---
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: 170
range_count: 10
html_url: "https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02?start=170&limit=10"
markdown_url: "https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02.md?start=170&limit=10"
previous_url: "https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02.md?start=160&limit=10"
next_url: "https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02.md?start=180&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=170 | PRINTED_PAGE=162
AI-capital is an abyss, communism a bridge across, but a perilous, shaky one, partially in flames, and with an obscured arrival point on the other side: nonetheless, advance.
162
inhuman power
## PDF_PHYSICAL_PAGE_INDEX=171 | PRINTED_PAGE=NOT_DETECTED
Notes
INTRODUCTION
1. One of us (Kjøsen 2013a), without prior knowledge of Land’s argument, also arrived at the same conclusions through a discussion of whether androids could labour and create surplus-value.
2. For more on Land and his influence on academia, Silicon Valley and alt-right/neo-fascist/neo-reactionary thought, see Wark (2017), Goldhill (2017) and Haider (2017).
3. Throughout this book we use AI to refer to AI as a whole, including ML. Where the specific properties of ML are important to the discussion at hand we make specific reference to ML. While in most cases the existing AI systems we discuss are ML, this is not always the case. How or whether machine learning is differentiated from statistics are disputed topics we leave to experts. Regardless of one’s perspective, there are substantial areas of overlap between the two fields (Srivastava 2015; Fawcett and Hardin 2017).
4. The point deserves emphasis: what is called learning in machines is not the same as learning in humans. Nor does machine perception, or any kind of machine cognition, function the same way that analogous processes or functions do in humans.
5. Marx divides constant capital into ‘fixed’ and ‘circulating’. The former refers to buildings, tools and machines, whereas the latter refers to raw material and other inputs into the production process. Marx distinguishes between the two based on how they transfer their value to the commodity during the production process: either piecemeal (fixed capital) or wholesale (circulating) (see Marx 1992: 237–48).
CHAPTER 1
1. This is far from a complete survey of the commercial possibilities of ML. Indeed, its applications seem endless because ML’s ability to find patterns can be applied in any data rich field, which in the digital age is almost any field at all (Zilis and Cham 2016).
2. Srnicek (2016) suggests that Apple, because of its narrow and tightly controlled focus on its high-design computing products, does not really qualify as a ‘platform capitalist’ dependent on user data, but this assessment can be contested, given both its involvement with music streaming and the app-fuelled iPhone.
## PDF_PHYSICAL_PAGE_INDEX=172 | PRINTED_PAGE=164
3. For a recent review of various attempts to complete Marx’s thought on the state, see O’Kane (2014), and for discussion of the state’s role in creating and maintaining the general conditions of production, Läpple (1973).
4. However, the very fact that the privatization of initially state-led technologies is so advanced in the US creates some problems for the US government. The big-tech five, Google, Amazon, Facebook, Apple and Microsoft together spend almost as much on non-defence-related scientific research as the US federal government (Manjoo 2017). In instances where there is conflict between the US government and tech companies, as, for example, between the FBI and Apple over encryption, it is by no means certain that the state holds superior know-how.
5. When Marx discussed the particular conditions of production, he referred only to those elements of production that assumed the form of constant capital, and he gave special attention to fixed capital in particular. While this book focuses primarily on the form of fixed capital, it is important to remember that the economic form of fixed capital cannot be reduced to machinery. Buildings, tools, internal transportation networks, and facilities for storing the productive reserve and finished commodities all function as fixed capital in the same way as machinery does (Marx 1990: 510; 1992: 201).
6. Even the ostensibly human internet is increasingly travelled by machines. According to Zeifman (2017), humans only accounted for 48.2 per cent of traffic on the internet in 2016 with the rest being generated by various bots.
7. Post-operaismo thinkers have generally not directly addressed the AI issue, with the exception of Matteo Pasquinelli who has elaborated the necessity of understanding information machines and data, especially metadata, as key components of labour and capital in twenty-first-century capitalism and recognizes that capital and AI have deep and evolving affinities. His work represents an attempt to come to terms with some of the contradictions AI presents to post-operaismo thought, though he has yet to present a systematic critique. We therefore look forward eagerly to his upcoming monograph on capital and computation (2019).
CHAPTER 2
1. For a sample, see the papers from the conference ‘The Economics of Artificial Intelligence: An Agenda’, 13–14 September 2017, available from the National Bureau of Economic Research at https://www.nber.org/books/ agra-1, published in Agrawal, Gans and Goldfarb (2019).
2. Leo Impett (2018), however, argues that removing algorithmic bias from Deep Learning (DL) AI is all but impossible, as such systems automatically vectorize any and all data; in GOFAI and normal ML, it is possible to remove biases like gender, race, credit score, etc.; DL algorithms take everything into account – which is also why they are so powerful.
164
inhuman power
## PDF_PHYSICAL_PAGE_INDEX=173 | PRINTED_PAGE=165
CHAPTER 3
1. Discussion of ‘the singularity’ was popularized by Raymond Kurzweil (2005a). Competing accounts adopt different markers for its arrival. Marxian discussions of the singularity are rare, but include Rikowski (2003); Dyer-Witheford (2010); Kjøsen (2013a; 2018); Rectenwald (2013); Steinhoff (2014).
2. This chapter is an expansion and critique of the arguments one of us presented in the article ‘Träumen Androiden vom Mehrwert?’ (Kjøsen 2018).
3. HLMI may also refer to ‘high-level machine intelligence’ (Nadin 2018: 2) or can be understood as ‘Human-Level AI (HLAI) (Nilsson 2005).
4. Each mode of production has its own definition of human (Read 2017). That the definition of HMLI ties human intelligence to labour strongly suggests that anthropogenesis in the capitalist mode of production concerns a being’s capacity to labour. Indeed, Nils J. Nilsson has suggested that the Turing test be replaced by the ‘employment test’, which an AI would pass if it is ‘able to perform the jobs ordinarily performed by humans’ (2005: 68).
5. As active AGI projects, Baum includes projects that ‘either identify as AGI or conduct R&D to build something that is considered to be AGI, humanlevel intelligence, or superintelligence’ (2017: 13). Baum includes projects aimed at ‘brain emulations’ because they would be ‘computational entities with general intelligence’ and are as such ‘a type of AGI’ (2017: 8). There is, however, disagreement as to whether brain emulation falls under the umbrella of AGI (Wang and Goertzel 2007: 7). Baum’s survey is based on publicly available information which means that there are likely more active AGI projects that have yet to be made publicly known or are deliberately kept in secret.
6. Baum’s (2017: 18) breakdown is as follows: academic institutions (20); private corporations (12); public corporations (6); non-profits (5); governmental; no formal institution (2). Four of the projects are split across two of these categories, which is why these numbers add up to 49.
7. As we discuss later, technical achievement is of course by no means the same as economic deployment. For an attempt to estimate the temporal horizons of the technological singularity by an eminent mainstream economist, putting it at least over a century away, if ever, see Nordhaus (2015).
8. Marx referred to this fetish as the ‘trinity formula’ (1991: 953).
9. As raw material inputs, animals would still appear as constant capital, but of the circulating variety and, therefore, with a different social function than that of fixed (constant) capital.
10. In Moore and Aveling’s translation of Capital, both Kopf and ideell are translated as ‘imagination’ (Marx 2011: 198). While Ben Fowkes (Marx 1990) correctly translates ideell, the correct translation of Kopf is ‘head’.
11. Intuition can be linked to creativity, although the connection is tenuous (Pétervári, Osman and Bhattacharya 2016).
notes .
165
## PDF_PHYSICAL_PAGE_INDEX=174 | PRINTED_PAGE=166
12. CNNs are specific types of neural networks that are modelled on the visual cortex and well-suited to processing image data.
13. They therefore argue that the specific architecture of ANNs is significant to solving the generalization problem and needs to incorporate lateral and feedback connections rather than just resembling the feed-forward hierarchy of cells in the neocortex.
14. ImageNet is a visual database consisting of over 1 million labelled images and 20,000 categories. It was designed for research on visual object recognition software research.
15. BAYOU, which we discussed in Chapter 2, is yet another example.
16. Marx points out that ‘machinery is most valuable for capital when its value = 0’ (1993: 766).
17. The legal framework suggested by the Committee on Legal Affairs, however, is to ‘ensure that robots are and will remain in the service of humans’.
18. This bundle of goods may even be cheaper than the bundle human beings need to reproduce their labour-power. Thus AGI proletarians may emerge as a new pool of very cheap labour because AGI workers would spend a shorter part of their working day on reproducing their wage, i.e. performing necessary labour relative to surplus labour.
19. For a lucid and sympathetic treatment by a non-Marxist economist of Marx’s ‘surplus population’ thesis in relation to AI see Skidelsky (2018).
CONCLUSION
1. This, we note, is the view of AI held some of China’s Marxist scholars who believe their government is still on the road to socialism, and that under its tutelage AI will spell the end of capitalism: ‘If AI rationally allocates resources through big data analysis, and if robust feedback loops can supplant the imperfections of “the invisible hand” while fairly sharing the vast wealth it creates, a planned economy that actually works could at last be achievable’ (Xiang 2018).
2. The same can be said about a predictable attack: Luddism. As far as we are concerned the hour of Luddism has come and gone: the phrase is exhausted. The historical conditions that generated the terms and made them a signifier of seriously threatening or promising subversion have changed so much that neither the adoption nor the damnation of Luddism is of much relevance. All struggles are in and against the conditions of actually-existing AI-capitalism. In terms of destroying digital infrastructures or rendering them uninhabitable by humans, capital is doing the job itself in cyber-wars and cyber-crimes, not to mention eco-disasters, all of which however spur the development of technologies. The question is not the human sabotage of capitalist machinery, but capital’s machinic sabotage of humanity, and the possibility of a different social trajectory.
3. The philosopher Reza Negarestani has recently argued for a conception of AGI as the ultimate, emancipatory goal of philosophy and takes issue with both sceptical accounts such as Golumbia’s and laudatory ones such as Land’s, holding that ‘the project of artificial general intelligence … [is] a natural extension of the human’s process of self-discovery through which the last vestiges of essentialism are washed away’ (2018: 117).
166
. inhuman power
## PDF_PHYSICAL_PAGE_INDEX=175 | PRINTED_PAGE=167
4. Such ‘posthuman feminism’ stands in a complex relation to the accelerationist ‘xenofeminism’ (Laboria Cuboniks 2015; Hester 2018) that, building on the legacy of socialist-feminist Shulameth Firestone (1979), looks enthusiastically to cybernetic and biotechnologies as means for the ‘abolition of gender’, and can be seen as a feminist and queer version of transhumanism. The question of whether the radical anti-naturalism of xenofeminism can be reconciled with the ecological concerns of other posthuman feminisms is at this point unresolved. In this regard, the controversies surrounding Donna Haraway’s re-modulation of her famous invocation of ‘cyborg’ radicalism (1985) to her more recent posthuman ‘humus’ (i.e. earth) based politics (2016) are of particular importance, but also go beyond the scope of our study of AI.
notes
167
## PDF_PHYSICAL_PAGE_INDEX=176 | PRINTED_PAGE=NOT_DETECTED
Bibliography
Aarts, Emillie and José Luis Encarnação (2006) True Visions: The Emergence of Ambient Intelligence, New York: Springer.
Abdelrahman, Ogail (2017) ‘Deep Learning in Production and Warehousing With Amazon Robotics’, Medium, 2 May, https://medium.com/@teamrework/ deep-learning-in-production-warehousing-with-amazon-robotics-571e69 fea721
Accenture (2017) ‘Accenture Technology Vision 2017: AMPLIFYYOU’, www. accenture.com/us-en/insight-artificial-intelligence-ui
Acemoglu, Daron and Pascual Resteropo (2018) ‘Artificial Intelligence, Automation and Work’, National Bureau of Economic Research, www.nbr.org/ papers/w24196
Agrawal, Ajay, Joshua S. Gans and Avi Goldfarb (2018) Prediction Machines: The Simple Economics of Artificial Intelligence, Boston: Harvard Business Review Press.
Agrawal, Ajay K., Joshua Gans and Avi Goldfarb (eds) (2019) The Economics of Artificial Intelligence: An Agenda, Chicago: University of Chicago Press.
Ajunwa, Ifeoma, Kate Crawford and Jason Schultz (2017) ‘Limitless Worker Surveillance’, California Law Review 105:3.
Alba, Davey (2017) ‘The Hidden Laborers Training AI to Keep Ads Off Hateful YouTube Videos’, Wired, 21 April, www.wired.com/2017/04/zerochaosgoogle-ads-quality-raters
Alpaydin, Ethem (2016) Machine Learning: The New AI, Cambridge MA: MIT Press.
Amadeo, Ron (2018) ‘Google’s Iron Grip on Android: Controlling Open Source by any Means Necessary’, Ars Technica, https://arstechnica.com/gadgets/ 2018/07/googles-iron-grip-on-android-controlling-open-source-by-anymeans-necessary
Amazon (2016) ‘Introducing Amazon Go and the World’s Most Advanced Shopping Technology’, www.youtube.com/watch?v=NrmMk1Myrxc
Amazon Workers and Supporters (2018) ‘“Stop Treating Us Like Dogs!” Workers Organizing Resistance at Amazon in Poland’, in Jake Alimahomed-Wilson and Immanuel Ness (eds) Choke Points: Logistics Workers Disrupting the Global Supply Chain, London: Pluto.
Amblee, R.S. (2018) The Ugly Fight: Unleashing Artificial Intelligence Against Global Warming, New York: Gloture Books.
Amin, Ash (1994) ‘Post-Fordism: Models, Fantasies and Phantoms of Transition’, in Amin (ed.), Post-Fordism: A Reader, Oxford: Wiley-Blackwell.
## PDF_PHYSICAL_PAGE_INDEX=177 | PRINTED_PAGE=169
bibliography
169
## PDF_PHYSICAL_PAGE_INDEX=178 | PRINTED_PAGE=170
170
inhuman power
## PDF_PHYSICAL_PAGE_INDEX=179 | PRINTED_PAGE=171
bibliography
171
下一段：[继续读取](https://workshop.closeclaw.top/api/ai/read/a2eab101-adfc-4a0d-8142-63a4d11cee02.md?start=180&limit=10)