Ethical AI is a smokescreen for capitalist exploitation of global south

The European Union (EU) Commission’s recent proposal for a regulation of artificial intelligence (AI Act) [27] appears to signal a change in approach. Despite worldwide and growing concern for the need to regulate AITs, efforts remain nascent, abstract, and mostly self-regulatory [32], [35]. The EU Commission’s proposal stands out for a different approach: more hierarchical, coercive, and “dissuasive” [27]. It suggests banning some AI uses altogether (e.g., social scoring), certifying and risk-mitigating others (most public uses of AITs), and introducing sanctions1 that are meant to ensure high compliance rates.

The regulation is unlikely to take effect until 2023 and has already been contested by civil society organizations, specialists, and the European Parliament (see [21]). There are likely amendments on the horizon but these tend toward extending (rather than limiting) the proposed protections from some “high-risk” use cases (see [26]).

The overall proposal positions the EU at the forefront of developing a comprehensive regulatory approach to AI. Its rather pompous self-proclaimed aim is to produce and foster “cutting-edge,” yet “trustworthy” and “human-centered” AITs made in Europe [24], [39], [49], [52]. It suggests a dual aim to contribute to more equitable AIT production and applications through its common market and to generate future economic growth through ethical AIT design and development.

In this statement, we explore the narrative surrounding the AI Act and situate it in a problematic social scientific history of European integration claimed as a project of “peace and prosperity.” In an interpretive policy analysis [64], we explore how the relationship between ethical principles (“peace”) and economic rationalities (“prosperity”) is depicted in the draft Act and discuss what such depiction reveals and hides about the wider context of the EU’s regulation of AITs. We draw on colonial political economy accounts of capitalism [9] and European integration [33] to qualify the potential of the self-declared “normative power Europe” to lead the way to more equitable AI development and use through its regulatory powers.

Our argument runs in three steps: • First, we depict the EU Commission’s narrative on the AI Act in more depth, drawing on our own interpretive policy analysis of official documents. • Second, we situate the Commission’s “trustworthy and human-centered AI” narrative in wider explanations of European integration and market-making and show how AIT regulation reiterates the EU’s peace and prosperity myth for the digital age. • Third, we problematize the notion of an AIT common market that fosters prosperity in ethical ways. Instead, we locate this project as a late addition to long-running mythmaking around European integration that disguises its colonial political economy, including for AI, and from which we identify four lines of inquiry that should orient our assessment of current and future regulatory efforts.

Note that we decided to include quite a large variety of references for a short reflection paper to allow our readers to deepen the several—so far separate but deeply interconnected—discussions on the regulation of AITs, EU common market-making, and colonial political economy.

EU’s narrative of “cutting-edge, yet trustworthy” AI made in Europe

The development of “cutting-edge, yet trustworthy” AI in the EU was identified as a key priority for the EU in late 2019 [24], alongside the European Green Deal and managing the knock-on effects of the SARS-CoV-2 pandemic. Importantly, this level of priority also included the development of a regulatory approach from scratch. The Commission’s AI policies include a strategy paper and a coordinated plan in 2018, the establishment of a AI High Level Expert Group (AIHLEG) in 2018, a Commission White Paper in 2020 which draws on the publication of the AIHLEG’s policy and investment recommendations and ethics guidelines in 2019, and eventually a proposal for regulation in 2021.

The bottom line for EU regulation of AITs has been neatly summarized as “we want AI, but we do not want any AI” (foreword by Fredrik Heintz in [39]). Shortly after taking office on December 1, 2019, the EU Commission’s new President Ursula von der Leyen promised to develop the, worldwide, first set of binding rules for “trustworthy” development and application of AIT systems in the first 100 days in office. In that context, von der Leyen reiterated the goal of “reaping the benefits of technology and making it work for people” [24], an ambition that was already expressed in the EU Commission’s AI strategy in 2018 as follows: “the EU can lead the way in developing and using AI for good and for all, building on its values and its strengths” [23].

Indeed, similar formulations of regulatory goals have already surfaced in the context of the 2018 General Data Protection Regulation (GDPR) which is seen as a “major step for building trust” in the development of new technologies and is cherished, by the Commission, as a key example for how “the EU’s sustainable approach to technologies creates a competitive edge, by embracing change on the basis of the Union’s values” [23]. Amidst this focus on enhancing the competitiveness of the Union on all things AI, the Commission’s strategy highlighted its wish to “ensure” that AIT design, development, and use should respect “ethical principles” and “fundamental rights”:

…some AI applications may raise new ethical and legal questions, for example related to liability or potentially biased decision-making. The EU must therefore ensure that AI is developed and applied in an appropriate framework which promotes innovation and respects the Union’s values and fundamental rights as well as ethical principles such as accountability and transparency. The EU is also well placed to lead this debate on the global stage. This is how the EU can make a difference – and be the champion of an approach to AI that benefits people and society as a whole” [23]

This cursory summary of the EU’s own reasoning highlights twofold regulatory goals. On the one hand, the Commission wants to mitigate AIT-related risks and expresses this in terms of the trust, ethical principles, and fundamental rights. On the other hand, there is a strong focus on developing market leadership for the EU in the wider geopolitical economy of AI technology production. Of course, this agenda is also closely linked to ideological and security concerns about changing international relations that position the EU between the United States and China (mirroring the EU’s normative and security positioning between the United States and the Soviet Union in the post-World War II period). The role of France and other EU countries—along with Canada, in particular—in pushing for regulation of AIT use via the Global Partnership for AI, exhibits this tension especially strongly.

Indeed, some observers argue that the swift development of a regulatory framework signals “an increased awareness of ethical and value-based concerns surrounding applied AI” in the Union [39], [52]. This awareness certainly responds to the substantial and numerous concerns of institutions such as the UN [1], the Council for Europe [15], [16], and the European Parliament [25] about the transparency and accountability of algorithmic decision-making. This has centered on the protection of fundamental human rights, but also the use of AIT in law enforcement, the judicial system, and social welfare. The Commission, in its 2018 strategy paper, expresses the wish to “ensure an appropriate ethical and legal framework” which will include product liability rules, safety standards, ethics guidelines, and the “empowerment” of users and citizens vis-à-vis AIT-based data extraction and decisions [23]. It seems plausible then that scholars assert the EU Commission’s main goal is the development of a “human-centered approach” to AI systems where humans should “be the baseline upon which these systems should improve” and where these systems should abide by the “same requirements” as we usually put on people when regulating their behavior [39]

And yet, the EU’s approach to regulating AI and automated decision-making has, from the very beginning in 2018, been underpinned by market development, competition, and security considerations. The need to protect this was highlighted by a key position paper by 14 member states [18]. The Commission explains the “strong political endorsement” of 24 member states and Norway to work together on regulating and promoting AITs first and foremost with the requirement of “significant efforts to ensure that Europe is competitive in the AI landscapes with bold investments that match its economic weight.” Ethical considerations rank second and third and are expressed in ways that mask agendas that also enhance those investments: “no one is left behind in digital transformation” and “new technologies are based on values” [23]. The proposed regulation argues that it “aims at strengthening Europe’s competitiveness and industrial basis in AI” [27].

Myth continued: The common AI market as a force for peace and prosperity?

This specific narrative on the EU’s AI regulatory efforts taps into a hard-wired and encompassing narrative of the drivers of European integration and policymaking since the Treaties of Paris and Rome: the six founding members would have come together in the Coal and Steel Community and then the European Economic Community to ensure peace and prosperity in Europe after two devastating wars.

Robert Schuman’s famous Declaration of 1950 singles out “world peace” as a key driver for uniting Europe, arguing that “solidarity in production” will “make it plain that any war between France and Germany becomes not merely unthinkable, but materially impossible” [28]. Economic integration will also lead to prosperity, namely the “modernization of production and the improvement of its quality, the supply of coal and steel … to the markets of … member states … the development in common of exports … [and] the equalization and improvement of the living conditions of workers” [28]. The EU’s contemporary self-representation explicitly builds on Schuman’s publicized vision: “As of 1950, the European Coal and Steel Community begins to unite European countries economically and politically in order to secure lasting peace.”

The framing of the EU’s rationale for AIT regulation reflects this favored historical narrative. What is missing is Schuman’s hidden rationales for a united Europe—and especially his colonial views on “developing Africa,” thereby “revitalizing” an “old Continent” [28], [33]. This is despite—or perhaps because—of the ongoing relevance of these ideas to the framing of many of the Union’s policies: including development, technology, science, security, and migration. The emphasis on creating a normative market leadership between the United States and China disguises this significant colonial dimension and context of its emergent AIT regulatory framework. It does not diminish or displace it.

As the EU advances as a site of AIT design, development, use, and regulation, it apparently does so with two feets firmly set on its own founding myth. Notions such as technological innovation, bold investment, growth, and prosperity come across very explicitly in the proposed AI Act and the Commission’s supporting statements. At the same time, the vision of a “human-centered” AI economy that “works for the people” by respecting their basic rights and ensures trust and transparency speak to a self-declared higher moral ground that has, after all, earned the EU not only just the label of “normative power Europe” in international relations theory [54], but also a Nobel Prize.

Scholarly work on AI regulation plays its part in feeding the Janus-faced narrative of AI competitiveness and ethical AI. More prescriptive work in political science and legal studies conceives of regulation in the AIT and automation as a proportional correction of market failure [36]–[38]. These studies advocate for “risk-based” regulation that assesses the likelihood and relative impact of adverse effects of AI applications, and, on that basis, designs differentiated regulatory answers (e.g., prohibitive approaches toward unacceptably harmful uses; tight risk mitigation for high-risk uses; and permissive regulation for low-risk adoptions). Such risk-based approaches would, the authors suggest, facilitate innovation, investment, and AI marketing, key economic goals of the EU and other regulators, all while mitigating the greatest harms and ethical concerns.

The EU’s proposed AI Act justifies its own “riskbased” and “proportional” approach with precisely that kind of reasoning: the ban on social scoring and tightly curtailed application of AI-based automation in public decision-making for ethical reasons coexists with regulatory sandboxes and light-touch approaches to “low-risk” use cases [68]. In a nutshell, these accounts portray risk-based AI regulation à la EU as a somewhat magical rationalizer for achieving both: 1) the provision of trustworthy AI that respects human values and rights and 2) the creation and expansion of a competitive European AI market.

Of course, the AI Act is not without critics who have cast plausible doubt on the ability of the regulation to pursue the dual goal of peace and prosperity, or trustworthy AI. As it stands, this comes mainly from critical legal scholars, who castigate insufficient data protection [65], a failure to acknowledge the collective dimension of data-related rights, or a lack of enforceable standards [32], [38], among other issues. Such criticism creates important grounds for qualifying the actual drivers of the EU’s “trustworthy” and “human-centered” AI narrative.

Nonetheless, these important critiques do not explain the emergence of this specific form, mode, and/or scope of actual AIT regulation. First in the context of the wider political economy of European integration (i.e., why this form, here and now). Second, in the context of the wider availability of possible alternative approaches, from outright bans, to impact assessments and transparency or monitoring requirements (i.e., why this mode). Third, in the context of accelerated economic agendas to promote AIT developments (i.e., why this scope). In particular, there is a striking lack of focus on the regulation of AIT design and production, and rather more on its use.

Political economists offer a more profound contextualization when highlighting the overriding normative and functionalist force of the EU’s common market project. Rosamond [54], for example, fittingly describes the Union as driven by the “conquest for the common market” with vast implications for integration internally and for global governance externally. Internally, the EU regulates to minimize the adverse effects of globalization—such as environmental degradation, tax avoidance, or indeed mass data extraction—for its own citizens and companies. Externally, it regulates not only to “externalize” its own normative principles, but also to enhance and protect its own competitiveness in a global market [29], [54].

Some see the externalization of EU regulatory norms as a “by-product” of consumer protection goals rather than a case of “regulatory imperialism” [10]. In product regulation, scholarship has observed a “Brussels effect” [10], [57] where the EU unilaterally sets aggressive product standards to shield its consumers from chief risks. Firms in other jurisdictions—seeking access to the largest consumer market in the world—must comply, thus by default imposing a regulatory benchmark. The GDPR, in particular, has been credited with this role in a regulatory race to the top in data protection.

The EU Commission is well aware of the by now global reach of EU standard-setting in many areas of product regulation [10]. In that perspective, “ethical AI” could be a way to keep non-EU AI products and applications out “if they do not conform to national regulations serving valid health, safety or environmental purposes”; AI regulation could “serve as a certificate of superior product quality that is rewarded by the market” [57] (for global AIT regulation, see [70]). The EU AI Act could hence be a recent case of European quality product regulation where “trustworthy” and “human-centered” AI, while also promoting consumer protection and security (the moral high ground just short of “peace”), are integral to the competitiveness brand through which the EU seeks to protect its internal AI market and compete on a high-quality label globally (“prosperity”).

In addition, critical governance scholars highlight the intersection of Brussels’ AIT, digital and data regulation, with its own governance projects. This is likely to constrain, or at least shape, the EU’s regulatory impulses to protect EU executive bodies and member state governments as chief users of AITs.

A case in point is the EU’s approach to AITs in border control. Scholars depict a mutually constitutive logic of “business facilitation” and “border protection” in the EU’s “smart borders” approach—for example, green-listing of frequent trusted travelers, black-listing of risky individuals based on algorithms, and gray-listing of most people [40]. Digitalization is fuelled by a tech and big data industry “eager to cater to real and perceived needs of governments in the field of border management” [11], while governments are eager to ensure their products and digital and technology economies are supported.

For example, Carmel [12] traces how research and development for security and military technologies is conceptualized and funded through EU Horizon and other funding streams. These are strongly oriented to efforts to develop and enhance European security technology markets in ways directly proposed by corporate interests and to meet public agendas for promoting investment in these technologies, including AITs. This leads to research organized around the interoperability of data and systems across the EU to facilitate product development, subsidizing pilot-testing of security technologies and facilitating product sales. This example suggests that we need to understand the EU’s AI Act in the context of Brussels’ proactive and self-referential market-making agenda and its wider political economy.

Eventually, then, it seems that “the twin considerations of value capture and ethics” ascribed to AI regulation in the EU [39] might not be equal twins after all but imply a hierarchy of these goals. As the EU has entered into a regulatory competition with China, the United States, Canada, and other countries developing AIT specialisms, the “trustworthy AI” agenda may not be primarily a way of producing AI more ethically across the globe. Recent studies suggest that the strong focus on AI ethics and risk mitigation also serves the Union’s own competitiveness goals [70]. Meanwhile, we can see that AIT competitiveness is likely to conflict with ethical tenets, rather than harness them, given the reliance on exploitation of labor for data, energy for systems development, and earth/wider environment for hardware and infrastructure. In light of these observations, we now turn to examine how we might more usefully examine the framing and focus of the proposal.

Toward a colonial political economy of AI regulation

The dominant self-representation of the EU’s AI Act and data protection regulations are as standing in a heritage of “peace and prosperity,” of human rights protection and as moral leaders. However, when we revisit this claimed heritage, a more ambivalent set of relationships between market-making, power, and the political economy of regulation are revealed.

The above accounts consider the common market as a rather self-contained Western project. While acknowledging the interdependence of national economies on global markets as drivers of economic and political cooperation in the EU and beyond, these accounts disguise the colonial connections that underpinned capitalist modernization from the outset.

This is not only oversight in studies of European policymaking. As Bhambra [9] recently showed in her overhaul of the history and development of “Western” capitalism in Europe, most classic accounts of capitalism—from Karl Marx to Max Weber—fail to address that “the global connections forged through colonialism ... are the condition of capitalist-modernity.” This is true, she argues, for the myth of “discoveries” and appropriation of allegedly unchartered land as the foundation for capitalist expansion from the 16th century. The colonial apparatus was claimed as ontologically a product of the moral superiority, progress, reason, and civilization of the White European metropoles [45] (see [56]). This in turn erased the terms of capitalist progress and the destruction and dispossession on which it rested (e.g., [53], [67]). These insights draw our attention to whether and how the EU’s efforts in AIT market-making might not only express a particular competitiveness strategy, but also build on and reproduce this fundamental colonial political economy of domination, exploitation, and geopolitical influence.

Its relevance is shown when we consider that “peace and prosperity” for Europe relied on colonial power relations in its founding, as Hansen and Jonsson [33] argue. Their careful and thorough archival work on the early days of European integration shows that economic and political unification in Europe was thought of at the time—including by key figures such as French Foreign Secretary Robert Schuman or German Chancellor Konrad Adenauer—as fundamentally relying on the extraction of resources (raw materials and labor) from the French, Dutch, and Belgian colonies, the sharing and pooling of these resources across the union-to-be, but also the joint investment in harvesting these resources among the six. Drawing Germany and Italy, those member states that lost their empires after the Second World War, into a pooled form of European Economic Communities (EEC)-colonialism would help, in the words of Robert Schuman, to “revitalize” Europe as a whole and to secure peace and prosperity on the continent ([33] also see Schuman Declaration).

Reflections by decolonial scholars of political economy and European studies thus raise the analytical question of how Brussels’ current efforts to create a trustworthy AI market in Europe tie in with this “long-forgotten or suppressed” history of the European project [33]. We propose four lines of inquiry here:

1. Examine the interaction of EU AIT regulation with the global political economy of AITs’ physical infrastructures and the deep inequalities it generates. Critical AI scholars have described the extractive industry that supports AIT production as a new form of colonialism where the production of capitalist surplus for a handful of big tech companies draws on large-scale exploitation of the soil, minerals, and other resources (e.g., [17]). The proposed EU regulation does not address the harm that occurred in other parts of the world through the vast environmental and social costs of mineral and resource extraction and energy consumption for large-scale computing [14], [19], [71]. Indeed, the draft regulation says little to nothing about the conditions under which AITs are produced at all; the colonial political economy that sustains conditions of production, in which the “high-value” (scientific and design) elements of AIT design and development remain firmly in the hands of metropolitan centers.