Artificial Intelligence and the Black Hole of Capitalism: A More-than-Human Political Ethology

Abstract

This paper applies a ‘more-than-human’ theoretical framework to assess artificial intelligence (AI) in the context of a capitalist economy. Case studies of AI applications from the fields of finance, medicine, commerce and manufacturing elucidate how this capitalist context shapes the aims and objectives of these innovations. The early sections of the paper set out a more-than-human theoretical perspective on capitalism, to show how the accumulation of capital depends upon free flows of commodities, money and labour, and more-than-human forces associated with supply and demand. The paper concludes that while there will be many future applications of AI, it is already in thrall to capitalist enterprise. The primary social significance of AI is that it enhances capital accumulation and a capitalist ‘black hole’ that draws more and more human activity into its sphere of influence. AI has consequent negative social, political and environmental capacities, including financial uncertainty, waste, and social inequalities. Some ways to contain and even subvert these negative consequences of an AI-fuelled capitalism are suggested.

1. Introduction

A recent paper by Henriksen and Blond (2023) threw into doubt whether artificial intelligence (hereafter, AI) can ever be turned toward human-centred ‘intelligence augmentation’ (IA), rather than simply replacing human labour or increasing workplace efficiency (Henriksen and Blond 2023, p. 740; see also, Dreyfus and Dreyfus 1988; Makridakis 2018).1 This paper delves further into Henriksen and Blond’s doubts, using a perspective that is both materialist and ‘more-than-human’ (that is, engaged with both human and non-human matter) to assess AI’s capacities in the context of global capitalism.

To this end, the premiss upon which the paper is founded is this—that until we acknowledge its social, political and economic contexts, we have no idea what (else) AI can do: what is the extent of its capacities, now and into the future? The ontology that will be used to explore AI’s capacities and their capitalist contexts is relational, post-anthropocentric and monist. The proposition that the paper advances is that to make sense of the capacities of AI, a materialist political economic analysis is required. Later sections of the paper explore this premiss, this ontology and this proposition in full—the rest of this introduction supplies a swift overview to establish their significance and relevance for efforts to glean political economic insights into AI and its capitalist context.

Concerning the premiss, the paper will suggest that what AI can do is entirely contingent upon the human and non-human matter with which it interacts (assembles): until these interactions are specified, AI’s capacities remain unknowable. So, rather than enumerating essential differences from other tools or technologies or analysing its technical features, a sociological understanding of AI requires a contextual map of its capacities: in other words, what it can do. For instance, a cursory survey of extant AI applications suggest that in healthcare, AI can aid diagnosis, predict outcomes and personalise treatment plans; in engineering, it can enable real-time analysis of large volumes of data in applications, from industrial robots to electric vehicles; in finance, it can optimise trading strategies, detect fraud and assess credit risk; it can optimise public transport systems; in education, it can personalise learning and provide intelligent tutoring; and it can power virtual assistants, chatbots, and customer service interfaces (Das et al. 2015; Lu 2019, pp. 15–20). This recognition undermines any glib assessment of AI as either inherently beneficial or harmful toward humans or ‘the environment’ (Makridakis 2017, pp. 50–51; Peters 2020, p. 28; Strickland 2019; Walton and Nayak 2021).

To explore AI’s capacities, the ontology to be applied here is relational. Congruent with the premiss, it focuses not on the essential attributes of matter (human and non-human) but upon its interactions with other materialities. It is post-anthropocentric or ‘more-than-human’ (that is, concerned with both human and non-human matter): it acknowledges the capacities of non-human matter (including AI) to affect, rather than simply being ‘used’ by, human actors as tools or technologies. And it is monist, flattening out the ontologies of the social that assert ‘another level’ acting beyond or beneath the surface of the everyday world (Karakayali 2015, pp. 741–42), as in Marxist analyses of an economic ‘base’ and a social, political and cultural ‘superstructure’, or sociological dualisms of ‘agency’ and ‘social structure’ (Karakayali 2015, pp. 741–42). Instead, there is a focus on the ‘everyday’ interactions and assemblages that not only produce daily life, but also more lasting continuities (Latour 2005, pp. 130–31) including cultural formations, social stratifications and inequalities (Fox and Alldred 2022a; Fox and Powell 2021; DeLanda 2016, pp. 14–18).

Finally, the proposition derives from the suggestion by Engster and Moore (2020, pp. 203, 212) that AI is a ‘specific capitalist mediation’—one that contributes to capitalism’s overarching objective: the accumulation of capital (see also, Berman 1992). From such a perspective, artificial intelligence and the capitalist mode of production (the dominant contemporary global economic model) are deeply interwoven or imbricated (Leonardi 2011), and thus need to be analysed together. Furthermore, a political economic analysis of AI and capitalism should be congruent with this relational, more-than-human and monist ontology (see, for example, Rella’s (2024) analysis of cryptocurrencies and blockchain technologies). The political economic framing adopted in this paper is consequently relational, more-than-human and monist, as opposed to the anthropocentrism and dualisms exhibited in classical, neo-classical and Marxist perspectives.

The paper’s early sections review current work on AI and capitalism, and then set out the more-than-human ontology and its methodological application (Fox and Alldred 2022b; Deleuze 1988, pp. 125–28) to be used in the analysis. These sections set the scene for an exposition of ‘integrated world capitalism’ (Guattari and Negri [1985] 2010, pp. 48–53) and a micropolitical exploration of these flows in real-life capitalist assemblages founded on this author’s recent work (Fox 2023a, 2023b). This framework is used to assess examples of AI from four different economic sectors, using this a more-than-human political economy, concluding with an assessment of the immediate and future challenges AI poses, and an action strategy to decouple AI from its co-option by the capitalist free market.

2. Artificial Intelligence in the Context of Capitalism

The imbrications (interweavings) of AI and capitalism have been examined variously. Some scholars have noted how capitalism has been instrumental in driving the adoption and advancement of AI technologies (Demir and Çakmak 2023, p. 197), with priorities for economic growth, efficiency and productivity, which have driven significant investment in AI research and development (Henriksen and Blond 2023, p. 739; Walton and Nayak 2021, pp. 1–2). Private enterprises, motivated by profit and competitive advantage, have harnessed AI to enhance decision-making, automate processes, and improve customer experiences (Paschen et al. 2020; Soni 2020; Verdegem 2024, pp. 1–2; Walton and Nayak 2021, p. 2). Moreover, the free-market dynamics of capitalism have created a fertile environment for AI startups to flourish, with the OECD (Tricot 2021, pp. 5–7) estimating that USD 75 billion of venture capital was invested in the AI sector in 2020, enabling entrepreneurs to develop innovative AI applications and bring them to market. The growth in AI applications has resulted in transformative technologies across various industries, including healthcare, finance, manufacturing, and transportation (Abduljabbar et al. 2019; Bawack et al. 2021; Briganti and Le Moine 2020; Paschen et al. 2020).

Others have focused on the profound economic implications of integrating AI into capitalism. On the one hand, AI has the potential to boost productivity and economic growth (Walton and Nayak 2021, p. 1). AI-powered algorithms can analyse mountains of data, supplying insights and informing decision-making that can improve resource allocation and business outcomes (Enholm et al. 2022, pp. 1714–15; MacKenzie 2017). Furthermore, by leveraging consumer data, Demir and Çakmak (2023, p. 201) suggest that AI algorithms can tailor products and services to individual preferences, enhancing customer satisfaction and driving sales. This, in turn, fuels economic activity and encourages further innovation. Meanwhile, AI can release human labour from repetitive or simple tasks, enabling more creative and strategic contributions (Robinson 2023).

On the other hand, researchers have flagged how the integration of AI within capitalism also raises concerns about job displacement and precarity (Crawford 2022, p. 219). As AI technologies advance, there is a risk of certain job roles becoming obsolete, leading to unemployment and social disruption (Peters 2020, p. 27), while broadening social inequalities (Zajko 2022). Low-skilled workers may face significant challenges in the labour market as AI systems outperform them in efficiency and accuracy (Demir and Çakmak 2023, p. 204; Robinson 2023). Furthermore, the benefits of AI adoption in capitalism may not be equally distributed. Large corporations with substantial resources can leverage AI to gain a competitive edge, consolidating their market dominance (Makridakis 2017, p. 56).

Finally, studies have identified ethical considerations, as AI integrates with capitalist production and markets. AI systems are built upon data, and the gathering and use of ‘big data’ poses issues of privacy, security, and consent (Crawford 2022, p. 224). Moreover, the potential biases embedded in databases or AI algorithms pose a challenge: perpetuating discrimination or reinforcing societal inequalities (Daneshjou et al. 2021). Efforts must be made to address these biases and ensure fairness, transparency, and accountability in AI decision-making processes (Buiten 2019, pp. 57–59) and mitigate the potential socio-economic disruptions caused by AI-driven automation (Makridakis 2017, p. 58).

These imbrications of AI and capitalism indicate a pressing need to analyse them together. However, if there were ever a need to explore the agentic capacities of non-human matter within capitalist economies and societies, the study of AI is surely it. What is missing from these anthropocentric assessments of AI and capitalism is an acknowledgment of the more-than-human capacities of AI itself. To that end, the following two sections develop such a framing for the analysis subsequently undertaken in this paper.

3. A More-than-Human Ontology of Capitalism and AI

The ‘turn to matter’ (Diener 2020, p. 45) in social theory emphasises the active agency and vitality of all matter (Fox and Alldred 2017; Bennett 2010), and that matter is relationally entangled with social, cultural, and ecological realms (Coole and Frost 2010, p. 27; Deleuze 1988, p. 123). This ‘turn’ has been variously denoted as ‘new materialist’ (Coole and Frost 2010), ‘posthumanist’ (Braidotti 2013), ‘materialist feminist’ (Alaimo and Hekman 2008) and ‘more-than-human’ (Kuby 2019). It includes actor–network theory (Latour 2005), ethology (Deleuze 1988, p. 125; Deleuze and Guattari 1988, pp. 256–58), non-representational theory (Thrift 2004) and agential realism (Barad 2007), while also acknowledging indigenous ontologies that dissolve a culture/nature dualism (Rosiek et al. 2020).

Acknowledging the relationality and vitality of matter challenges essential perspectives that assert the inherent and fixed characteristics or qualities of human and non-human matter, while prioritising human agency and subjectivity (Braidotti 2013, pp. 60–61). Further, new materialism expands the notion of relationality to encompass a broader ecological and ontological perspective, acknowledging the entanglement of human and non-human actors, cultural and natural systems, and the complex interplay between various material forces. Finally, new materialist ontology replaces the dualism in historical materialism between economic base and social/political superstructure with a ‘flat’ or monist ontology, in which there is not ‘another level’ beyond the everyday (van der Tuin and Dolphijn 2010, p. 154). These features (relationality, post-anthropocentrism and monism) supply a basis for a new materialist inquiry that seems particularly apt for the study of AI and capitalism, focusing on the more-than-human interactions of human bodies, technologies, commodities and money.2 This ontology requires an appropriate methodological framework within which to conduct such an exploration. Here, such a methodology is supplied via the ‘ethological’ toolkit first set out by Deleuze (1988) and later applied by Deleuze and Guattari (1984, 1988), and, more recently, by social science scholars to explore topics including ageing (Cluley et al. 2023), education (Mulcahy and Martinussen 2023), environment and climate change (Fox 2022), health (Duff 2014; Potts 2004), infrastructure (Bennett 2010), sexualities (Fox and Alldred 2013; Alldred and Fox 2015), social inequalities (Fox and Powell 2021; Fox and Alldred 2022a), and well-being (McLeod 2017). Ethology enables analysis of the breadth of interactions between human and non-human matter, ensuring that the contexts of these interactions are recognised (Coole and Frost 2010, p. 28). It also replaces a dualism between economic base and social/political superstructure with a ‘flat’, immanent, or monist ontology, in which there is not ‘another level’ beyond the everyday (Deleuze and Guattari 1988, pp. 20–21; 1994, pp. 48–50).

The ethological toolkit (Malins 2004, p. 84) comprises four concepts: affect, assemblage, capacity and micropolitics. Ethology explores the relationships between bodies (human and non-human) and their immediate contexts, focusing on matter’s affects, meaning its capacities to affect or be affected. Affect is a pre-personal, non-representational force that operates independently of intentionality (Deleuze and Guattari 1988, p. 257). Conceptually, it signifies the potential for becoming, change, and transformation, as bodies affect or are affected (Deleuze and Guattari 1988, p. 256), and emphasises relational capacities rather than fixed attributes (Deleuze 1988, p. 126). It is consequently the sole means by which materialities interact and make new connections, and is also the basis for the operation of power and resistance in social encounters (Fox and Alldred 2018).

The affects between materialities establish the more-than-human assemblages (Deleuze and Guattari 1988, p. 22) that shape the actions, interactions and becomings of the material and social world. Assemblages are not fixed structures, but fluid and contingent arrangements of disparate materialities; ‘a series of heterogeneous elements that are organised and held together through temporary relations’ (McLeod 2017, p. 16). They may be very short-lived or longer-lasting, and supply the context within which matter gains its capacities: what it can do. As Deleuze and Guattari (1988, p. 257) note, until we can list its capacities, we know nothing about a body (human or non-human)—what its affects are, and how it may enter into assemblage with other human and non-human matter. It follows that a focus upon relational capacities is key to ethological inquiry.

Finally, micropolitics (Deleuze and Guattari 1988, p. 216) explores the ways in which power operates within assemblages, shaping what bodies can do, feel and desire. Micropolitically, assemblages produce particular capacities, contingent upon the affects between the assembled materialities. Deleuze and Guattari described micropolitical ‘territorialisations’ (Deleuze and Guattari 1988, pp. 88–89), which specify what matter can do in specific contexts (for instance, specifying a piece of metal as a ‘tool’ or a ‘weapon’), and ‘de-territorialisations’, which generalise these capacities, freeing matter from physical, psychological or social limits or constraints (Deleuze and Guattari 1988, p. 277)—opening up new possibilities for what that piece of metal (or any other human or non-human materiality) may do. By exploring assemblage micropolitics, it is consequently possible to explore both the forces that shape human bodies and societies and identify the potential for change that exists in the everyday practices and interactions of matter.

These four concepts operationalise new materialist ontology, and, later in this paper, will provide a methodology of inquiry to analyse AI applications as ‘machines’ (Deleuze and Guattari 1988, p. 511) that gain their affects, capacities and micropolitics when assembling with other human and non-human matter (Wellner 2022, p. 8). However, this conceptual toolkit also permits ethological analysis of capitalism itself, supplying a perspective on what capitalism actually does. The following section develops this assessment of capitalism, drawing upon Deleuze and Guattari’s (1988) analysis and Fox’s (2023a, 2023b) ethological explorations of capitalist assemblages, and thereby establishing the basis from which this paper can ask: what AI can do?

4. Integrated World Capitalism, the Capitalist Axiomatic and the Black Hole of Capitalism

Deleuze and Guattari’s collaboration also supplies one of the few examples within new materialist theory of a micropolitics of capitalist political economy—first in Anti-Oedipus (Deleuze and Guattari 1984), then developed further in A Thousand Plateaus (Deleuze and Guattari 1988) and in Deleuze’s (1992) brief discussion of control societies. In this immanent perspective, ‘integrated world capitalism’ (Guattari and Negri [1985] 2010, pp. 48–53) is quite distinct from other socioeconomic modes: it is an ‘international ecumenical’ social formation (Deleuze and Guattari 1988, p. 435) that transcends the nation-state, enabling the globalisation of production and exchange (see also, Hardt and Negri 2001, pp. 326–22). Nor is capitalism an evolutionary step along an inevitable path to communism, as argued by Marx and Engels ([1888] 1952, p. 60).

Capitalism has achieved this status, according to Deleuze and Guattari, because in comparison to other economic modes, the movements and interactions of money, commodities and labour are highly ‘de-territorialised’ (that is, generalised in terms of what they can do) (Deleuze and Guattari 1988, pp. 435–36). This de-territorialisation has freed the production and exchange transactions of a market economy from the social norms, stratifications or other constraints that regulated and limited the circulation of goods, capital, and information in other political economies such as slave economies or feudalism (Deleuze and Guattari 1988, pp. 453–54). On one hand, there is now a de-territorialised worker, ‘free’ to sell their labour-power, regardless of class, social position or other social stratification; on the other, money that has become ‘capital’ is now able to purchase both that labour-power and any commodity available in the marketplace (Deleuze and Guattari 1984, p. 225; 1988, p. 453).

This de-territorialised free market in money, goods and workers constitutes what Deleuze and Guattari called (Deleuze and Guattari 1988, pp. 453, 457) the ‘capitalist axiomatic’, by which they meant a world-wide drawing together (conjugation) of these three free flows, thereby enabling global capitalist enterprise to develop and flourish (Adkins 2015, p. 226). Guattari and Negri ([1985] 2010, p. 26) described this axiomatic as ‘one vast machine, extending over the planet’ and encompassing every aspect of human life, including ‘work, childhood, love, life, thought, fantasy, art’. This global process is progressively incorporating all sectors of social production into its ambit, with the over-arching objective of assuring capital accumulation (Deleuze and Guattari 1988, p. 464). Capitalist states are real-life manifestations of the capitalist axiomatic (Deleuze and Parnet 2007, p. 129), with the sole aim of sustaining free flows of capital, goods and labour by means of their legislative, fiscal and regulatory powers (Deleuze and Guattari 1988, p. 456).

However, Deleuze and Guattari’s texts (Deleuze and Guattari 1984, pp. 226–27; 1988, p. 461) only hint at how these free flows enable the two key transactions of production and exchange that are core to the capitalist economic mode. To address this gap, the current author’s recent work (Fox 2023a, 2023b) established an ethological ‘cartography’ (Deleuze and Guattari 1988, p. 260), which mapped the interactions between money, goods and labour in these transactions. This project began by re-analysing Marx’s ([1906] 2011) meticulous description of production and exchange in Volume 1 of Capital. While Marx’s focus was firmly upon human agency and the capacity of labour-power to generate capital, an ethological approach enabled analysis of these transactions in terms of more-than-human affects and capacities in actual concrete manifestations of production- and market-assemblages such as factories and physical markets (Fox 2023b).

However, this ethological analysis also revealed further affects in the market-place not acknowledged in Marxist theory, though associated with what classical and neo-classical economists called the ‘laws of supply and demand’ (Marshall [1890] 2009, pp. 284–87; Moore 1925).3 But whereas these economists considered supply and demand as a benevolent means to assure market efficiency and innovation, in this ethological analysis, these more-than-human affects operate beyond individual human intentionality (DeLanda 2006, p. 36). They link the supply of commodities’ relational capacities with the level of demand from customers for these capacities (Fox 2023b). For example, stainless steel has a capacity to resist oxidation (rust) when in contact with water or moist air: a capacity in demand from multiple consumers of cutlery, cookware, surgical instruments and other tools or components. These more-than-human supply-and-demand affects operate independently of both consumers’ ability to pay and producers’ ability to sell at that price.

It is these affects that not only regulate competition between producers and between customers, innovation and growth in production and consumption (Baumol 2002, p. 3; Smith 2010, p. 31), but are responsible for various undesirable features of the capitalist economy. For example, to sustain market share in a competitive market (Wrenn 2016, p. 63), producers face a high level of uncertainty about future sales of their goods or services, due to competition, and must make difficult decisions regarding pricing and staffing—in turn, leading to employment and income uncertainty for employees. Competition between producers also leads to excess supply, resulting in waste (Horton 1997, pp. 128–29) and pollution—including climate change (Keen 2021, p. 1162). Differential wages, and hence capacities to pay for access to commodities and services establish and sustain wide material inequalities between citizens (Fox and Powell 2021; Piketty 2014; Rueda and Pontusson 2000, pp. 353–54).

Crucially, it is these supply-and-demand affects that draw more and more economic and social activity into capitalism’s ambit. Once caught up in the assemblages of capitalist production and markets, both workers and entrepreneurs have no means to escape from what might be described, metaphorically, as capitalism’s ‘black hole’ (Fox 2023b). Advocates for a neoliberalised free market and the rolling-back of the public sector have accelerated the rate at which all human activity (including, for example, education, healthcare, welfare services, public transport and security services) are being progressively drawn into this black hole (Mudge 2008, p. 706).

With the benefit of this dual materialist and more-than-human analysis, which theorises capitalism as a global economic mode dependent upon free flows of commodities, money and labour (the capitalist axiomatic), and the supply-and-demand dynamics these flows enable, the paper now turns to an assessment of what AI can do within this capitalist context. The next section sets out the methods used to assess AI applications ethologically. The paper then reports ethological assessments of four case studies of AI applications, addressing each from the perspective of this dual analysis of capitalism.

5. Research Design and Methods

5.1. Methodology of Inquiry and Methods

To assess AI applications from the perspective of the capitalist axiomatic and the dynamics of supply and demand, the study applies the ethological methodology of inquiry set out earlier in this paper. To reiterate briefly, ethology is the study of the capacities of matter for affecting and being affected (Deleuze 1988, p. 125). These capacities, or ‘affects’ draw matter into assemblages, which are unstable and fluid interactions between human and/or non-human materialities. Within a specific assemblage, matter (for instance, and AI application, or an AI user) gains certain capacities that, in another assemblage, it might not possess. For instance, if a generative AI such as ChatGPT assembles with a school-student, it may provide the latter with the capacity to gain a good mark in an essay assignment. If the same AI assembles with a public-facing business, it can handle common customer inquiries, troubleshoot issues, and provide information, reducing the need for human customer-service agents. The consequent power relations within a specific assemblage describe its micropolitics.

This ethological framework supplies the basis for the methods to be adopted (Fox and Alldred 2022b). Data may be gathered from any source that can identify the affects, capacities and micropolitics operating within the assemblages being studied. Given the theoretical focus of this paper, it will be crucial that the data gathered on AI assemblages include the economic and social contexts of capitalist production and markets. Data analysis methods will address four questions:

• What human and non-human matter is assembled?
• What are the affects between these materialities?
• What capacities are produced in the human and non-human matter in the assemblage?
• What are the flows of power (the micropolitics) in the assemblage?

5.2. Research Question

The empirical data analysed and reported here aims to answer the following question:

‘What can an AI application do in the context of a society underpinned by the capitalist axiomatic and dynamics of supply and demand?’

5.3. Sampling

Four case studies of AI applications have been selected because (a) they engage with four different sectors of the economy: medicine, manufacturing, marketing and finance, and with both ‘productive’ and ‘rentier/platform’4 forms of capitalism; (b) each illustrates the complex assemblages within which they are applied; and (c) each is documented in the scholarly literature. No claims of comprehensive coverage are made for this selection; however, they reflect a breadth of current AI applications in real-life contexts.

5.4. Data Analysis

Following a description of each application, an ‘ethological assessment’ will explore the application cartographically, asking the four questions set out in Section 5.1 above to identify the human and non-human matter in the assemblage, the affects between them, the capacities produced in the human and non-human matter, and, finally, the micropolitical flows of power that these affects and capacities produce. Together, these will answer the overarching question ‘what does this AI do—materially, socially, economically and politically?’ The findings of these ethological assessments will be critically analysed in the following discussion.

6. Four Case Studies: What Can an AI Application Do?

6.1. Optellum: Lung Cancer Diagnostic Tool

Despite advances in immunotherapy, the principal means to increase life expectancy from lung cancer is early diagnosis, typically via tomographic imaging. Interpretation of scans falls to human radiologists, with a very high false-positive detection rate (epidemiologically: low specificity) according to Massion et al. (2020, p. 242), meaning that lung cancer screening is both inefficient for clinics and costly in terms of both clinician and patient time and money. A commercially developed AI tool addresses this problem of diagnostic specificity, successfully differentiating low-risk (likely benign) from high-risk (potentially malignant) pulmonary nodules (lesions) visualised on CT scans.

The Virtual Nodule Clinic tool was developed by a University of Oxford spin-out tech company, Optellum, using a deep-learning ‘convolutional neural network’ architecture developed for computer vision tasks (Massion et al. 2020, p. 243). The objective was to accelerate the diagnosis of malignant lesions and reduce the number of false positives. The AI tool was first trained by a team of radiologists and students using images of 14,761 benign and 932 malignant lesions. Once trained, it was tested on validation databases of positive and negative scan images from two lung disease clinics with different populations and levels of disease prevalence. When compared with clinically-validated risk prediction models in common use by radiologists to assess images of lesions, the Optellum AI supplied improved accuracy in differentiating high-risk and low-risk lesions, with benefits in terms of speed of treatment and more appropriate management of false positives (Massion et al. 2020, p. 248). This AI gained FDA approval in 2021 for clinical decision support in early lung cancer diagnosis in the US, and approval for use in the UK and EU in 2022. Optellum is rolling out the AI in cancer clinics, in partnerships with multinational medical technology companies GE Healthcare and Johnson and Johnson (see https://optellum.com/products-and-solutions/lung-cancer-prediction-ai/, accessed on 13 September 2024).

Ethological Assessment

The Virtual Nodule Clinic’s foundational capacity is to assess scans of suspicious pulmonary lesions, reducing false negatives (missed cancers) and more appropriate management of false positives (benign nodules), based on iterative learning from expert assessments of a large database of scan images. This capacity thereby increases speed of diagnosis and hence treatment, a highly desirable ability for health services. By mimicking the level of skills of well-trained human assessors such as highly-experienced oncology consultants, it increases the specificity of the scans (reducing false positives), but does this same job for a fraction of the financial and time cost of these experts. Such a capacity is in high demand from both socialised health services such as the NHS and commercial clinics, to reduce diagnostic costs and offer appropriate treatment options. As such, the application supplies a means to increase productivity in this diagnostic service and reduce time and financial costs to both clinics and patients. For commercial clinics, this increases turnover of patients, and hence both income and patient satisfaction.

6.2. Streaming-Service Content Recommender System

A recommender system (RS) is an information-filtering technology that provides recommendations that a user may find relevant or interesting. Organisations that use RSs include e-commerce companies (to recommend products); streaming services (content); social media (friends/connections); online news and content platforms; and travel and hospitality services (flights, hotels, destinations, etc.). The video streaming service Netflix uses an AI-enabled RS to assist its subscribers to discover content tailored to their preferences, and thereby enhance their watching experience (Steck et al. 2021).

The algorithm-based AI used by Netflix to power its RS has been enhanced by ‘deep learning’ (Cheng et al. 2016; Zhao et al. 2019, p. 214), enabling it to track and recognise complex patterns within user activity, while also drawing upon wider sources of data, beyond this activity (Steck et al. 2021, p. 12). The developers describe how multiple iterations of tweaking the deep learning models used by the AI and online tests during its development refined the effectiveness of this RS, while also establishing its scalability to a system available to hundreds of millions of users globally (Steck et al. 2021, p. 15). Furthermore, as it learnt, this AI gained capacities to extend the range of modalities (for instance, images and video) that could be captured as data upon which to base its recommendations, enhancing its utility as a means by which Netflix enhanced subscriber satisfaction and hence competitiveness (Steck et al. 2021, p. 16).

Ethological Assessment

The NetFlix AI-enabled RS can supply subscribers with sophisticated recommendations for ‘content that they will watch and enjoy to maximize their long-term satisfaction’ (Steck et al. 2021, p. 8). These recommendations are highly personalised, rather than generic. Underpinning the development of this RS is the capacity to enhance subscriber retention and, consequently, revenue, in an overcrowded market for media streaming services. In the highly-competitive streaming market, the NetFlix recommender can outperform a non-AI enhanced RS, thereby increasing the likelihood of successful subscriber retention over competitors, while also generating data about users that can be monetised (Zuboff 2019). As in the previous example, while in theory humans could perform at a similar level as the AI, the numbers of staff required would be prohibitively costly for the business.

6.3. Knowledge Graphs for Financial Risk Analysis

One of the key challenges for strategic and operational decision-making within financial institutions such as investment banks is ‘entity linking’: associating unstructured information from a wide variety of sources (for instance, company reports, media stories) with relevant entities (people, businesses, etc.). The key task for any entity-linking (EL) tool is to associate a potentially ambiguous ‘mention,’ such as a company name, in a data source such as a supplier invoice, with its corresponding real-world entity. An AI solution to the EL problem has been developed by the large banking corporation JP Morgan Chase (JPMC), using a technique known as knowledge graphs (Ding et al. 2021).

Knowledge graphs are a ‘compelling abstraction for capturing key relationships among the entities of interest to enterprises, and for integrating data from heterogeneous sources’ (Ding et al. 2021, p. 15301). Ding et al.’s paper describes how their team at JPMC implemented ‘JEL’: a large-scale knowledge-graph solution that integrated the bank’s own internal databases with third-party data. This system captures millions of entities, such as suppliers and investors, and the links between these entities (for instance, supply chains or investments), using AI neural-network learning techniques to enhance EL performance. In tests, JEL out-performed other EL systems in terms of both accuracy and precision of entity linking (Ding et al. 2021, p. 15307). Currently, the company is in the process of deploying the JEL model, and actively collaborating with its customers and others to process incoming news articles, extract information about companies, and link them to the corresponding entities in the knowledge graph.

Ethological Assessment

The knowledge-graph AI solution to the challenge of entity linking enables vast quantities of disparate data from multiple sources such as incoming news articles and publicly available company information to be trawled for links to corresponding entities in JPMC’s databases. This capacity has the potential to significantly enhance core aspects of JPMC’s financial business processes by deepening understanding of the financial and other networks within which these entities are located. These processes include analysing and managing business risk, strategic decision-making concerning investments and loans, supply chain assessment, detecting fraud, and offering investment advice. All of these business processes are—in one way or another—key to the economic fortunes of JPMC, giving the bank a commercial advantage over competitors without incurring additional costs of entity linking by human employees.

6.4. Additive Manufacturing Neural Network

In manufacturing operations, product and process complexity, customer variability and competitive pressures all challenge throughput, quality, safety, and cost objectives. AI has been advocated as a tool to address these challenges, by analysing the data generated by machines, sensors and other records of operational and performance data, in order to identify patterns that conventional methods may miss and establish a deeper understanding of manufacturing processes (Arinez et al. 2020).

One area where an AI-enabled analytical tool may be used is in additive manufacturing (AM), commonly known as ‘3D printing’ (Conner et al. 2014). The task of fabricating three-dimensional objects from digital designs poses a challenge of scalability, as it involves substantial and sophisticated decision-making to enable the correct manufacturing processes to be employed to deliver a suitable product. Elhoone et al. (2020) suggest an AI solution to identify the optimal AM technique to and dynamically allocate production to networked AM devices. The authors trained artificial neural networks based on three different algorithms on 300 Computer-Aided Design (CAD) images of complex three-dimensional machine parts or appliances. They built an AM decision model based on nine ‘input’ variables, such as product dimensions, material, surface finish and speed of manufacture, and five AM processes including sheet lamination, material extrusion and binder jetting (3D printing) (Elhoone et al. 2020, p. 2845). The trained neural networks were tested on a total of 100 new CAD images, with a ‘general regression neural network’ achieving the highest accuracy of 97-percent appropriate decisions. The authors also designed an interface that enables the AI’s decisions to be sent automatically to the most appropriate AM device for real-time production.

Ethological Assessment

The AI-enabled solution integrates inputs from user preferences, digital designs and machine availability to dynamically allocate different part designs to the most appropriate AM machine (Elhoone et al. 2020, p. 2859). If implemented, this both enables wider adoption of AM within industry and facilitates AM supply chains (Elhoone et al. 2020, p. 2842), providing customers with faster production and automated allocation of designs to the most appropriate manufacturing device. This AI solution effectively replaces all human intervention in the AM process, while also increasing the volume of AM undertaken, thereby enhancing income to manufacturers and potentially lowering costs for clients.

7. Discussion: AI, Neoliberalism, and Supply and Demand

These case studies of current manifestations of artificial intelligence technologies supply evidence of what AI can do in the contemporary context of integrated world capitalism and the capitalist axiomatic. An ethological analysis that acknowledges the economic, political and social contexts of AI applications has revealed that all four of the case studies analysed are justified both in terms of benefits to clients in terms of increased accuracy or effectiveness (in diagnosis, recommendations, entity-linking and allocation of designs to AM devices) and supplying cost savings over non-AI processes, typically by eliminating the need for (or replacing) low- or high-skilled human labour. Case studies 2, 3 and 4 offer direct commercial advantages to their users; the first example (the lung cancer diagnostic tool)—though delivering a healthcare service of great value to both patients and healthcare professionals, has been established as a commercial enterprise by a university research spin-out company. It has used venture capital to fund its development, with the aim of returning a revenue for investors, as the AI tool is sold to health providers in the US, EU and UK, including the National Health Service.

In their analysis of AI, Henriksen and Blond (2023) identified a disconnect in organisations applying AI solutions. While workers saw AI as an opportunity to enhance their work (‘intelligence augmentation’), managers considered it primarily as a means to increase efficiency, by automating human tasks (Henriksen and Blond 2023, p. 754). The analytical opportunity supplied by the more-than-human (micro-)political economic perspective developed and used in this paper is to disclose this disconnect as consequential upon the de-territorialised flows of commodities, money and labour in capitalist economies and the dynamic of more-than-human supply-and-demand affects these flows enable.

In the contemporary world of privatised and neoliberalised global trade and consumption, the capitalist axiomatic has progressively incorporated most sectors of social production into its ambit (Guattari and Negri [1985] 2010, p. 26). Its concrete realisation—the capitalist state (Deleuze and Guattari 1988, p. 456; Deleuze and Parnet 2007, p. 129)—uses its legislative, fiscal and regulatory powers to facilitate free flows of commodities, labour and capital through the economy and society (Fox 2023a). Within this context, markets are driven by more-than-human supply-and-demand affects that require enterprises to continually strive to sustain or increase market share by keeping their costs low and prices as close to, or lower than, those of their rivals (Wrenn 2016).

All four of the AI applications analysed in this paper exemplify the operation of these more-than-human supply-and-demand affects. The AIs supply capacities for which there is a demand (for instance, to expedite additive manufacturing of machine parts or link entities for business planning), while offering competitive advantage to their providers, whether a pulmonary disease clinic or a streaming platform. All four offer the potential to compete, and increase the market share and consequent profitability of a business, corporation or platform. The capacities of each have been refracted through the lens of integrated world capitalism.

This assessment suggests that multiple AI applications have been caught up with these supply-and-demand affects, as enterprises vie to offer customers products with more attractive capacities than those of their rivals, while attempting to sustain or enhance their market share by reducing labour or other costs. While AIs will no doubt continue to be developed for multiple everyday purposes, from assisting essay-writing (Sharples 2022) to ensuring the affluent home larder is fully stocked with groceries (Schiller and McMahon 2019, p. 185), the main thrust of AI developments will service the ends of competitiveness, profit, and capital accumulation. Furthermore, these developments will be entrepreneurial, funded by corporations or venture capital, rather than from the public purse. Just as the trajectory of the ‘digital economy’ since the 1970s has been underpinned by the competitive market (Wittel 2015, p. 69), and has come to be dominated by a handful of global players such as Microsoft, Google and Apple (Fox 2023b), a similar trajectory for AI innovation may be predicted.

Beyond the various imbrications of AI and capitalism discussed earlier, the analysis here indicates that the story of AI is foundationally also the story of capitalism. This matters because of the negative consequences of capitalism and of unregulated supply and demand. Advocates of the ‘free market’ in commodities and labour, from Adam Smith to contemporary neoliberal politicians, considered supply and demand as an unrivalled engine of growth and prosperity (Bishop 1995, p. 165). By contrast, an ethological assessment suggests that the free rein of supply and demand in the era of integrated world capitalism is a tale of cut-throat competition; of an inexorable drive for economic growth, regardless of cost to society, the environment and the climate; of the waste of resources and human labour as competition ravages enterprises; and of widening social inequalities. While AI-enabled products and services are proving attractive to customers and profitable for businesses exploiting their capacities, all they do is hasten the immurement—for both workers and entrepreneurs—within the metaphorical ‘black hole’ of global capitalism.

Consequently, this analysis asserts that AI will exacerbate the current neoliberal socioeconomic trends of labour precarity and social inequalities (Herod and Lambert 2016, pp. 25–26; Piketty 2014, p. 277), as competitors introduce ever-more powerful applications, profit margins tighten, competition becomes more and more intense, labour becomes progressively precarious, and global business platforms achieve market dominance. This will also contribute further to a divide between an AI-enabled global North and a consequently work-poor global South (Makridakis 2017, p. 58), exacerbating mass migrations, and conflicts and wars over territory and natural resources (Baer 2018, pp. 35–37), all fuelled by the more-than-human affects of supply and demand that operationalise the capitalist axiomatic. This is the principal sociological significance of AI.

Despite this conclusion, the ethological analysis of capitalism outlined here also suggests the way back from the edge, with the potential for AIs to contribute to an alternative political economic future: one requiring a less drastic economic and political revision than proposed in neo-Marxian political economy. Deleuze and Guattari’s (1988, pp. 460–63) analysis dissolves the notion of a unitary coercive capitalist state directing the accumulation of capital. Instead, the capitalist state comprises multiple sociomaterial assemblages or ‘machines’, which together enable and promote the free flows of capital and labour throughout the economy (see also Deleuze and Parnet 2007, pp. 129–30). If the political will is there, these assemblages can be turned toward the objective of undermining the affects linked to supply and demand strategically and operationally, locally, nationally and globally. This intervention can thereby unsettle, arrest and eventually reverse the dynamics of the capitalist axiomatic, including the part that AI is playing within the contemporary capitalist economy. This strategy moves far beyond efforts to simply regulate AI, and can be sketched out here in terms of three related objectives.

The first objective is to counter the proposition asserted by some that capitalism’s success universally benefits society, instead highlighting its detrimental effects on people, the environment and the climate due to resource extraction, growth, and waste. Illustrations of how capitalist applications of AI such as intelligent robotics can lead to de-industrialisation, unemployment and increased job precarity, or how facial- or gait-recognition AI applications increase surveillance and authoritarianism (Dauvergne 2021, p. 295) can contribute to this communication. This element of the strategy is primarily targeted at supporters of neoliberalism and marketisation who are unwittingly advocating for the black hole of capitalism, but also to developers of AI. In addition, this message needs to be disseminated throughout civil society.

Second, to support financially commercial alternatives to entrepreneurial capitalism, including not-for-profit organisations, social enterprises, cooperatives, alongside public ownership of essential services. AI applications such as artificial neural nets and generative language processing can be used by governmental agencies, co-operatives and social enterprises to increase their efficiency and effectiveness, while also supporting information management to support these alternative economic models; to develop new public and civil society services; and to enhance green energy use and environmental protection (de Sousa et al. 2019, p. 7).

Finally, to undermine the unfettered operation of supply and demand in capitalist markets by increasing regulation and controls on economic activity, including the use of AI applications to supply price advantage or replace human labour. For classical economists, such regulatory interventions are market ‘distortions’ (Palley 2005, p. 22)—here, the aim is not distortion but disruption of a ‘free’ (that is, de-territorialised) market. Such disruptions include introducing price controls on commodities, fiscal measures to promote socially useful innovations, penalising resource extraction and waste, ending free-trade deals that advantage the global North, and using windfall taxation to increase welfare benefits. Machine learning and generative AI can be used to model the management of the capitalist market, and enhance policy development and implementation in all these aspects of political economy. This includes supporting poorer countries to enact this strategy, while working towards resolving conflicts caused by territorial and resource disputes exacerbated by international capitalism.5

8. Conclusions

The analysis undertaken in this paper indicates that many artificial intelligence applications are already in thrall to capitalist enterprise. The ethological, more-than-human (micro-)political economic analysis used to unpick AI’s capacities has disclosed that this association is exacerbating the global capitalist axiomatic, via the assemblages of capitalism and the more-than-human dynamics of supply-and-demand affects operating beyond human intentionality. As such, AI is accelerating a future of deepening social inequalities, precarious work, conflicts over resources, and environmental catastrophe. AI is not the problem, but it is part of the problem: the problem is capitalism and the more-than-human supply-and-demand affects that drive competition, waste and social inequality. However, AI can also be part of the solution. It has—if applied as an element in a wider strategy—the capacity to contribute to an economic and social transformation of the supply-and-demand affects that sustain the worst excesses and unintended consequences of global capitalism.