A Reading of “Data as Labor”: Reinterpreting Narratives of the Digital Economy

The Mechanical Turk – a prototype of the “ghost in the machine”

4An image aptly illustrating the ambiguous nature of the “ghost in the machine” is the so-called Mechanical Turk – probably the first known specimen of the attempt to create the appearance of a fully automated, artificial intelligence: a chess-playing “machine”, built in the late 18th century to impress a seemingly gullible public by creating the impression of playing chess “mechanically”. After having toured the world and defeated such famous human opponents as Napoleon und Benjamin Franklin, the Turk was revealed to be a hoax: The box under the chess player model did not contain a complex mechanism but a human chess master guiding the model’s moves.

Fig. 1¹

5It is of course far from clear to what extent the illusion of the chess machine had initially “really worked”: Was the appeal of the Chess Turk due to its convincing rendering of what a mechanical intelligence might really look like? Or was it rather a certain lack of convincingness in its construction – its partial transparency as a hoax – that was at the core of its attractiveness for the public? At least part of its effect was certainly the kind of curious awareness that one encounters when watching conjuring tricks, when what is clearly an illusion appears like a surreal reality, which, far from eliciting credulity, rather provokes fascination with how the deceptive magic operates. This ambiguity at the heart of the first fake AI model makes it all the more interesting as a starting point for a contemplation of its contemporary successors (comp. Douetteau , 2018).

6Incidentally, almost 2.5 centuries later, this strange case of fake AI is used by the tech giant Amazon as a model for its crowdsourcing platform MTurk which it actually dubs “artificial artificial intelligence”. Here, the not-so-artificial mechanism is “hiding in plain sight” as it were, with the fictitiousness of the self-sufficient automatism of the MTurk-tool openly described as part of its conception. According to Amazon,

MTurk enables companies […] to streamline business processes, augment data collection and analysis, and accelerate machine learning development. While technology continues to improve, there are still many things that human beings can do much more effectively than computers, such as moderating content, performing data deduplication, or research. Traditionally, tasks like this have been accomplished by hiring a large temporary workforce, which is time consuming, expensive and difficult to scale, or have gone undone. Crowdsourcing is a good way to break down a manual, time-consuming project into smaller, more manageable tasks (Amazon MTurk Website).

7To achieve such effective results, Amazon invites companies to “access a global on-demand 24/7 workforce”, and workers to “make money in their spare time” (Amazon MTurk , official website). In its own way, Amazon’s MTurk thus creates, just like the fake chess-machine of yore, the appearance of rendering results mechanically or “virtually”, without recourse to human labor recognized or recognizable as such, i.e. without having to hire and pay “a large temporary workforce, which is time consuming and expensive”. The microtasks are performed in the workers’ “spare time” and as such are not compensated as work in the stricter sense. Shortly after the launch of MTurk by Amazon, the Newspaper Economist fittingly defined the target group for MTurking as “people who have computer access and some time to kill“ (Economist, 06/2006), i.e. bored housewives and unemployed persons, whose activities are not considered as work but rather as pastime and thus do not require a remuneration but rather some pocket money of sorts: the real hourly rate at MTurk varies depending on the specific tasks, but in average does not exceed 5 USD (Anderson 2022). The real workers steering the “mechanical Turk” are thus concealed – in this case not in a box, but behind the appearance of leisure activity. And this time around the concealment is practiced in a paradoxically overt way, rather straightforwardly announced by Amazon to its customers and the general public – almost as if the company wanted to share its skillful conjuror’s tricks with an admiring audience.

8To grasp the implications of the MTurk-phenomenon – not only within the scope of crowdsourcing data work, but also in the larger context of data-driven digital technologies – , it is enlightening to consult Eric Posner and Glen Weyl’s book Radical Markets: Uprooting Capitalism and Democracy for a Just Society (2018). In light of Posner and Weyl’s analysis, the pseudo-artificial, non- or super-human automaton that obfuscates the strictly human efforts and interests sustaining it (i.e. hiding the real player in a narrow box), proves to be a fitting metaphor for the workings of today’s global digital market. As the authors argue, completing simple virtual tasks for the benefit of tech companies, without being paid appropriately (or at all), is something not only microtasking human intelligence providers do on MTurk: it is the daily practice of anyone who uses social media platforms, downloads apps, plays video games etc. – so basically all of us.

9According to Posner and Weyl, the typical business model under the rules of today’s digital market can be summed up as follows: Tech companies offer their users ostensibly free services, and, in exchange, collect the users’ data. Accordingly, the authors describe platforms such as Facebook as “Siren Servers” or “data collectors, delivering services that lure users into providing information on which they train AIs using ML” (Posner and Weyl 2018: 220). In this constellation, a service of the tech company (in the form of entertainment or information) is paid for by a return service tacitly extracted from the users (providing their data), with the terms and conditions of this opaque transaction being under the sole control of the tech company, given that the users are left in the dark as to the purpose and value of the data they are providing. It is precisely this economic model, termed “technofeudalism” (Posner and Weyl 2018: 230), that allows for a quasi-monopoly position of a small number of tech giants. And as Posner & Weyl argue, this model can only be transformed in favor of an economically more balanced and socially more just system, if the legal basis for accessing user data is radically reconfigured – not through restrictive legislation on data protection, but through a re-evaluation of “data as labor” (Posner and Weyl 2018: 205ff).

10Such a perspective destabilizes the misleading notion of the body of data available to a tech company as the capital owned and deployed by it: “Data giants” seemingly create value out of nothing, without the use of an extensive labor force, but solely through the individual achievements of a small number of Silicon Valley code wizards, who, through their computing skills and entrepreneurial ingenuity, bestow innovative products on a mass of passive consumers. In contrast to this view, Posner and Weyl claim that this very mass of ostensibly passive consumers, who provide valuable data to the operators of these systems, are to be re-evaluated as data workers, i.e. active contributors to the creation of the digital products developed from the data material provided by them. Thus, AI products seemingly unilaterally designed and developed by “data titans” should rather be conceived of as a return on investment of collective resources, or as “collective intelligence”. In other words, speaking in terms of the fake chess automaton: The human contribution sustaining Artificial Intelligence should not be conceived of as a secret maneuver hidden in a box whence magic products are pulled like a rabbit from the hat. Rather, it needs to be acknowledged as the engine and the guidance system of digital technologies and the economic systems relying on their development and deployment.

11As I would like to show in the following, the MTurk construct can serve as an extended metaphor for the economic, political, and cultural function of AI-driven digital technologies – relying on the appearance of a coherent and self-sustaining system of knowledge producing meaningful actions and decisions, which is actually masking a multitude of divergent and often contradictory human actions, motives, and power relations hiding within its core.

12As formulated by Kate Crawford in her Atlas of AI, “AI is neither artificial nor intelligent” (Crawford 2021: 8, italics orig.) but is rather to be considered as a “massive industrial formation that includes politics, labor, culture and capital” (ibd : 9). In order to understand, how and why the latter – a vast organic conglomerate of complex relationships and messy enmeshments – is made to appear as the former – a neat technological tool with clear-cut applications and purposes – , it is interesting to look into the workings of this conjuring trick through the lens of psychoanalysis.

A psychoanalytical perspective: The Subject as product and producer of its big Other

13From a psychoanalytical perspective, the MTurk-construct aptly illustrates the structure of belief systems and the ambivalent relationship between the human subject and the object of its belief – the religious or ideological, economic or sociopolitical signifying order it refers to – a superordinate reference system termed the big Other (Žižek 2008 : 45).² While the big Other typically appears to be an external, autonomous authority guaranteeing meaning and purpose and providing coordinates for the right course of action to its subjects (such as a God, or the inexorable logic of the Market), at a closer look it reveals itself as the result of the collective input of the actions and intentions, proclaimed interests and hidden motivations, open and covert power struggles of its subjects.

14According to the French philosopher and psychoanalyst Pierre Legendre, this ambivalence of belief systems goes as far back as to the medieval papal state and its complex system of “God-dictated” laws and customs: It is not only that the medieval catholic community of believers receives God’s message through the voice of the Pope and the comments of its designated interpreters (the canonists), but the interpreters also create, shape and manipulate their God through their interpretive activities.

15As Legendre states about the function of the medieval pope: It is the incarnate voice of the Law, “viva vox juris” (Legendre 1974: 75), and at the same time “the answer of the crowd to the crowd” (Legendre 1975: 71)³. This principle can also be detected in any other system of belief such as politico-historical imperatives, coercive forces of the market, or, for that matter, the power of algorithms to provide scientifically backed guidance – from business strategies to public governance. And while the new messengers of our present-day big Others (tech corporations and their audience-grabbing CEOs, policy makers, regulative bodies, etc.) are seemingly merely proclaiming unshakeable truths about the economic, technological, environmental realities and necessities we are facing, with all the seemingly unavoidable benefits and sacrifices they entail, they are also, like the pope and his canonists of yore, actually creating, shaping and manipulating these very realities, with the “crowd” of their audience (the present day community of believers, customers and suppliers, providers and consumers of digital products) participating in the construction of their narrative, or “crowdsourcing” it as it were.

16Legendre subsumes the cultural, social and political implications of this ambivalent relationship under the term “Society as Text” (Legendre 2001) – with the question as to who is actually writing and who is being written always remaining undecided. And this very undecidability between product and producer opens up room for interpretation: the freedom of human subjects to extract new meaning out of the reference text, not just reading it, but interpreting it, confronting their visions and decisions referring to a big Other with the question: Are the ostensible messages from the big Other really what we take them to be? Are we confronted with unequivocal truths of inexorable market dynamics, political necessities, socio-biological facts? Or are these messages mediated by shortsighted, biased, inadequate interpretations that may be revised, reshaped, reformulated so as to convey more nuanced, less coercive or even entirely different truths? It is by tackling these questions, by interpreting the message of the big Other, that the subject shapes and designs the contents of this big Other.

17So, to come back to the image of the Mechanical Turk, what it shows us is precisely this: while the non-human big Other seemingly automatically generates meaning out of chaos, giving a sense, a direction to the subject which looks up and refers to it – it actually turns out to be generated and steered by that very human subject. As Pierre Legendre further clarifies in this context, the danger to avoid here is relegating this steering function of the subject (i.e. the fact that it’s the player in the box who is in charge, not the Turk that magically leads the game) to oblivion, misconceiving the big Other not as the output of human intentions, expectations etc., but as the expression of an objectively determined truth: not as a narrative to be interpreted, but as a formula to be applied (Legendre 1982: 155).⁴ The absolute Other thus constructed dictates orders, which are to be heeded unconditionally, without letting any room for interpretation. According to Legendre, however, interpretive freedom is not just a benefit to be enjoyed, but, first and foremost, a duty to be accomplished (Legendre 1992: 392f): what is required from the subject is hard (interpretive) work, whereas, strictly speaking, the narrative construction of an absolute Other who guarantees meaning without the subject's intervention (a God’s will, an economic necessity, an algorithm dictating the only right path to follow) stems from the desire to get rid of the burden of this interpretive work – the desire to be absolved from the weight of responsibility for one’s actions and decisions. It is something of a lazy shortcut that seemingly leads us to a divine epiphany or a scientific breakthrough delivering a universal formula, but is actually just an attempt to avoid the effort of reflection, a refusal to think. Economically speaking, it is an attempt to outsource liability, or, as Legendre puts it: “to abandon oneself, to surrender to something that knows absolutely” (Legendre 1982: 155), i.e. to succumb to the fantastic idea that a big Other leads the game, that our input is not required, that we can just relax and let ourselves be guided.

AI in data analytics: Deep Data as a big Other “that knows absolutely”

18And isn’t this “something that knows absolutely” precisely the kind of big Other suggested by ideological representations of present day’s data analytics? As social scientist Rob Kitchin argues, the purpose of data analytics does no longer consist in using data to check the validity of a given theory, but rather in directly accessing insights generated by the data themselves (Kitchin 2014). Deep Data as a new big Other speaks for itself, automatically generates knowledge and meaning, in short: it renders obsolete the human theorists with their imprecise hypotheses (ibd.). A case in point of such a post-theoretical analytical practice is, according to Kitchin, the software company SymphonyAI, formerly Ayasdi, which launched its services under the label of “automated insight discovery”, advertising them as follows: “Today, ‘Big Data’ Fails to Solve Big Problems Because Query-Based Approaches are Flawed” (Ayasdi AI LLC 2013) – i.e. queries are per se postulated as epistemologically detrimental, whereas the problem solving strategy proposed by Ayasdi ostensibly offers the benefit of directly taking the short cut to the answers without losing time asking unproductive questions. “The Ayasdi Breakthrough” consists precisely in “Discovering Answers to Questions You Didn’t Know to Ask” (Ayasdi AI LLC 2013).

19This is a claim that computer scientist Eric Siegel would certainly endorse for the whole scientific and economic branch of data analytics. In his book Predictive Analytics. The Power to predict who will click, buy, lie or die, featuring a picture of a fortuneteller’s crystal ball on its front cover, the author sums up the methods and objectives of his discipline with the stunning remark: “We usually don’t know about causation, and we often don’t necessarily care. The objective is more to predict than it is to understand the world. It just needs to work; prediction trumps explanation” (Siegel 2016: 132).

20Against the backdrop of such a fantastic narrative presenting AI as access to automatic insights, i.e. to knowledge without understanding or reading without interpretation, Posner and Weyl’s concept of data as labor proves to be a constructive critical tool: Through their deconstruction of the idea of a magically self-generating surplus value uncoupled from any noteworthy labor inputs, the authors also open new perspectives on the epistemological surplus value of data analytics: Just as the development of AI through data collection is not simply based on isolated brilliant inventions, but is rather to be considered as the work output of vast crowds of data providers and annotators, the deployment of AI in data analytics, too, is not just the implementation of a finished product, but a human work process, i.e. a complex set of intentions and actions which necessarily involves multiple possibilities and responsibilities.

21It is this very complexity and multiplicity of agencies and responsibilities in both the development and the deployment of AI-driven tools and systems that needs to be accounted for if what we want is to understand and shape the world instead of predicting its ostensibly automatically generated future. If we relegate these underlying complexities to oblivion and instead choose to marvel at the conjured appearance of smooth, artificial production of “insights”, we will sooner or later inevitably stumble over a kind of ironic twist in the narrative of the digital revolution as paving the way to progressive, knowledge-based economies/societies: Historically, the dichotomy of extraction- vs. knowledge-based economies ascribes corruption, dysfunction, inequality to the former, and integrity, sustainability, democracy to the latter. More recently, however, economic, political, and cultural narratives around AI-driven tools and systems increasingly rely on the implicit premise that the massive amounts of data used to build and maintain these systems are to be treated as though they were extractive raw materials, while tech giants enjoying preferential access to these materials use the disproportionate benefits flowing from the technological transformation of these materials to wield increasing influence on state and world politics. Thus, with the digital turn, not only do we encounter the rise of phenomena typical for extraction economies, such as corruption of liberal market principles (increasing monopolization and erosion of antitrust safeguards) and threats to democratic structures (concentration of capital in tech giants leading to quasi-oligarchic powers such as increasing control of information outlets, influence on state governments etc.). But also, with the growing reliance in matters of policy making, jurisdiction, compliance and control, etc. on automated systems and technological tools providing ready-made insights required for complex predictive and decision-making processes, it is the knowledge itself in our so-called knowledge-based economies that is increasingly treated as extractive raw material that calls for the physical effort of mining instead of the mental effort of reflection, critical analysis, and in-depth understanding. Paradoxically, the risk faced by our digital knowledge-based economy is thus that not only our economy but also our very knowledge increasingly depend on what is treated as extractive raw materials. To avoid this risk, it is crucial to look beyond the appearance of AI-driven tools as automated access to insight and knowledge, and scrutinize this present-day “answer of the crowd to crowd” (Legendre 1975: 71) as to its hidden meanings and as yet unexplored potentials. This is precisely what is at stake in the demand “to go beyond neural nets and statistical pattern recognition to instead ask what is being optimized, and for whom, and who gets to decide” (Crawford 2021: 9, italics orig.).

The self-steering AML-Turk

22To further explicate this point, let us take a look at the purpose and the functioning of SymphonyAI products – “innovative software and solutions for regulatory compliance, fraud, and financial crime detection” (SymphonyAI , official website) – most prominently so-called AML (AntiMoneyLaundering) and KYC (KnowYourCustomer) solutions. As I would like to show, these financial compliance tools are an essential part of a fantastically autonomous system, a whole market even, seemingly smoothly running according to its automated internal rules, but at the same time functioning as a direct expression of a human failure – the refusal to formulate decisions and the endeavor to outsource responsibility perceptible in the AML policy making processes.

23The purpose of AML systems can be briefly summarized as follows: Pursuant to a range of international and state regulations, financial institutions, such as banks or investment companies, are under the obligation to put in place systems and processes so as to prevent the circulation of “dirty money”. To facilitate this task, software companies such as SymphonyAI specialize in elaborating time- and cost-effective AI-solutions helping banks to comply with their obligations. Based on a thorough knowledge of the banks’ customers (data on their previous business history, public exposure, criminal records, etc.), AML-software scans transactions for suspicious activities hinting to an illicit provenance or destination of funds, and issues so-called SARs (Suspicious Activity Reports) to be submitted to national FIUs (Financial Intelligence Units) for further processing and potential transmission to law enforcement authorities.

24Now, what SymphonyAI offers to its customers, financial institutions seeking to comply with government regulations, are predictive and generative AI products allowing them to “move toward a crime-free future” (SymphonyAI, official website). To be sure, AML-software has been extremely productive lately. In Germany, for instance, the number of SARs transmitted to FIUs more than doubled within three years, rising from 59.845 in 2017 to 144.005 in 2020 (IT Finanzmagazin 2021), and the picture is quite similar in a range of OECD-countries adhering to AML standards (Wintour 2021 a).⁵ Meanwhile, however, the United Nations assess the amounts of money laundered globally as constantly rising, with the amount of illicit funds actually detected representing only a fraction of a percentage point of the total criminal gains, estimated at about 3.6% of global GDP (Pol 2020). Ronald F. Pol, author of the paper “Anti-money laundering: The world’s least effective policy experiment?”, comments this development with the remark:

Letting criminal enterprises retain up to 99.95 percent of criminal proceeds, the modern anti-money laundering experiment unwittingly enables, protects and supports terrorists, drug, human, arms and wildlife traffickers, sex and labor exploiters, and corrupt officials, fraudsters, and tax evaders on a global scale (Pol, 2020).

A plea for an effort to understand the world

25One could argue that the AML system is a perfect example for the practical implementation of Eric Siegel’s predictionist conviction that “the objective is more to predict than it is to understand the world. It just needs to work”. The AML-system actually does work, it even accounts for a flourishing digital market of its own, with tech companies providing financial institutions with software generating floods of SARs, providing FIUs flooded with those SARs with software filtering out high-risk SARs to be treated with priority, and thus, incidentally, also providing criminal entities with clear guidelines for the construction of low-risk profiles undetectable by the system. So, one could say that the system works autonomously and automatically, and: completely independently from the external reality of money laundering, i.e. without any noteworthy impact on it.

26Take the example of another AML-software company, Sentinels, whose website features an impressive real-time count-up of money being laundered globally. Naïve laymen might be rather surprised to learn that while they explore Sentinel’s AML-solutions, astronomical sums of money will be laundered within a matter of seconds, notwithstanding the company’s mission “to solve the problems in transaction monitoring solutions across the globe” and “detect money laundering in real time” (Sentinels, official Website). However, this apparent paradox reflects the everyday reality of money laundering: On the one hand, the OECD blacklist of so-called unco-operative tax havens not adhering to AML standards has continually diminished and actually completely disappeared in 2009 (OECD 2018), suggesting that the ML-phenomenon is under control. On the other hand, revelations from Panama to Pandora papers keep appearing, suggesting that the money laundering business is still thriving. An interesting anecdote in this context are the G7-Talks of December 2021, preceding the Russian invasion of Ukraine. As part of a discussion on possible dissuasive measures against a Russian aggression, “the UK foreign secretary, Liz Truss, said all forms of economic sanctions against Russia were on the table if it made an incursion into Ukraine, and hinted she may be prepared to look again at the UK’s laws against money laundering” (Wintour 2021b). It might come as something of a surprise that these laws needed to be looked into, given the supposedly smoothly working regulative machinery of AML already in place. However, according to sources such as the UK foreign affairs committee, the workings of this machinery never actually prevented London from being a “laundromat for Russia’s dirty money” (Ellyatt 2018) – a role provoking “dismay and frustration” amongst US-officials: “The fear” so they said, “is that Russian money is so entrenched in London now that the opportunity to use it as leverage against Putin could be lost” (Curzon 2022).

27The recent experience of political efforts to introduce large-scale financial sanctions against the Russian regime and the difficulties encountered in their practical implementation sheds a disturbing light on the effectiveness of regulative and controlling mechanisms at work within our financial landscapes. Even while the thriving AML-industry reached a market size of USD 2.34 Billion in 2022 and was expected to grow to USD 9.61 Billion by 2023 (Reports and Data 2023), at the same time, colossal amounts of funds of questionable provenance continued to flow through our economic systems. How are we to account for such a discrepancy between the declared purpose of a seemingly working system and its actual outcomes? According to Ronald Pol, the problem does not lie with the AML-solutions applied by banks, but with the design of the policies pre-defining the implicit function of these solutions – the fact that regulators delegated the entire AML issue to banks, thus outsourcing their responsibility and defining money laundering not as a problem of public interest to be fought by law enforcement institutions, but rather as a private business risk of banks, whose interest, of course, is not to fight crime, but to make sure the money they deal with is labelled “clean”. Looking at the actual output of the AML-system, we might as well conclude that it was never designed to actually remove dirty money out of financial institutions, but rather to deliver proof that efforts have been made to clean money from suspicion, as a guarantee for the regulators, policy makers, representatives of the public interest, that the problem has been addressed, as a certificate of equitable finance that the banks and their software providers are required to produce for the sake of formality. Strictly speaking, the purpose of AML is not to fight profit driven criminal enterprises, but to make sure that the money circulating in our financial institutions can be considered clean, which of course is an entirely different matter.

28That the reality of money laundering remains untouched and unimpressed by such measures is not an accident, but part of the design of these measures, and it is in the failure of a perfectly working system to obtain any impact on an external reality that we see: The AML-System is not only a set of rules and their automatic application, but a narrative, not only rendering explicit outcomes – a mechanical Turk waging a strategic game against organized crime – , but also conveying the hidden intentions of the human regulator sitting in the box, outsourcing the responsibility for crime prevention to financial institutions, while indulging in the tacit belief that it would be a waste to throw away the baby with the bathwater: that once it’s washed clean and made presentable with a compliance certificate, the formerly dirty money could as well be put to productive use. This belief is accompanied and sustained by the uneasy question: What would happen if our financial markets and economic sites were no longer supplied with laundered profits from the theft of collective resources that are cheaply fed into our heating systems, gas tanks, car and cell phone batteries? It is the refusal to fully confront this question that prevents the AML policies from moving beyond the stage of mere conscience-laundering, with financial institutions simultaneously washing dirty money and their hands of the matter.

29The example of the AML-system shows how multiple divergent political, economic, legal, etc. interests with oftentimes conflicting motives and purposes are intertwined and reflected in a system that appears to be a straightforward regulative policy which only needs to be properly applied to ensure a smooth running of ethically and economically efficient financial operations. Considering all of the covert factors contributing to the setup and implementation of AML technology, this technology cannot possibly “succeed” in the sense of establishing a financial system entirely independent from funds of criminal or questionable provenance. They can only serve to create the stable appearance of “clean” while still relying on variable scales of “dirty”. The crucial question is therefore not whether the AI-driven AML system works (e.g. how many, how accurate and how relevant SARs it generates), but to what ends it is used within the larger framework of regulative policies. To obtain any noteworthy impact on money laundering, what might be required is not more optimizing of automated processes already in place, but rather, precisely, an effort to “understand the world”. Generally speaking, when it comes to regulative procedures, the highest stakes might not be in technology and its optimization for the application of rules, guidelines, and standards. The obsession with ever more performant technical tools (or mechanical Turks) for their application actually obfuscates the fact that it is the definition of these standards, the articulation of rules, the formulation of guidelines that is at stake. It’s the complex interplay between economic interests, political motives, implicit or explicit expectations of lobbies/public stakeholders/electorates, etc. hiding in the box which need to be revealed, re-evaluated, and revised. And in order to be able to do this, we need to acknowledge that, no matter the technical progress we make, the responsibility for any actions or inactions within our socio-politico-cultural reality will never be something that can be outsourced. 

References

1. Amazon Mechanical Turk. https://www.mturk.com/ .
2. Anderson, Martin (2022): Amazon Mechanical Turk Pays Less Than 40% of US Minimum Wage, Research Suggests. In: unite.ai, December 9. https://www.unite.ai/amazon-mechanical-turk-pays-less-than-40-of-us-minimum-wage-research-suggests/ .
3. Ayasdi AI LLC (2013): Press Release. January 16. https://www.fiercebiotech.com/it/ayasdi-launches-10-million-from-khosla-ventures-and-floodgate-to-pioneer-automated-insight, consulted on 14 February 2024.
4. Crawford, Kate (2021): Atlas of AI. Power, Politics, and the Planetary Costs of Artificial Intelligence. New Haven, CT: Yale University Press.
5. Curzon, Michael (2022): “Putin’s Money is in Knightsbridge”. US in Broadside at UK as Moscow Tears Alliance Apart. In: Daily Express, January 28. https://www.express.co.uk/news/politics/1557168/russia-news-vladimir-putin-money-Moscow-us-uk-alliance-ukraine-joe-biden .
6. Douetteau, Florian (2018): The Tyranny of Appearances in AI and the Tale of the Mechanical Turk. In: Medium, September 19, https://medium.com/@fdouetteau/the-tyranny-ofappearances-in-ai-ef3c34515711 .
7. Ellyatt, Holly (2018): London is a Laundromat for Russia’s Dirty Money, UK Report Warns. In: CNBC, May 21. https://www.cnbc.com/2018/05/21/london-is-a-laundromat-for-russias-dirty-money-uk-report-warns.html .
8. Hayles, Katherine N. (1999): How We Became Posthuman: Virtual Bodies in Cybernetics, Literature, and Informatics. Chicago: University of Chicago Press.
9. Kitchin, Rob (2014): Big Data, New Epistemologies and Paradigm Shifts. In: Big Data and Society, April–June: 1–12.
10. Legendre, Pierre (1974): L’Amour du censeur. Essai sur l’ordre dogmatique. Paris: Seuil.
11. Legendre, Pierre (1982): Paroles poétiques échappées du texte. Leçons sur la communication industrielle. Paris: Seuil.
12. Legendre, Pierre (1992): Les enfants du texte. Etude sur la fonction parentale des Etats. Paris: Fayard.
13. Legendre, Pierre (2001): Da la Société comme Texte. Linéaments d’une anthropologie dogmatique. Paris: Fayard.
14. Nash, Jim (2022): AI is Already Thinking the Way We Do. In: BiometricUpdate.com, April 7. https://www.biometricupdate.com/202204/ai-already-is-thinking-the-way-we-do .
15. No author (2023): Anti-Money Laundering (AML) Market. Report ID: RND_002045. In: Reports and Data, July. https://www.reportsanddata.com/report-detail/anti-money-laundering-market .
16. No author (2006): Artificial Artificial Intelligence. Software. For many tasks, humans outperform computers. So why not farm out computing tasks to people, not machines? in: The Economist, June 10. https://www.economist.com/technology-quarterly/2006/06/10/artificial-artificial-intelligence .
17. No author (2021): Deutschland versagt im Kampf gegen Geldwäscher. In: IT Finanzmagazin, September 20. https://www.it-finanzmagazin.de/geldwaesche-wird-zu-einem-zunehmenden-problem-in-deutschland-125335/ .
18. Pol, Ronald F. (2020): Anti-money Laundering: The World’s Least Effective Policy Experiment? Together, We can Fix it. In: Policy Design and Practice 3 (1): 73–94. https://doi.org/10.1080/25741292.2020.1725366.
19. Posner, Eric. A. and E. Glen Weyl (2018): Radical Markets. Uprooting Capitalism and Democracy for a Just Society. Princeton, NJ: Princeton University Press.
20. Sentinels official Website. https://web.archive.org/web/20220814104852/https://www.sentinels.ai/.
21. Siegel, Eric (2016): Predictive Analytics. The Power to Predict Who Will Click, Buy, Lie or Die. Hoboken, NJ: Wiley.
22. SymphonyAI official Website. https://www.symphonyai.com/financial-services/get-started/.
23. Wintour, Patrick (2021a): UK’s Anti-money Laundering Measures need Complete Reset, says Report. In: The Guardian, December 8. https://www.theguardian.com/law/2021/dec/08/uks-anti-money-laundering-measures-need-complete-reset-says-report .
24. Wintour, Patrick (2021b): G7 leaders warn Russia all Sanctions on Table over Ukraine Border Buildup. In: The Guardian, December 12. https://www.theguardian.com/world/2021/dec/12/g7-leaders-sanctions-russia-buildup-ukraine-border .
25. Žižek, Slavoj (2008): Enjoy your Symptom! Jacques Lacan in Hollywood and Out. London: Routledge.