Online manipulation in China? : rational choice perspective on the incentive structure of China's tech giants Baidu, Alibaba and Tencent

(1)

Thesis for the master’s degree program in Political Science at

Universiteit van Amsterdam

Online Manipulation in China?

-

Rational Choice Perspective on the Incentive Structure of China’s Tech

Giants Baidu, Alibaba and Tencent

June 21, 2019

Research Project:

21

st

_{Century Power Shift: China’s Rise in the Global Political Economy}

Student ID: 12295196 Supervised by

Raphael von Aulock

Julian Gruin PhD

2nd_{semester MSc. program Political Economy}

Jan van Galenstraat 1004 Faculty of Social and Behavioural Sciences NL-1061 AZ Amsterdam Nieuwe Achtergracht 106

Tel: +49 (0) 1744208220 NL-1001 NB Amsterdam E-Mail: raphael.vonaulock@student.uva.nl E-Mail: J.Y.Gruin@uva.nl

(2)

2

I Abstract

As Research on the phenomenon of online manipulation has finally caught up with current technological developments, this paper addresses the research gap that has resulted from a lack of treatment of the topic within the context of China and of its leading tech companies Baidu, Alibaba and Tencent. Faced with the difficulties resulting from manipulation being inherently hidden, this paper opts for a semi-theoretical approach that mostly draws on the existing literature, secondary sources and several expert interviews to assess the wider choice architecture that these companies currently face regarding the potential costs and benefits of engaging in online manipulative practices for the purpose of obtaining economic gain. Despite the ample limits of such a broad outside-in approach the analysis uncovers how the dynamics between different actors lead to a current situation where the country’s technological leaders are in a seemingly perfect position for exploiting their market power without having to fear any severe consequences.

II Introduction

The proliferation of digitization has given rise to tremendous opportunity and improvement across much of the globe and in many areas of personal life. Despite all the advantages of an economy that increasingly relies on digital platforms for just about anything, there are unavoidably some problems and threats emerging from these changes. One aspect that has come under increasing scrutiny by academics is that digital markets and the technologies that come with them can facilitate online manipulation, i.e. “the use of information technology to covertly influence another person’s decision-making” (Susser et al. 2019: 24-25), with unprecedented effectivity and at scale (Calo 2014). Because technology is moving at a remarkable pace, research addressing the topic of online manipulation has, so far, been mainly preoccupied by catching up with the latest developments. Studies have almost exclusively focused on theoretical elaboration (Calo 2014; Yeung 2017; Susser 2019) or the evaluation of the few known cases of manipulation that make use of tools like big data and artificial intelligence (AI) to influence individual behaviour (Susser et al. 2019; Calo and Rosenbladt 2017).

(5)

5

This study aims to apply the understanding gathered from previous research to the Chinese context by assuming the perspective of country’s largest technology companies Baidu, Alibaba and Tencent, that are commonly known as the BAT. Because of China’s rising influence in the global economy and since the BAT are not only China’s leading digital platform providers but also spearhead the national AI initiatives in their respective fields (Webb 2019), it seems rather astounding that their presumed ability to influence in hidden ways has been almost entirely neglected throughout the English literature on online manipulation, as western academics have addressed the activities of American companies within liberal democracies.

With a deep diving investigative study into each company not being possible due the limited scope of this paper and because of the glass floor that is the language barrier and scarce data availability, this paper doesn’t treat each company individually but mostly sees the BAT as representatives of China’s exceptional digital economy.

Grounded within the wider framework of rational choice theory and drawing on adapted criminological methods embedded in the theory (Becker 1968), this paper develops a variable-based approach that allows for a holistic analysis of the incentive structure that would-be manipulators are facing. Taking the Chinese focus into account this inquiry is best expressed in a dichotomous fashion through the following research question:

Is the use of online manipulative practices to influence Chinese individuals through their digital platforms an opportunity or a threat for China’s tech giants, within their domestic socioeconomic context?

I would like to point out that, despite the hypothetical structure of the question, the purpose of this paper merely to shed light on the dynamics that tech companies face within the Chinese socioeconomic context, while previously unknown activities are neither uncovered, nor are they alleged.

Starting this paper by reviewing the literature on online manipulation or setting up a political economy context could be valid approach, yet, due to the methodology of this study, such discussions would already directly contribute to the analysis of the research question by partially answering sub-questions like Why would the BAT want to manipulate? and How could the BAT manipulate?. Because the goals of companies are assumed to be largely the same and since the mechanics of online manipulation are similar regardless of geographic location or individual perpetrator, most aspects of the analysis will have to be addressed exclusively at the Chinese level. These topics are, therefore, left untouched until they are addressed later on, since their

(6)

6

treatment at the beginning of the thesis they would otherwise cannibalize the analysis, or lead to confusion by splitting it, if nothing else.

Postponing these sections is definitely unconventional, yet I argue that it ultimately benefits the continuity of this paper, since the study can systematically move forward with a keen understanding of who is being studied and how. Why this makes sense is outlined in the following paragraphs that detail the overall structure, briefly touch on the methodology and hint at the overall results.

As the evaluation of the research question essentially consists of a cost versus benefit analysis the first chapter starts by outlining rational choice theory as a means to make incentive structures palpable. Subsequently, social power, data as the root of it and online manipulation as a special case of are introduced. Following the resulting understanding about the unique properties of manipulative practices and after settling on the scope of analysis, a framework that evaluates the probability for their use along four distinct variables is introduced based on criminological methods to predict the occurrence of crime. Essentially, the simple cost versus benefit analysis at the basis of rational choice theory is expanded by a probabilistic dimension that takes into account how, similar to crime, the use of manipulative practices could become uncovered (Pc2),

which would lead to the incurrence of additional costs for the manipulator (C2) on top of the

efforts already needed to reliably engage in these practices in the first place (C1). As a result,

varying outcomes would limit the perceived utility from their use (U) by more or by less. While these variables can and will be broken down to along various strands to give a more detailed insight into the choice architecture of the companies, their relationship with each other in this simplified form can be expressed through the following ratio that is derived from Becker’s economic approach to crime (1968):

C1 > (Pc2 * (U + C2) + (1- Pc2) * U)

Due to the unavailability of the quantitative data required to support a mathematical approach, the analysis is conducted in a qualitative way. As such, the equation can and will not be formally solved, however, it structures and guides the inquiry and based in its results the subsequent discussion will attempt to formulate a tendency for or against the benefits of using online manipulative practices from a company perspective. This will ultimately serve as the answer to the research question.

The second chapter directly moves into the argument of why this paper focuses on manipulation to advance the economic goals of companies instead of looking at political influence and governments and it also outlines why the BAT and the wider Chinses case are especially relevant

(7)

7

not only to the Chinese population but possibly also to the west. Following this, the methodology outlines how the research question will be addressed. Essentially, the variables are not individually explored but they are carried along throughout an analysis of three sub-questions, two of which were already introduced just above. As much of the analysis is applicable to other parts of the world the China specific parts of the analysis mainly address the risk dimension that was raised by the research question and which is embodied in the third sub-question namely, ‘Would manipulation be worth it in China?’.

Each chapter of the analysis starts from a broader and sometimes outright theoretical angle and is subsequently narrowed down to the Chinese and ultimately the company level, when needed. The analysis proper is conducted by means of reviewing existing literature, secondary and sometimes primary sources and, most notably, by drawing on three expert interviews that were conducted to give this mostly theoretical study an additional practical dimension that is grounded within the hands-on experiences of the interviewees from working with and within the Chinese market as well as in the field of economic psychology respectively.

During the discussion learnings from the sub-questions are distilled back into the variables to facilitate the evaluation of the research question. The broader takeaway is that, from an economic perspective, much could be gained by the BAT from the use of these practices at an arguably very low cost, while there is little chance that that they will adequately be held to account even in the event that their activities are uncovered. It is fully clear that the results are not final and merely a snapshot of the current situation. Nevertheless, this exploratory approach is justified by the danger of manipulation itself, the clear research gap and China’s overall growing relevance while its value comes from holistically addressing the developments so far, thus revealing several areas where further scrutiny is needed to get a firmer understanding of the risks of online manipulation in China.

(8)

8

Chapter 1: Operationalising the Invisible

This first chapter is characterized by the attempt to make manipulation palpable for the purpose of qualitative analysis. Following the introduction of rational choice theory as the wider theoretical framework of this paper, different types of social influence are presented and delimited from manipulation. Based on the understanding that

the understanding of data is at the root of the power to influence others, online manipulation and online manipulative practices are introduced as specific manifestations of it. Finally, because manipulative practices are hidden and harmful, rational choice theory is refined from a criminological point of view that ultimately allows the

probability for manipulation to occur to be predicted within certain limits.

Rational Choice Theory

We make choices every day, all day long and while one choice seems insignificant, the next one can be life changing, with some decisions being intuitive and others very hard to make. Philosophical debates about the existence of free will can, at this point, easily be left aside, since any social rule that leads to consequences, no matter if positive or negative, necessarily assumes the existence of at least some form of decision-making responsibility inherent to mature and healthy human beings. In different parts of the world, laws and norms give different types of guidance but the choices we make, really boil down to individual opinions and priorities, no matter how clear cut or incomprehensible the ‘right choice’ might seem to everybody else. This ability to make choices is called autonomy (Raz 1986) and, as will be discussed throughout this chapter, it is decisively impaired by manipulation. In fact, there are countless ways by which choices are shaped from the outside, yet, not all go hand in hand with a limitation of autonomy (Susser et al. 2019, Spreer 2018). Before addressing manipulation more closely, a better understanding of choices in general and how they are influenced is necessary.

Across the social and behavioural sciences, the notion of actors making decisions based on their preferences and acting on them in a purposeful way is called rational choice theory and it ultimately boils down to weighting costs against benefits based on the individual perception of

utility. The concept of utility was introduced by mathematician Daniel Bernoulli (1738) who introduced the notion that personal preference can come to be valued higher than mere mathematical gain. Similarly, the works of Adam Smith (1776) and John Stuart Mill (1836) popularized the idea of a homo economicus that acts in his own interest and aims to maximize wealth and happiness has long dominated economic thought.

(9)

9

Still, the assumption of entirely rational actors unavoidably lends itself to criticism. Therefore, the rational choice theory has been adapted and expanded over the years. Two of these limiting concepts are crucial for this paper.

First, bounded rationality was introduced by Herbert Simon in the late 1950s and assumes that, due to the limited time, information and cognitive ability to calculate all possible outcomes, actors are only able to make sufficiently optimal decisions instead if ideal ones (1957).

Second, while the rational model of decision making was long assumed to be the only one, this changed when Kahneman and Tversky (1979) proposed Prospect Theory. It has displaced utility theory as the exclusive model for economic decision making and explains why actors that make decisions under risk do not only consider the expected gain in utility but also factor in possible losses in relation to a reference point. Furthermore, the theory assumes that actors will become more risk seeking when faced with losses either way (Ibid).

Since both these limitations of an entirely rational decision model guide individual choices, they also apply to larger structures such as companies and governments. Because no actor is omniscient, bounded rationality will always apply to some extent, yet, especially the risks of irrational behaviour as a result of the dynamics presented in prospect theory can be mitigated through adequate structures of (corporate) governance (Wen 2010). Arguably, the same applies to bounded rationality because, as we have come to understand throughout human history, absolutism is rarely the best model of governance, as a single person can’t possibly be expected to consistently make the optimal choices, even when acting in good faith.

Decisions never take place in a vacuum and while an actor might make an autonomous judgement according to their inherent preferences and goals, they can hardly do so while disregarding the relevant social context (Susser et al. 2019). This means that, depending on the physical and especially the psychological stimuli someone is exposed to, their choices will turn out differently. These influences can be referred to as ‘incentive structure’ (Dowding 1996) or as ‘choice architecture’ (Yeung 2017) and they shape all aspects of society by determining what options are reasonably available to rational decisionmakers in the first place, be it because of and external limitation or because acting in a certain way is regarded as ‘simply not an option’. As a result, autonomy can never be absolute, but it is worth protecting nonetheless as it determines whether our decisions are our own or those of others (Raz 1986; Calo 2014; Susser et al. 2019). Again, the same applies to companies and governments who are faced with their own ‘choice architectures’.

(10)

10

Social Power

Any Influence, especially when intentionally being exerted upon by one actor over another, can be referred to as power, of which there are two fundamental kinds that can be nested in rational choice theory (Dowding 1996). First, there is the ‘power to’, which can be regarded as the most basic instance of influence, as it only requires one actor. Often described as ‘outcome power’ it is “the ability of an actor to bring about or help bring about outcomes” (Dowding 1996: 4). Second, the ‘power over’ entails at least two actors and necessitates the ‘power to’ (Dowding 1991; Dowding 1996). It can, therefore, be described as ‘social power’, namely, as “[T]he ability of an actor deliberately to change the incentive structure of another actor or actors to bring about or help to bring about outcomes”, or in more direct terms: “A has the power to make B do x” (Dowding 1996: 4). Thus, exerting power becomes all about the influence over the perceived costs and benefits of another actor.

Considering the online and corporate focus of this paper, it is important to note that this ability does not exclusively extend to human beings. In line with Latour’s Actor-Network Theory (ANT), inanimate objects, say a piece of software, can be regarded as actors just as well (Muniesa 2015).

The actor in power (A) can change the ‘incentive structure’ in such a way, that choosing an option other that the one favoured by A (x), would result in a higher cost or a lower benefit for the actor that power is exerted upon (B). At the same time, as will be elaborated upon when dealing with manipulation, B is not necessarily aware that of the influence exerted over their choice architecture by A.

The social sciences often divide social power into the categories ‘hard power’, ‘soft power’ and ‘smart power’ (Wilson 2008; Nye 2009a), with the recent addition of ‘sharp power’ (Walker and Ludwig 2017). Whereas these terms are usually applied in the context of international relations, they are by no means out of place when talking about the influence of companies on individuals, since both are social interactions and the mechanisms remain essentially the same. These common demarcations are, therefore, highly useful for understanding different intensities of power along with some techniques of social influence that should be delimited from manipulation prior to diving into the heart of the matter.

(11)

11

Hard Power

Hard power is mostly attributed to the use of coercion (Nye 2009a), which can be understood as making someone an offer that they cannot reasonably refuse (Rudinow 1978). Aside from the projection of military or economic power (Wilson 2008), one interpersonal example for this would be to hold a loaded gun to someone’s head and while they could, again, technically still choose to not comply with whatever the demand of the moment might be (Susser et al. 2019), to resist coercion would take “heroism, madness, or something similarly extraordinary” (Rudinow 1978: 341), which is why hard power is usually very effective, despite rarely being absolute.

Soft Power

In the literature, the term soft power has been widely used and sometimes overused, as different understandings are often overlapping and sometimes contradictory (Nye 2009b). Therefore, it can be understood as almost any form of social influence that is present in the absence of hard power with “make B do x” characteristics (Cooper 2004; Nye 2009b). In the IR context, Nye originally presented it as a form of influence that facilitated compliance through persuasion and attraction (1990). While persuasion became the defining element for some (Wilson 2008), Nye later emphasized more on attraction (Nye 2009a; Nye 2009b). He argues that that persuasion alone cannot constitute soft power (Nye 2009b), however it is relevant to note that he refers to the narrow meaning of persuasion as a form of social influence (Nye 2009b). In broad terms, persuasion fits the IR definition of power given above (Wilson 2008: 114), meaning that any form of influence that alters an actor’s decision making can be regarded as persuasion (Susser et al. 2019). In narrow terms, however, persuasion is constituted by a rational attempt to influence and merely “means changing someone’s mind by giving reasons he or she can reflect on and evaluate” (Susser et al. 2019: 11).

Sharp Power

As recently as 2017 the term sharp power was introduced by Walker and Ludwig as actions that “pierce, penetrate, or perforate the information environments in the targeted countries” (2017: 6). These environments can include institutions, the media or academia, among others. Sharp power’s main methods of influence are “manipulation and distraction” (Ibid.: 10). Manipulation will be more closely defined in the following section. As if today, the use of sharp power has been explicitly attributed to authoritarian regimes who aim to export the oppressive techniques that

(12)

12

they have perfected at the domestic level, to increase their influence in the world (Walker 2018). While the cases of content manipulation described by the authors can be regarded as manipulative, their effectiveness is arguably a far cry from the manipulative abilities of private firms that will be described below. This is mostly because in the IR context, sharp power is strongly mediated in the sense that it usually has to cross borders and subvert existing systems along with their inherit checks and balances, before it can possibly affect its targets. In terms of the demarcation of sharp- against soft power it should be noted that the two are not mutually exclusive (Walker 2018) and, as argued by Nye (2018), a line can be crossed where soft power turns into sharp power. When exactly this happens, will be laid out in the next chapter as manipulative practices are discussed.

Smart Power

Lastly, combining the techniques of hard- and soft power in a, well, ‘smart’ way, smart power emphasizes on the idea that more favourable outcomes can be achieved when the two powers are sensibly applied to more complex situations (Wilson 2008; Nye 2009a). As Nye argues, “[i]n today’s information age, success is the result not merely of whose army wins but also whose story wins” (2009a: 162-163). In this sense, smart power can, at least for some actors, clearly also be extended to include the techniques can that constitute sharp power. Smart power itself doesn’t have a certain intensity in relation to other forms of power, however, it is arguably overall more efficient at attaining goals, since the use of smart power allows for the right type of influence to be applied to the right situation in order to get to a desired outcome. This makes it very similar to the principle of proportionality that is fundamental to the rule of law an important part of international law (Hirschberg 1981). The type of influence that is most adequate to get to the desired outcome of a given situation, as this section has demonstrated, is ultimately a matter of choice, regardless of the actor in question being a world power or merely one person influencing another.

Power from Knowledge

An understanding about the nature of social influence almost inevitably leads to the question of how that power can be obtained. The widely known dictum ‘knowledge is power’ provides an

(13)

13

answer that is as compelling as it is easy to understand. If know something, i.e. if they identify a promising gap in the market, they can act upon that knowledge with the certainty that the expected outcome will happen.

However, as anyone will likely have experienced on multiple occasions, it is also possible to act and reach the expected outcome based on a mere belief. Yet, no matter how strong they are, beliefs do not necessarily turn out to be true. Why this distinction would matter is equally simple to grasp, since true power can only result from knowledge as opposed with beliefs, who could turn out to be wrong at any moment, potentially bringing about unintended consequences. But how can one be sure whether they know or just believe?

Defining Knowledge

Setting knowledge apart from belief is a matter that is still subject to ample debate in various fields, most notably among epistemologists (Roush 2006) and system theorists (Rowley 2007; Frické 2007; Jennex 2019) Seemingly going as far back as Plato (Fine 2003), knowledge was long universally defined as being ’justified true belief’ (JTS), meaning that when a belief is not only justified but also true, then it is knowledge (Boghossian 2007). Of course, this gives rise to the question when a belief is sufficiently justified to be regarded as knowledge and, in short, there is still no definite answer and true knowledge arguably doesn’t even exist (Musgrave 1993). Additionally, the prevalent JTS approach was severely challenged when philosopher Edmund Gettier published an article where he presented a series of cases in which a subject has a justified true belief for knowing something and still ends up being wrong (1963). These so-called Gettier problems have resulted in the development of new and often rival theories of knowledge over the consecutive decades.

Notably, Sherrilyn Roush (2006) has introduced a theory that gives knowledge a probabilistic character, meaning “…that the better one’s evidence for a hypothesis is, the better one’s chances of knowing it.” (2007: 2). She further elaborates that “it is because we need justification that we need knowledge rather than mere true belief” (2007: 3). Roush weakened and combined the requirements of other theoretical approaches in favour of a model that has been described as closely related to the way that both science and technology work today (Goldman 2009; Godfrey-Smith 2009). System theorist Frické (2009) equally sees knowledge as being ‘weak’.

(14)

14

For the scope of this inquiry the classic JTS approach, enhanced by an understanding of the probabilistic aspect of knowledge from Roush’s approach, appears to be a more than sufficient definition of knowledge, since it emphasizes on the properties of how knowledge is applied on a daily basis. Knowledge can, therefore, be understood as belief that is sufficiently justified to act upon it and whose probability for being true is proportional to the amount of evidence supporting it. While this might not satisfy epistemologists, it is more than suited to lead to the next question regarding the root of social influence, namely, what evidence is and how it can be adapted to a purpose.

From Data to Knowledge

To understand the origin of knowledge, yet another dominant approach, namely, the so-called data, information, knowledge and wisdom hierarchy (DIKW) that is generally attributed to system theorist Russel Ackoff (1989), must be deconstructed. Again this dominant concept is a highly contested one (Rowley 2007; Zins 2007; Frické). As authors almost unanimously agree, the basis of knowledge can be found in data. In her critique of the model, based on aggregating definitions of the different concepts of the DIKW hierarchy throughout the literature, Jennifer Rowley concludes:

“Typically information is defined in terms of data, knowledge in terms of information, and wisdom in terms of knowledge. However, there is less consistency in the description of the processes that transform elements lower in the hierarchy into those above them, and some consequent lack of definitional clarity.” (2007: 177).

A similar and equally inconclusive verdict was also given by Chaim Zins (2007) following his survey-based analysis of the DIKW model. While both authors have merely acknowledged the ambiguity of the definitions presented by the DIKW hierarchy without severely challenging its validity, Martin Frické has harshly criticized the model as relying on operationalism and inductivism, both outdated philosophical ideas. He has, therefore, argued that the model “…should no longer be part of the canon of information science, and such related disciplines…“, nevertheless, he acknowledges that this “…would leave an intellectual and theoretical vacuum over the nature of data, information, knowledge and wisdom, and their interrelationships, if any.” (2009: 131). Since then, both his and Jennifer Rowleys article were widely cited, yet, ten years

(15)

15

on, there seems to be no generally accepted alternative (Kebede 2010; Case and Given 2016; Jennex 2019).

To avoid a lengthy discussion about which single definition of the terms might be the most accurate, it seems reasonable to take advice from Keen and Tan (2007), who have argued that that overly precise definitions of these concepts would come at the expense of progress in research. Therefore, the missing elements will be defined according to a 2004 analysis by Bellinger et al., that is among the more concise ones, and doesn’t seem to have had significant opposition over the years.

With wisdom being higher up in the hierarchy it should be addressed first. It appears to be present if one grasps all the relationships of cause and effect that make a system work, while knowledge merely approximates such holistic understanding and is, therefore, not exhaustive (Bellinger et al. 2004). In that sense, wisdom is an ideal, at least within the context of this paper. The terms data and information are often used interchangeably in daily life, yet, the two are clearly distinguishable from one another (Bellinger et al. 2004; Liew 2007; Rowley 2007; Zins 2007). Information can be inferred from data by analysing it and thus creating a relationship, such as cause and effect (Bellinger et al. 2004: 3), while data merely exists and is always correct (Rowley 2007). Information on the other hand, i.e. what we infer from that reality, can be false, just like apparent knowledge can turn out to have been nothing more than a belief. ‘Wrong data’, therefore, simply becomes ‘inaccurate information’ (Frické 2009).

Moving between the layers

Bellinger et al. (2004) have identified understanding as the transition mechanism from one level of the hierarchy to the next, thus contradicting Ackoff who had originally seen that as a fifth layer to the hierarchy (1989). Bellinger and his co-authors characterize understanding (Fig. 1) as a “…cognitive and analytical…process by which I can take knowledge and synthesize new knowledge from the previously held knowledge.” (2004: 1).

(16)

16

Figure 1: The Classic DIKW Hierarchy

Source: Own Illustration based on Bellinger et al. (2004: 2)

Gregory Chaitin sheds further light on this concept by writing that "…comprehension is compression…The simpler the theory, the better you understand something” (2006: 77). This compression approach also fits into what we have learned about knowledge so far, meaning that the process of understanding is mainly the process of informed theorizing based on a, hopefully accurate, depiction and interpretation of the data at hand. This understanding obviously doesn’t formally address how the process of understanding works, however, as will be laid out in the following throughout the analysis it is mostly an arbitrary one, even in corporations.

(Online) Manipulation according to Susser et al. (2019)

Be it in interpersonal relationships or in the many political examples throughout history, psychological manipulation aims to subvert a target’s ability to act upon their free will in a

(17)

17

hidden and usually harmful way (Handelman 2009). It can be defined as “imposing a hidden or covert influence on another person’s decision-making” (Susser et al. 2019: 22).

Manipulators intentionally exploit their targets’ trust in them, to make them act on the manipulator’s behalf without realising that they are doing so. This makes manipulation potentially useful in any situation where the manipulator wants to make a target to act in a certain way, without being recognized as the source of influence.

Usually, manipulation involves the deception of a target, however, Susser et al. have delimited deception from manipulation by pointing out that, while being a useful tool when attempting to manipulate, the latter can also take place without “instilling false beliefs” (2019: 18). Instead, they argue, manipulators can also prey on emotions or desires as well as on psychological weaknesses that are called cognitive biases (Idib.) but can also be referred to as behaviour patterns (Spreer 2018). Their discovery relates back to the works of Kahneman and Tversky (1979), among others.

The harm from covertly making others act according to one’s goals, if nothing else, is ultimately a harm to their autonomy, since the decisions of the manipulated target are not their own anymore (Calo 2014; Yeung 2017; Susser et al. 2019). Beyond that, the harm from successful manipulation is as unlimited as the reasons it might be applied for and as a result the potential rewards for the manipulator are theoretically only confined by their ability to manipulate successfully.

Only recently the manipulation was formally introduced to the digital sphere through the definition that Susser et al. settled upon, following their review of the literature about the threats of digital markets and new technologies. They have defined online manipulation as “the use of information technology to covertly influence another person’s decision-making” (2019: 24-25).

(Online) Manipulative Practices

At the same time, as every known case of (attempted) manipulation exemplifies, it is not a failsafe concept and the attempt to manipulate can become uncovered. This leads to two basic types of outcomes. Either the manipulator doesn’t get caught and they reach their goals, or the manipulation is unsuccessful, and the manipulator must face the consequences of their actions, if any. There is also a third case, nested in between, when the manipulation is successful at first but gets discovered later. This occurs if manipulation is not punctual but rather an ongoing

(18)

18

phenomenon that only remained successful up to a certain point. As a result, the costs and benefits for both the manipulator and the manipulee can strongly vary from situation to situation. This probabilistic perception of the phenomenon will be the foundation of the analysis of this paper and it will be laid out in greater detail throughout the remainder of this chapter.

Inconveniently, manipulation itself remains nearly impossible to spot or to empirically study, since it “only exists where the attempt to manipulate succeeds and the manipulee’s decision-making is affected by the influence” (Ibid.: 22-23). Susser et al have called “strategies that a reasonable person should expect to result in manipulation” (Ibid.: 23) manipulative practices.

Of course, the such activities are not unheard of throughout the literature but at this point it seems redundant to look any further than a recent and informed definition that is based on what was previously said.

They authors identified three characteristics of these practices, namely, that they are usually (1) “hidden”, (2) “exploit cognitive, emotional, or other decision-making vulnerabilities” and (3) “targeted” (Ibid.: 23).

Elaborating on this, they argue:

“The only necessary condition of manipulation is that the influence is hidden; targeting and exploiting vulnerabilities are the means through which a hidden influence is imposed. Indeed, targeting is best understood as an exacerbating condition. The more closely targeted a strategy is to the vulnerabilities of a particular manipulee the more effective we can expect that practice to be. While a strategy for imposing hidden influence that is not targeted still counts, therefore, as manipulative, we ought to worry more about manipulative practices the more targeted they are.” (Ibid.: 23).

Ultimately, it is impossible to say whether manipulative practices succeed at influencing their targets. Still, they make the concept of manipulation palpable by boiling it down to activities that, while equally hidden, are not entirely elusive and can, therefore, be consciously evaluated. Susser et al. also carried this concept into the digital world and they define online manipulative practices as “applications of information technology that impose hidden influences on users, by targeting and exploiting their vulnerabilities.” (Ibid.: 25).

(19)

19

A Criminological Approach to Understanding Manipulative Practices

Based on our understanding of online manipulative practices from the previous section they are a form of social influence where, like with any other, the goal is for A to make B do x. Of course, and in line with rational choice theory, A wouldn’t even think about influencing B, if A didn’t expect some increase in utility from doing so. Yet, manipulative practices don’t necessarily result in manipulation and can, if uncovered, lead to negative consequences for the manipulator. Because of these properties, manipulative practices are very similar to criminal activities and while it is not inherently illegal to manipulate, the mechanics remain essentially the same. As a result, a rational choice-based understanding of the circumstances under which crimes are committed can also shed light on the choice architecture a would-be manipulator is facing. Yet, evaluating the probability for a crime to occur, or in this case manipulation, from a largely hypothetical rational choice perspective is not beyond criticism. Bouffard and Wolf (2007) have pointed out the need to consider bounded rationality to obtain meaningful results. This criticism can largely be omitted in the present study as governance structures of companies serves the purpose to counter exactly that. A far greater limitation, however, is the inevitable outside-in approach, i.e. the inability to look ‘inside’ the companies, at least within the scope of this paper. This is yet another criticism of rational choice theory in a criminological context (Ibid.), however, it possibly makes the rational choice framework more fitting than others. For scrutinizing the broader dynamics of why and how companies might resort to manipulation and if it could be worth it to them, it appears rather perfect

Two common approaches can be found in the literature. First, Becker’s subjective expected utility model (SEU) presents a rational choice-based approach where the probability for crime to occur can be expressed in economic terms through a mathematical formula (1968). This calculated approach is exactly what Cornish and Clarke (1987) have criticized Becker for. They argued that crime would rarely take place following such sober economic deliberation and then went on to introduce their own model where they presented so called choice-structuring properties that a would-be offender is likely to take into consideration. These included nuanced factors such as

(20)

20

Both models, of course, have their own advantages, however, Becker’s approach strikes as the more meaningful one for this paper, since it delimits the potential for an event to occur through a set number of variables instead of creating new ones for every individual type of activity. Additionally, because this paper deals with the possible manipulation at the hand of companies, such a calculated approach is by no means unlikely to occur, even when considering a problematic concept like manipulation. The examples of companies deliberately accepting collateral damage of any kind are innumerable, the one worth mentioning, however, is probably the infamous Ford Pinto Case of the late 1970s that illustrates the inexorable rationality on with corporations often ground their actions. The company Ford had evaluated the cost of a human life to society against the cost of implementing an additional safety upgrade to the car and ultimately decided that the human cost was more bearable than upgrading the cars (Danley 2005). This, of course, doesn’t mean that the subjects studied here might not be limited by their own morality.

In fact, this is one aspect that strongly limits the validity of any answer to the research question. Its most comprehensive criminological explanation can be found in opportunity theory (Felson and Clarke 1999) which stipulates that actors willing to engage in criminal behaviour are more likely to do so if given greater opportunity but that the mere opportunity for crime does not automatically turn an unwilling actor into a criminal. Nevertheless, a greater opportunity for crime does lead to a greater probability that it is actually comitted (Idib.). From this it can be inferred, that the likelihood for companies to resort to manipulative practices increases proportionally as their perceived utility of doing rises in relation to the overall cost of doing so.

This proportionality can be illustrated by adapting Becker’s formula for the occurrence of crime to the manipulation context. Four variables can illustrate the wider choice architecture that any potential manipulator is facing:

U refers the expected utility, namely, what advantages manipulators could get out of the successful use of online manipulative practices

C1 symbolizes what it would take, i.e. the costs to apply online manipulative practices with a

reasonable probability of success.

C2 relates to additional costs that might be incurred refers to the social and legal costs that

result from being caught

(21)

21

It follows that if C1 > (Pc2 * (U + C2) + (1- Pc2) * U) an attempt to manipulate can be considered

profitable, at least from a purely mathematical perspective. What this depiction omits is the probability for the attempted manipulation to succeed. This was primarily done for the sake of simplicity, but it can be justified with the argument that if a manipulator wouldn’t believe that they have a reasonable chance of succeeding at their attempt they wouldn’t even consider it in the first place. This also fits with the criminological methods who also disregard the probability of failure (Bouffard and Wolf 2007).

(22)

22

Chapter 2: Case Selection, Methodology and Data

This second chapter elaborates on the relevance of China and introduces its leading tech companies as the narrower cases before laying out the methodology and data that will be used in the attempt to tackle the research

question.

Economic Reality over Political Possibility

Thinking about technology enabled manipulation, images of 20th_{century dystopias with a 21}st

century twist for extra totalitarianism are quickly evoked, however, as should be made clear right from the start, manipulation by governments is not the focus of this paper and they are arguably not the most imminent threat. This could seem counterintuitive in light of events such as the Russian interference with the 2016 U.S. presidential election (Alcott and Gentzkow 2017) or of China’s development of a social credit system which is aimed at micromanaging the lives of its nearly 1.4 billion citizens (Dai 2018).

While some success at imposing a hidden influence cannot be denied to Russia’s disinformation campaign (Alcott and Gentzkow 2017), the spreading of fake news which characterises one of the main aspects activities found in the execution of sharp power (Walker and Ludwig 2017) seems to be the furthest that governments have managed to extend manipulative influence to date. In fact, as it was alluded to above, the international arena limits the manipulative abilities of sharp power by default because manipulators have to rely on an outside-in approach. Surely, however, a population control system as copious as China’s could possibly be extended beyond its known purpose. Yet, as many of its more advanced technological features are merely planned it has become unlikely that the system will be even be fully operational in by the 2020 deadline (Creemers 2018; Dai 2018). As such it appears to be too early to address its potential manipulative abilities in detail. In fact, instances of hidden government influence through the internet that go beyond deception remain yet to be disclosed.

The British company Cambridge Analytica, however, used the personal information of over 50 million Facebook users to allegedly impact political elections in the U.S. and UK, by individually targeting voters with political ads that were based on their personality traits (Cadwalladr 2018). Whereas Cambridge Analytica didn’t spread falsehoods, it had obtained the personal information of users under problematic circumstances, which were ultimately disclosed by whistle-blower

(23)

23

Christopher Wylie the spring of 2018 (Ibid.). The resulting Cambridge Analytica Scandal became what is probably the most prominent case of attempted psychological manipulation in the private sector.

Whereas Cambridge Analytica’s approach was analysed by Susser et al. (2019) and identified to fall under the definition of online manipulative practices, the effectivity of psychographics, i.e. character-based targeting, is arguably limited outside of laboratory settings (Matz et al. 2017). The effect was likely further enfeebled because the company’s psychological profiles where mostly based on a single 2014 dataset obtained from researcher Aleksandr Kogan (University of Cambridge 2018) and because the company, like Russia, was only able to influence potential voters through the means granted by Facebook’s advertising platform.

As such, despite being individually problematic, the two cases of attempted manipulation are still a far cry from a scenario where an actor combines surveillance, exploitative targeting and unrestricted access from within. What seems like it could only exist in a future China with a fully operational social credit system beyond what is it is currently intended to, does already exist, at least to some extent and not for political purposes. Facebook itself successfully conducted a controversial ‘emotional contagion’ experiment on over 689.000 unsuspecting users, aimed at influencing their feelings by means of altering their newsfeeds through an automated tool (Kramer et al. 2014). Considering the company’s over two billion users (Facebook 2019), this can hardly be regarded as ‘large scale’, but the impact of similar activities with a purpose beyond research could arguably be larger than that of any government.

A 2017 study (Calo and Rosenbladt) on how the ride-hailing application Uber exerts influence on its ‘freelance’ drivers in order to ensure their compliance was able to identify various problematic features, some of which have later been labelled as online manipulative practices by Susser et al. (2019). It is hard to deny that the motive behind this was profit and in an age of constant A/B testing to determine the value of new ideas, it if safe to say that Uber wouldn’t have implemented manipulative features if they hadn’t previously tested their effectivity. Upon reflection, even the Cambridge Analytica scandal was ultimately made possible by the monetary interests of the company clients and while others might attempt to influence political opinions in the future, they have yet to come up with effective ways to do it.

Because government success at manipulation has been limited and since companies are already successfully employing manipulative practices for economic gain, this study omits what could one day become the bigger threat in favour of the more imminent one and focuses on online manipulation for economic gain.

(24)

24

Case Selection

American digital platforms such as Facebook, Amazon and Google are, by far, the global market leaders in their respective fields. In Mainland China, however, things look different as the country’s heavily censored internet (Freedom House 2018) has given rise to a range of domestic companies that that have moved from being copycats of western firms to becoming global technological innovators (Woetzel et al. 2017; Jia et al. 2018; Webb 2019). China’s online platform giants Baidu, Alibaba and Tencent that are commonly known as the BAT share most areas of digital business between them (Web 2019; Kantar Media 2018). Because of this they are almost unavoidable picks for any analysis focusing China’s digital economy. In line with the arguments made above, Lv and Luo have recently called for increased scrutiny of private companies instead of the government (2018).

It should also be noted that due to the structure of the analysis and because of the broad approach of this paper, the three companies will mostly be addressed together as embodiment of the Chinese Digital economy and they only treated separately when the explicit need for it arises.

Baidu

The company got started in 1998 and is often referred to as the Chinese version of Google as the its core product is a web search engine. In 2005 the company went public on the NASDAQ (Postelnicu 2005). Baidu offers a range of other digital services, most notably several video and entertainment platforms, an encyclopaedia, a map application as well as several other apps like a private cloud service and a wallet app (Baidu 2019). The company is also highly active in the area of artificial intelligence through its intelligent voice assistant DuerOs, and its open source autonomous driving platform Apollo. With Baidu clout the company also offers IaasS (Idib 2019)

Alibaba

Founded in 1999, Alibaba is commonly understood to be a hybrid between amazon an China Amazon of China, since the company is the country’s leading e-commerce player for both B2B and B2C marketplaces and online-shops (Abacus 2018). Alibaba went public on the New York Stock Exchange in 2014 (Baker et al. 2014). The company’s service Alipay is one of the two main mobile payment services used in China and while the application’s focus remains finance,

(25)

25

it offers features that address various areas of life, including health, entertainment (Woetzel et al. 2017). With Alibaba cloud, the company has built the country’s leading infrastructure as a service (IaaS) platform and within Alibaba Cloud the company pushes its Artificial intelligence initiative ET City Brain that is explicitly aimed at improving governance, among other things (Alibaba Cloud 2019) It is also noteworthy, that the company acquired the popular Video platform Youku in 2016 (George Tharakan and Krishna Kumar 2015), and owns the South China Morning Post, a large English-language newspaper based in Hong-Kong (Lhatoo 2016) as well as the movie studio Alibaba Pictures.

Tencent

Not really having an analogue in the West, Tencent was founded in 1998 and is listed on the Honkong Stock Exchange since 2004 (Abacus) and does almost everything one could imagine for a digital platform provider. The company is most commonly known for its messaging app WeChat that is often referred to an ‘app for everything’ that also addresses various areas of life beyon it main purpose of enabling communication (Woetzel et al. 2017) Aditionally, Tencent runs the very popular social platform QQ, along with various related services. WeChat Pay is the second mobile payment leader of the country, and the company makes much of its revenue in the entertainment sector, most notably with online games. Tencent also has a movie studio and various news outlets. In the field of AI the company the company’s motto is literally “Make AI Everywhere” and the company has integrated its machine learning, facial recogtion etc. into over 100 of its products (Tencent 2019).

Relevance and Justification

One of the main assumptions carried throughout the analysis is that Chinese digital platforms are not so different from their American counterparts. They continuously expand their ecosystems; develop new technologies (Woetzel et al. 2017) and as publicly traded companies they are intrinsically motivated to pursue the interests of their shareholders. As can be gathered from a quick look at their ownership structure and the fact that all three companies are traded outside of Mainland China, they are subject to substantial foreign oversight and they are affected by China’s control over its domestic capital markets only to a limited extent.

Beyond the sheer size of the three companies there are several reasons that emphasize the relevance of looking at China and at these companies specifically. As an understanding of the

(26)

26

mechanics and dangers of online manipulation was only recently shaped in the English literature (Calo 2014; Susser et al 2019; Susser 2019), there are very few case studies (Calo 2017; Susser et al. 2019) and no systematic attempts at understanding the incentive structure that the few with the ability to engage in large scale online manipulative practices are facing. Chinese examples have, so far, been eerily absent from the social science literature on the topic and the areas related to it. The possible reasons for this are manifold, ranging from simple language barriers, over the sheer inexistence of (known) cases, to the fact that the American technology companies are simply more relevant to western authors. Why especially the latter could be a misconception will be the subject of several of the next points and through them, the focus of the analysis will be further sophisticated.

A Large Population under one Jurisdiction

Aside from the obvious research gap that was just pointed out, China’s population of almost 1,4 billion alone (The World Bank 2017a) could easily justify an inquiry about online manipulation in China. Just like anyone using technology today, they are all potential victims of online manipulation that have, so far, been virtually excluded from scrutiny. Because these people all live under one jurisdiction, an understanding of the potential for online manipulation in China can grant insights that equally apply to a large portion of the world’s population at once. It should be noted that the focus will remain on Mainland China. As cooperation between businesses and the government can be very close due to the governments regulatory power over its digital market (Jiang 2012), a hidden government influence that is facilitated by the digital platforms cannot be excluded. Yet, even if the government uses if legislative or executive power to demand explicit action, the one’s in immediate control and with the greatest understanding of their technology will likely remain the companies themselves. As such, they would be the ones that influence. Additionally, because the aforementioned social credit system will grant the government its own platform to influence the population, the relevance of this aspect is unlikely to increase in the future.

Economic Relevance

China’s economy has been growing both in absolute terms as well as in relevance for the West. After having seen unprecedented economic growth since it started to open its markets (The World Bank 2017b), the has become one of the world’s leading economic powers and steadily

(27)

27

moves towards becoming more of an innovator rather than a mere producer of goods. This change is reflected by the country’s various development policies that are aimed at accelerating development and ultimately exporting Chinese technologies. Plans such as Made in China 2025

are designed to bring the country to the same level as industrialized nations in a matter of years and the project explicitly focuses on innovations in the fields like Information technology and artificial intelligence (Wübbeke et al. 2016). In fact, while still not being able to compete with American companies by sheer size and global market share, Chinese tech companies are already increasingly taking on international leadership positions in many technological areas and in the domestic market the BAT are the uncontested leaders of China’s digital platform economy (Woetzel et al. 2017). While the focus on Mainland China might arguably limit the relevance for the West, it just seems like a matter of time until Chinese digital platforms expand to the west more aggressively (Jia et al. 2018).

A Possible Power Vacuum in Global Digital Markets

The potential for the BAT to take on an leadership role in the global digital economy was possibly raised since U.S. Senator Elisabeth Warren recently voiced the idea that powerful digital conglomerates such as Facebook, Google should be broken up due to theor monopolistic position and the resulting potential for abuse of market power (Waters 2019). As a result of the still ongoing discussion, Eric Schmidt, the former executive chairman of Google and of its Parent company Alphabet Inc., indirectly hinted at a Chinese risk in a recent Interview: "It’s very important to understand that there is a global competition around technology innovation, and China is a significant player and likely to remain so.” (Pagnamenta 2019). The Chief Operating Officer of Facebook, Sheryl Sandberg, also pointed towards the consequences that might result from a possible power vacuum but was even more direct. In an Interview with CNBC she directly alluded to a chance for Chinese tech companies, that would result from a break-up of the American tech conglomerates: “While people are concerned with the size and power of tech companies, there’s also a concern in the United States with the size and power of Chinese companies, and the realization that those companies are not going to be broken up,” (Desaulniers 2019). These statements can be understood as diversion tactics, however, hard to argue against the notion that Chinese companies might end up with a significant opportunity if their western rivals are weakened.

(28)

28

Artificial Intelligence

Due to the expected social and economic disruption that especially artificial intelligence can bring with just a single breakthrough (Grace et al. 2017), Amy Webb (2019) has argued that Chinese artificial intelligence (AI) companies have just the same potential to change the world as the one’s from the United States. While she argues Due to the nature of AI and its sheer infinite scalability it is easily conceivable how the right breakthrough might result in significant economic advantages for one side over the other. How the West is already falling behind in AI technology is exemplified by a 2018 case, where the U.S. State Department justified it’s pick of French company IDEMIA’s facial recognition software as “the most accurate non-Russian or Chinese software according to the Department of Commerce - National Institute of Standards and Technology's Face Recognition Vendor Test.” (U.S. Department of State 2018: 2).

Exporting Ideas

It is no mystery that, akin to the example set by the U.S., China not only wants to export its economic goods but also engages in cultural diplomacy with the Confucius Institutes being the most prominent example (Liu 2018). While it would be a stretch to directly tie Confucius Institutes to manipulation, the concept of Sharp power was developed based on subversive and mostly online activities by China and Russia. Although very western-centric, Walker and Ludwig’s influential paper (2017) illustrates that the Chinese Government has an interest to spread its views overseas and doesn’t shy away from doing so by using hidden methods. While this study only focuses on the economic aspects of manipulation, the political dimension is implicitly carried along but as illustrated by each of the cases mentioned so far global awareness about the risks of manipulation has substantially increased. Nevertheless, all risks cannot be excluded.

While the points raised above obviously don’t mean that China and its tech companies will necessarily take economic leadership and manipulate everyone thanks to their supposedly superior technological abilities, it becomes clear that out of the countries that might deserve a closer look through the lens of online manipulation, there is much to be learned from China. And while not each of these aspects is equally relevant to both the domestic and the international sphere a better understanding of the dynamics that these companies face domestically can undeniably contribute to the literature.

(29)

29

Methodology

With the case selection completed, an understanding of online manipulative practices established, and a framework of analysis derived from rational choice theory, a sound methodological framework must be now built in order to answer the research question in the most reliable way possible despite the various limitations that result from the exploratory nature of the inquiry and the inherent difficulties of researching manipulation (Susser et a. 2019) as well as conducting research on China in general (d’Hooghe et al. 2018). While they are implicitly enclosed in the research question itself the two rival hypotheses should be explicitly phrased at least once:

H1: The use of online manipulative practices to influence Chinese individuals through their digital platforms is an opportunity for China’s tech giants, within their domestic socioeconomic context H1: The use of online manipulative practices to influence Chinese individuals through their digital platforms is a threat to China’s tech giants, within their domestic socioeconomic context

The main assumptions are that there are any diverging outcomes and that the Chinese incentive structures will somehow be different the one’s faced in the west and while this might be a given, due to the entire socioeconomic context it remains to be seen if any clear differences can be discerned among the individual companies, at least at the broad level on which the analysis is conducted.

As already announced in the introduction, the data needed to reliably simulate the outcome of the formula introduced in the previous chapter is simply not available and the research question can therefore not be addressed through a quantitative approach. Even if that data was available, the validity of such a feat would be questionable, not only due to the limitation of the rational choice framework itself but because even the most modern crime prediction technologies struggle with issues like bias and inaccuracy (Rieland 2018). Resulting from this, the analysis will explore the variables in a qualitative way with the mathematical ratio serving as a guideline for the discussion that follows their exploration. To facilitate this, it should be said that from the perspective of a willing manipulator the following properties of the variables would play in favour of the occurrence for online manipuilation. This would be the case if…

(30)

30 …expected utility (U) is higher.

…costs of applying manipulative practices (C1) are low or inexistent.

…costs from being caught are (C2) low or inexistent.

…the probability of incurring additional costs (Pc2) is lower.

Clearly, the ability to discuss the results in a meaningful way strongly depends on how explicit the results ultimately are. In the case of the large corporations that are the objects of this paper this structure becomes especially palpable since, unlike petty thieves or other criminals, most of their actions are very much open to scrutiny, as is the wider incentive structure that they face through their position as market actors. This is prominently illustrated by the circular flow model of income (Daraban 2010). To ensure the best possible reliability the analysis is spread across three chapters that embody the variables through the following sub-questions in relation to the original research question:

Why would the BAT want to manipulate? (U)

How could the BAT manipulate? (C1)

Would manipulation be worth it in China? (Pc2 & C2)

Just briefly addressing the probability of being caught, this chapter scrutinizes much closer if being caught could lead to any consequences that might have a detrimental effect on the use of online manipulative practices by the BAT. The focus is on social and legal costs and thus attempts to explore population sentiment and some aspects of Chinese legislation and law enforcement. As this chapter is essentially the risk analysis and also the only chapter that focuses exclusively on China it is also the heart of the analysis conducted throughout this paper. In terms of structure, each of the three chapters starts from a broader theoretical perspective and is subsequently narrowed down to the Chinese and ultimately the company level, when needed and if possible. As such they become their own analyses of the variables that they are designed to explore, and the results are ultimately tied together in the discussion that aims at giving a tendency that would answer the research question in favour or against an overall usefulness from the use of online manipulative practices.

(31)

31

To address the research question, this paper mostly draws on previous research and secondary data analysis. Three reports are mainly used, namely The China Social Media Landscape 2018, by the market research Firm Kantar, The Discussion Paper China’s Digital Economy from the McKinsey Global Institute and, the China Internet Report 2018, that was jointly published by the news site Abacus and the U.S. venture capital firm 500 Startups. It should be noted that Abacus is a subsidy of the South China Morning Post, which in turn is owned by Alibaba, however, due this seems unproblematic as the existence disproportional of vested interests is unlikely due to the joint nature of the publication. Additionally, the analysis draws on news reports, some information that is disclosed by the BAT directly and most importantly on three expert interviews have been conducted to give a more practical perspective to the analysis

Interview Methodologies and Dialogue Partners

Interview I

(The interviewee asked to remain anonymous in order to be able to reply openly and without fear of repercussion.)

The Interviewee is a Chinese consultant with extensive work experience in China and Germany. He additionally works as a high-tech Startup Coach and Investor and facilitates market entry for Chinese companies into Germany and vice versa.

The interview was conducted in a semi-structured way and not recorded. Therefore, a protocol has been drafted from the notes taken during the conversation and was sent to and approved by the interviewee.

Interview II

The interview was conducted with the Dutch businessman Michael Pronk. He has worked with and in China since 2011 and has been closely connected with Chinese culture since childhood. After more than 8 years of working for the family business, a Dutch consultancy firm that specializes in helping European companies break into the Chinese market, he became the acting CEO in 2018. Mr. Pronk is also the founder and CEO of another consultancy that he co-founded in 2017.

Again, the interview was semi-structured and not recorded while Equally, a protocol of the conversation was drafted and approved by Mr. Pronk.

(32)

32 Interview III

Dr. Philipp Spreer is a German consultant. He has a background in marketing, psychology and computer science and works for a consultancy firm that specialises in e-commerce and digital business. In 2018 he published the bestselling book PsyConversion that describes how psychological behaviour patterns can be used to optimize e-commerce platforms. He regularly applies these techniques in practice on the behalf of business clients.

The interview was conducted over the phone and recorded. Because the conversation was held in German, the transcript has been translated and was simplified by cutting some elaborations and digressions not to the topic of this paper. Dr. Spreer was instructed of this and has agreed to it.

Mr. Pronk and the anonymous consultant share their knowledge of the Chinese market and society and provide differentiation through both a western a Chinese perspective. Their contributions lead to a better understanding regarding the probability of being caught and the dynamics within the Chinese market and its political economy. The Chinese Consultants technical abilities and knowledge of China’s digital market yielded to additional insights about the BAT. While Dr. Speer has very limited experience with China, he was interviewed to share his hands-on experience with facilitating company influence over customers. His chands-ontributihands-on directly relates to a better understanding about the probability of success and the costs of manipulation. Throughout the analysis, the interviews will be referenced by the interview number, the number of the conversation section and the paragraph designation (ex: II-3-b). There are two ways in which this will take place. First, statements from the interviews will only be alluded to or directly quoted. In that case, usual in-text references are used. Second, the approved protocols allow for the use of entire paragraphs or individual sentences that can almost seamlessly be integrated into the analysis. If this is done, the reference will be in bold and placed at the beginning of the sentence or paragraph that that taken from the protocols and italicized for better noticeability. This essentially prevents the superfluous rephrasing of information that was already prepared in a concise manner.

The two protocols can be found in full, in the annex, but they transcript is not included in its entirety as it would be too long. Still, appropriate context will be given.

Online manipulation in China? : rational choice perspective on the incentive structure of China's tech giants Baidu, Alibaba and Tencent

Thesis for the master’s degree program in Political Science at

Universiteit van Amsterdam