produces man’ (2007: 103). As István Mészáros noted, for Marxists there can never be ‘a point in history at which we could say: “now the human substance has been fully realized.” For such a fixing would deprive the human being of his [sic] essential attribute: his power of “self-mediation” and “self-development”’ (1970: 119). While transhumanism has origins in socialist thinkers (Bogdanov 1984; Haldane 1924; Bernal 1969), it came of age in radical libertarian circles and proliferated with little criticism of capitalism (Hughes 2012). If transhumanism is understood, as it defines itself, as an ‘intellectual and cultural movement that affirms the possibility and desirability of fundamentally improving the human condition through applied reason, especially by developing and making widely available technologies to eliminate aging and to greatly enhance human intellectual, physical, and psychological capacities’ (Humanity+ n.d.), there is, as one of the authors of this book has argued (Steinhoff 2014), nothing inherently anti-communistic about it, despite its longstanding liaison with capital. This perspective is easily associated with Marx’s humanist and high modernist moments, for as transhumanists acknowledge, their philosophy ‘can be viewed as an extension of humanism, from which it is partially derived’ (Humanity+ n.d.). Ongoing transhumanist techno-transformations could, however, raise increasingly difficult problems about the nature of the communality to which communism refers. While our respective stances on these issues differ, we all agree that Marxists should seriously engage with transhumanism, to decouple it from its blindly capitalist trajectory, reflect on Marx’s own high modernist tendencies, and delineate a social project to embrace or escape.
160
inhuman power
## PDF_PHYSICAL_PAGE_INDEX=169 | PRINTED_PAGE=161
The alternative form of communist ‘inhumanism’ is ecological. To struggle for human autonomy from capital is also to struggle for a recognition of the ecological and cosmic human enmeshments and imbrications that capital obscures and obliterates: to recognize that, in actuality, ‘we have never been autonomous’ (Nelson and Braun 2017). It deposes the fixity of the human by attending to the species’ dependence on, and imbrication in, other living systems, rather than re-centring analysis and politics upon the machines some humans have created to dominate other humans and the natural world. In this regard, capital’s AI gambit is perhaps human, all too human: communism must play otherwise. A powerful impetus to this line of thought has come from currents of ‘ecosocialism’ developed by thinkers such as John Bellamy Foster (2002) and Jason Moore (2015), with the latter more strongly representing the inhuman flattening of ontological distinctions between humanity and nature we think is necessary to challenge capital’s tendency to a machinic autonomization. We believe, to varying degrees, that these perspectives can be deepened by linkage to Jane Bennett’s analysis of the ‘vital materiality that runs through and across bodies, both human and nonhuman’ (2010: 178) and Timothy Morton’s provocative project to ‘turn up the volume of the nonhumans within Marxism’ (2017: 61). Such positions are ‘inhuman’ not, as with transhumanism, through embrace of technology, but rather in the sense pioneered by posthuman feminists (Braidotti 2018), in epistemologically and practically overthrowing both gendered and racialized definitions of humanity and the dominative concept of hominid supremacy over ecological networks.4
These intellectual and political currents provide resources for an ecologically oriented departure from the trajectory of automatic AI-capital, forming a line of flight and fight that thinks of human relations to nature, the nonhuman and other humans not in terms of domination or competition, but of mutualism and cooperation. It may be out of the combination of climate change, war and the slow erosion of wage-labour that there will appear the lineaments of a mode of production that, while it must initially formally subsume capital’s already-existing AI, goes on to really subsume it into entirely new technological configurations in which AI automation is of far less importance than the cultivation of collective and individual human habits, subjectivities, associations and practices necessary for ecological and social sustainability and species survival. In this respect, one of the most promising directions is that of a possible articulation between Marxism and ‘de-growth’ movements, such as those adumbrated in recent works by the value-theorist Anselm Jappe (2017) and the autonomist Emmanuelle Leonardi (2019). It could be in such intentionally decelerationist movements that the deep reservoirs of a ‘biocommunist’ alternative to AI-capital are established.
conclusion: communist ai
161
下一段：[继续读取](https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02.md?start=170&limit=10)