The Responsibility of Social Media in Times of Societal and Political Manipulation.

The way electorates were influenced to vote for the Brexit referendum, and in presidential elections both in Brazil and the USA, has accelerated a debate about whether and how machine learning techniques can influence citizens' decisions. The access to balanced information is endangered if digital political manipulation can influence voters. The techniques of profiling and targeting on social media platforms can be used for advertising as well as for propaganda: Through tracking of a person's online behaviour, algorithms of social media platforms can create profiles of users. These can be used for the provision of recommendations or pieces of information to specific target groups. As a result, propaganda and disinformation can influence the opinions and (election) decisions of voters much more powerfully than previously. In order to counter disinformation and societal polarization, the paper proposes a responsibility-based approach for social media platforms in diverse political contexts. Based on the implementation requirements of the "Ethics Guidelines for Trustworthy Artificial Intelligence" of the European Commission, the ethical principles will be operationalized, as far as they are directly relevant for the safeguarding of democratic societies. The resulting suggestions show how the social media platform providers can minimize risks for societies through responsible action in the fields of human rights, education and transparency of algorithmic decisions.

Introduction and research methodology

Aims and research question

During the Corona-crisis in 2020, the degree of disinformation has reached a level which could endanger the proper functioning of democratic decision-making. Crises have always been a time of rising emotions and anxiety. These seem to culminate on social media platforms, where citizens and self-proclaimed experts give unsubstantiated advice dealing with Covid-19, or try to identify assumingly guilty parties and fabricate conspiracy theories, through amalgamating facts and false interpretations. The more exciting, the even more weird ideas that are shared, including vaccine anxiety, and doubtful, or even potentially lethal health recipes.

In the United States, these developments have fuelled the long-standing debate on whether misrepresentations and recommendations still fall under the freedom of speech, or should be accompanied, e.g. with a fact check advice, or should be filtered out and deleted. The European Commission (EC) aims at combating disinformation and appeals to the social media platforms to install a transparent and consistent moderation of disinformation (EC, 2020a, 2020b).

Bell's observation (2018) that “… techniques for fabricating, editing, and reframing news in harmful ways develop faster than they can be detected and countered …” describes the current situation (p. 5). C. West Churchman asks “which end results are good in an objective sense?” (Churchman, 1970). This question is applies to the current issues of social media platforms: The paper asks whether and how social media platforms can (practically) and should (ethically), deal with risks of societal and political manipulation.

The EC's Action Plan on Disinformation (2019) defines disinformation as is verifiably false or misleading information created, presented and disseminated for economic gain, or to intentionally deceive the public. The reasons given for their action are:

the potential for far-reaching consequences such as public harm,

threats to democratic political and policy-making processes,

the risk of endangering the protection of EU citizens' health, security and their environment.

This research focuses on the threats to democratic decision-making processes. Based on the developments in 2019 and 2020, the EC (2020a) expresses concerns which relate to the main reasons for the research: “Disinformation erodes trust in institutions and in digital and traditional media and harms our democracies by hampering the ability of citizens to take informed decisions. It can polarise debates, create or deepen tensions in society and undermine electoral systems, and have a wider impact on European security. It impairs freedom of opinion and expression, a fundamental right enshrined in the Charter of Fundamental Rights of the European Union.” This paper analyses how disinformation endangers democratic decision-making and how social media platforms could contribute to tackling those challenges.

Research methodology

Reflecting on the reasons and impacts of disinformation on social media, a hermeneutic process of understanding and interpreting the ethical and societal consequences has been chosen. What can be expected is a set of ethically grounded suggestions for actions which could be discussed and implemented by the platform owners.

Hermeneutics confronts “quantitative methodologies with qualitative questions” (van Dijck, 2014: p. 206). The tools for this process are rooted in the humanities and in social science. Critical reflectivity (Gregory, 2000) is one such method that could be applied to a topic that raises societal and political questions. The results cannot resolve the existing problems of political manipulation completely, due to societal complexity (DeTombe, 2002), associated with almost ubiquitous social media. The philosopher Wilhelm Dilthey (1833–1911) established hermeneutics in the humanities as interpretative sciences in their own right. He studied the relationships between personal experience, its realization in creative expression, and the reflective understanding of this experience; and, finally, the logical development from these to the understanding of social groups and historical processes (Dilthey, 2013).

Using the concept of hermeneutics and reflective understanding, the operation and impacts of social media with regard to the current societal trends are discussed in Part 2. Examples for manipulation are given in Part 3. Concepts of digital media ethics and responsibility are presented in Part 4, and compared with the traditional media's accountability approaches. Ethics codes for AI are shown and suggestions for responsible action of social media are provided. Based on those, a reflective and strategic corporate responsibility of digital media is introduced as a new concept. Part 5 gives some reflections and limitations. Finally, the conclusion in part 6 offers a global outlook on societal responsibilities.

The role of social media platforms and their impacts on societies

To explain why societal and political manipulation is a special issue for social media, the goals, functioning and legal status of the respective corporate actors need to be clarified. Social media platforms such as Facebook, Twitter, Instagram, YouTube and TikTok facilitate an interactive one-to-few or many-to-many-communication in an international scale. “In the web 2.0 era, an infinite “crowd” of users can anonymously and with almost no cost “voice” criticism and protest, via Twitter and Facebook” (Fengler, 2012, p. 184).

The business model of social media platforms

In their official prospectus for the computerized US stock market NASDAQ, Facebook claimed their vision was “to make the world more open and connected” (Facebook, 2012). Nevertheless, their business model is commercial. When going public they disclaim that their goal is profit and shareholder value: “Advertisers can engage with users … on Facebook or subsets of our users based on information they have chosen to share with us such as their age, location, gender, or interests. We offer advertisers a unique combination of reach, relevance, social context, and engagement to enhance the value of their ads.” (Facebook, 2012). Users do not pay but give their data to Facebook who then sells them to their customers, the advertisers. Dwyer and Martin (2017) explain that data capture, data-mining and behavioural advertising are typical activities of social media businesses. The consequence for democracies is that market-driven media tend to favour advertisers and consumers over citizens (Bardoel & d'Haenens, 2004). On such platforms, traffic and user time (and their data) are important parameters.

Originally social media had positive connotations for creative engagement, political participation and cross-promotion (Dwyer & Martin, 2017). Through media users’ participation, “citizens journalism” should provide manifold perspectives and help to avoid content filtering through governments or editors of traditional media. Inspired by technical advancements, the internet followed the vision of freedom – freedom of thought, freedom of opinion, freedom of speech and expression, and freedom of information. Digital media allow individuals and smaller groups to become visible and audible. Social media were supposed to allow “the small citizen to have a loud voice”, and bring their opinions to a large audience or group of followers. This opportunity is a blessing and risk, as Evan Williams, founder of Twitter, said in 2017: “I thought once everybody could speak freely and exchange information and ideas, the world is automatically going to be a better place … I was wrong about that” (quoted by Streitfeld, 2017). According to Beckers and Harder (2016) who analysed content in Dutch and Flemish news websites on Twitter, those are regularly used as a representation of public opinion which contain, as they say, “strong—mostly negative—emotions”.

Social media platforms are subject to US (business) culture, law and convictions: The USA's “First Amendment” of 1791 protects the freedom of speech, even in cases where other people are insulted (Downing, 1999). Usually, freedom is limited through the freedom of fellow citizens and their rights, such as dignity. Print media have editors who usually think about the content, and their responsibility and reputation. Most of the social media platforms act under California law, and prefer not to be classified as publishers or media companies. This way they avoid the legal responsibilities that being publishers entail, and avoid being subject to the same legal restrictions and public service obligations as traditional media are (Frau-Meigs, 2017). “… Facebook and other social platforms have long sought safe harbour under Section 230 of the Communications Decency Act and have held firm that it is their legal right to be shielded from liability for the content their users post to their platforms.” (Bell, 2018, p. 53). This regulation of the US Code for Telecommunication dates back to the year 1934 and still protects online intermediaries that host or republish speech against a range of laws that might otherwise be used to hold them legally responsible for what others say and do (Legal Information Institute of Cornell Law School, 2018).

In the current setting, neither the social media platforms nor their users are subject to journalistic accountability standards. Nevertheless, social media platforms should be guided by what serves the good of the society. Risky developments such as political manipulation call for considerations regarding education, responsibility and eventually, as Dwyer and Martin (2017) suggest, monitoring diversity and pluralism. Lipschultz (2017) says: “Social media communication raises important ethical issues because it can be perceived as anonymous – crossing borders and cultures worldwide.” (p. 7) Zweig, Deussen and Kraft (2017) therefore call for an assessment, regulation and control of social media similar to those applying to print media and TV (p. 323). How any kind of regulation should be operationalized and implemented depends on the respective society exercising its authority and right to ensure the good use of social media, especially during election time. Possibilities and forms of an ethical governance are discussed in chapter 4 of this paper.

How AI is used for tracking and profiling

Artificial Intelligence (AI) facilitates real-time tracking and predictive analysis through transforming social action into online quantified data (Mayer-Schoenberger & Cukier, 2013). Big Data volumes allow algorithms to analyse and cluster personal profiles of users much faster and more specifically than before. Through the almost global connectivity of users on social media platforms, machine algorithms can use data for dynamical learning, classifying and predicting the audiences’ online behaviour within and across borders. This technique has been developed to increase the efficiency of marketing: it allows selection of the consumers who are most likely to purchase the respective products or services (Domingos & Richardson, 2001, and Ballings & Van den Poel, 2015). Advertisers customize their services to address targeted audiences. They try to keep users on a platform as long as possible. This allows them to learn more about their interests and to trigger more sales through adverts which are presumably, as the algorithm suggests, based on the users’ interests and previous choices. According to Usher (2017) and Wall (2017), corporate media companies appropriate such kind of content because they operate within an internet system dominated by similar commercial values and profit motives.

The same type of targeting and prediction is used for manipulative pieces of information or recommendations which reach voters who have been categorized as “potentially prone” to absorb this kind of information. Personalities can be classified and preferences can be predicted through their data such as friends’ lists, content they read or watch, and online behaviour such as researching, shopping and writing. The volume of such data has increased through cross-domain tracking (Ammicht-Quinn et al., 2018, p. 7). This is done by collecting data from the activities of their platform users on other webpages. This way, social media platforms can track which newspapers’ apps or webpages and articles users read, which political parties’ or candidates’ profiles they visit, how long they remain there. Facebook, for example, collects data from persons who do not even have a Facebook account. This activity is called “cross-domain tracking”. Under US law, this type of data collection is no mayor issue: Any action against data collection needs evidence that the data has been abused (Legal Information Institute of Cornell Law School, 2019). While the collection of data is legal in the USA, and thus encourages many data driven business and AI models, this is not the case in Europe.

How social media may be used for influencing political opinions and election decisions

Albright (2018, p. 55) has asserted: “It is not just the Cambridge Analytica scandal, Facebook, Google, and Twitter each fell prey to election-related meddling by outside actors. And most (sic) every platform is struggling with manipulation, extremist content, and uncivil behaviour.” Among the different types of fake news (Tandoc, Lim & Ling, 2018), fabrication, deception, manipulation, and propaganda are those which are frequently used for influencing political opinions. Frau-Meigs (2017) defines disinformation as a toxic, but generally discernible mixture of truth and lies and sees fake news in the category of disinformation. Their “… malicious intent is unprecedented, because information technology makes it trans-border and trans-media, and therefore viral.” (Frau-Meigs, 2017). Viral news is defined as networked news stories which often carry social significance and unexpectedness and spread online mostly through social media in a much faster and wider manner than other news stories (Al-Rawi, 2019).

Political profiling may cause the following types of issues for democracies:

Political ads are highly optimized and targeted to specific individuals. Individual users often do not know how, through whom, and why they are being influenced.

“Virality” of online content means an exponential dissemination of information. It is created by “retweeting, favouriting, and replying” and has become an indicator of online message effectiveness (Alhabash & McAlister, 2015). This effectiveness is favourable for advertisers, but also for manipulators who use triggers to gain attention. Novelty, aggressive claims, exciting news and assertations are forwarded to like-minded group members. Vosoughi, Roy and Aral (2018) analysed the diffusion of news on Twitter and classified true and false information by using six independent fact-checking organizations. They found out that falsehood diffused significantly farther, faster, deeper, and more broadly than the truth in all categories.

Because certain topics have been classified as “relevant” for the users, they get “more of the same”, meaning the users’ assumptions and prejudices are echoed and reinforced. Participation in different networks leads to a segmentation and polarization of different groups in a society (Collier, 2015) and diminishes the likelihood of political consensus or compromise. Additionally, extreme opinions can fortify false beliefs, and so can false accusations and conspiracy theories that relate to political candidates, or societal and political topics and strategies (Sunstein, 1999).

Disinformation is usually not contextualized and their origin often remains opaque. Content cannot always be verified/falsified by fact-checking, and filtering algorithms struggle with coded terminology and irony.

The anonymity of sources, paired with privacy protection, makes it impossible for police to detect and sue persons who upload illegal and illegitimate content on social media. Since alarming news is shared more frequently, journalists and politicians are experiencing defamations or threats, which diminish their preparedness for working, or and standing as a candidate, in difficult times.

A communicative “race to the bottom” decreases respect both for human dignity, and for a fact-orientated and balanced societal dialogue. Fake news and digital agitation diminish the common ground on which societies are built, and could erode trust in the democratic process of societal discourse and co-operation.

This list is not exhaustive. It contains the most important issues, and shows how impacts reinforce each other. “Hate speech, fraudulent material, deliberate propaganda, and misinformation all grew largely unchecked in an environment where platforms did not police the content they hosted with enough rigour.” (Bell, 2018, p. 5). After the Brexit vote, the European Data Protection Supervisor (2018) emphasized: “The principle of electoral transparency is not met if the voters have no freedom to seek, receive and impart information about the process and the candidates, including about the source and spending of financial support received by a candidate or a party.”

How societies can be manipulated: examples

Social media: advertising and propaganda use the same tools

People share content in disruptive phases when they are not sure what to believe or decide. They often cannot differentiate between fact-based information and rumours/opinions. Humans are not rational and often react intuitively (Akerlof & Shiller, 2010), therefore negative campaigning creates more reactions and personal, and emotional content wins over complexity. The current problem is the co-incidence of availability of digital media as “burning lenses” for an already rough democracy: conflicts are debated in shrill tones, opposing views clash with each other, while a more balanced majority remains silent. People and journalists rely on online information aggregators, which create a self-fulfilling prophecy with a popular piece of information becoming even more popular (Hong & Kim, 2018). Even mainstream TV and print media unwillingly contribute to the bias through their seemingly neutral framing, e.g. through repeating a statement by a person without mentioning that it is unsubstantiated.

Psychological nudges

Profiling can detect the person's wishes or fears and may use psychological nudges (Thaler & Sunstein, 2009) and group pressure to trigger emotions. Three examples (a)–(c) illustrate this:

During the US elections 2016, several people absorbed some “false information” that Pope Francis would endorse Trump. The idea originated from a satirical website (Schaedel, 2016), and was then spread by partisan groups to gain voters who had previously shown interest in the Republican Party and/or Trump and/or the Church. Allcott and Gentzkow (2017) analysed the effects and found that fake news had been shared through social media, but that the impact on the voters’ election behaviour is more complex: people are more likely to believe stories that favour their preferred candidate, and absorb news through manifold channels. Since social media news gets quoted in newspapers, TV and offline talks, the effect might have been larger was measurable through web data.

Influences of viral disinformation are reported from Brazil, Pakistan and Mexico (Shaban, 2019). In the “Global South”, WhatsApp plays a huge role in political campaigning (Rennó, 2018). This messenger App is frequently used in Africa, South America, and Southeast Asia because it is free of charge and seemingly “personal”. For rural areas without WiFi but a telephone network, WhatsApp offers a chance to communicate, and for (political) advertisers is a formidable tool, because voters can be reached through their telephone numbers. In lesser developed countries, telephone numbers can be purchased and sold, by telecommunication providers or by advertisers, event managers, local associations, groups, influencers and private persons. Telecoms offer their services for free (called “zero-rating”) if users exclusively use Facebook or WhatsApp (Rennó, 2018). This facilitates the targeted dissemination of “state-sponsored WhatsApp content” (Sircar, 2018) as well as propaganda in large groups. The circumstances in Africa, Brazil and India do not raise hope for a well-informed usage and electoral decision-making.

Cadwalladr (2019) explains a similar effect, showing the false information that citizens of Wales received before the Brexit referendum: It claimed that the EU would open the doors for more than 70 Million Turkish people – which frightened citizens who were classified as potential recipients for this kind of fake news. Even recipients who had not fully believed this bogus claim could have been emotionally affected and/or got a small psychological nudge. Polonski (2018) researched the role of social media during the Brexit campaign in the UK and found that the “Leave Camp” was much more active on all social media platforms than the “Remain camp”. Through their emotionally charged messages, their hashtags were shared by many users and may have attracted undecided voters.

Politics 2.0

Echo chambers have been extensively discussed and blamed for societal divide, political polarization and an increase of extremism and conspiracy theories, e.g. through the works of Sunstein (2007) and Pariser (2011). During election campaigns, many different claims and statements contradict each other, and several users welcome simplification to reduce complexity. Since adoption behaviour is common in situations of uncertainty (Fang, Hu, Li & Tsai, 2013, p. 2), social media and opinions of like-minded persons have strong influences. The distinction between legitimate self-advertising, exaggeration of problems, or of the opponent's deficits, and the reality is difficult. An expressive revolution, an increased subjectivity, groups striving for their unique identity by differentiating themselves from the mainstream and through spreading their opinions, finds formidable tools in the online world. They gain more followers, and most of them remain aligned with their groups, even though they know that some of their assumptions are wrong (Mercier & Sperber, 2017). This also applies to voting behaviour: a vote can be an expression of identity rather than be based on rational self-interest (Hamlin & Jennings, 2011).

“Social bots” are agents steered by unidentifiable developers who engage in the process of disinformation, often in order to make (advertising) money (Graber & Lindemann, 2018). Faking a human identity through common names and friendly pictures, they are programmed to send fake news, comments, pictures or retweets related to certain key words or hashtags. Most of them simply copy or repeat existing messages, that meet their objectives and some try to involve or provoke others in discussions (Stecher, 2017). They account for 9–15% of all “users” (Varol, Ferrara, Davis, Menczer & Flammini, 2017) and often are not recognized as such by human users (Graber & Lindemann, 2018). Misinformation, social programming, and identity cloning are some of their main activities (Jones & Flaxman 2015). The virality of attention-catchers is supported by the advertising mechanism: Having many readers or followers means they can offer lots of potential customers to advertisers. Consequently, extraordinary content (including defamation, obvious lies, and in worst cases, violence) enhances the opportunity to place adverts. Coordinated through a “botnet”, disinformation can effectively be multiplied and used as a powerful political tool.

Narratives and manipulated videos

Narratives play the role of an enhancer because they are understood, believed and remembered better that any factual analysis (Zak, 2014). This creates room for political influencing, as Collier (2015) explains in the cases of Russia regarding the Ukraine and several others in Europe and Africa.

People tend to believe in the evidence of pictures and videos; but pictures can be manipulated, and so can videos. In 2018, Stanford researchers presented an AI based on a neural network with a novel space-time architecture, which creates stunningly realistic videos by using only one input video (Kim et al., 2018). Those are colloquially called “deep fake”, because it is hard for ordinary people to recognize that this is not the real person speaking. The researchers hope that their demonstrations will inspire people to think more critically about the video content they consume every day, especially if there is no proof of origin (ibid.). But current practice shows that speech and video manipulations are used to defame politicians and reach millions of people. For example, Nancy Pelosi, the Democrat leader of the US House of Representatives (Harwell, 2019), and other politicians and journalists suffer from embarrassments, where voice distortions and deep fake videos are used as a digital weapon to fight against political opponents. Although algorithms could find out whether a video is a fake or not, the societal and political risks remain. Synthetically generated videos fool people, groups or institutions with lesser knowledge and/or lesser money and access to machines which could reveal the deception. If one side has more money, then they can pay for the better deep fake. In dictatorships, this tool can be used for any kind of cyber-disinformation war and a visual re-writing of history, while people might still believe in what they see with their own eyes.

Concepts of digital media ethics and responsibility

In digital media ethics, the term responsibility is used when referring to individuals and groups who are free in their decision making. Ethics is based on well-founded standards of right and wrong that prescribe what humans ought to do, usually in terms of rights, obligations, benefits to society, fairness, or specific virtues (Velasquez, Andre, Shanks, S.J. & Meyer, 2010). Ethics can be applied to analyse the aims, procedures and outcomes of direct and indirect human actions. In a famous speech “Policy as a vocation”, the sociologist Weber (2014) differentiates between the ethics of conviction, based on (teleological) values and principles, and the ethics of responsibility, which considers the impact of a (political) decision, and is therefore consequentialist. Philosophers have discussed the importance of these two contradictory views. Many agree that for political decisions the consequentialist view is more appropriate (de Villiers, 2018). Weber's ethics of responsibility applies well to social media: Here corporate responsibility means considering the impact of their corporate decisions on all societies in which they operate. Immanuel Kant's Categorical Imperative underlines the importance of the universal view with a universal law: “So act that the rule on which you act would admit of being adopted as a law by all rational beings” (Kant, 1993). This universal principle is applicable to many fields, including social media. This means that “freedom requires responsibility” turns into “freedom of speech requires responsibility”, and this call applies to “… a variety of political regimes, public uses, media cultures and traditions of political engagement in various places” (Zelizer, 2010, p. 69).

The concept of accountability is applied to traditional mass media organizations and their journalists such as newspapers, TV, radio broadcasting and the respective professionals. It can be interpreted as “… responsiveness to the views of all with a legitimate interest in what is published, whether as individuals affected or on behalf of the society. It includes a willingness to explain, defend, and justify actions (and general tendencies) of publication or omission.” McQuail (2003, p. 204). In organizations, governance comprises the rules and processes and the practice of leadership and decision making, including operationalizing norms like openness, participation, accountability, effectiveness and coherence (EU 2001a, p. 8). As businesses, social media platforms work according such kind of rules in their internal processes, while trying to defend their role of “being just a platform”, a tool for the freedom of expression (Zuckerberg, 2019).

Since they never took on the role of being publishers, they are lacking …

the regulations and media accountability which apply to traditional media,

the (virtue) ethics which Plaisance (2016, p. 465) recommends for professional media workers,

the “polis” (Plaisance, 2016, p. 466), meaning the context and environment which is necessary to lead an informed and open dialogue between different opinions and perspectives,

the media governance which would mean a more complex guidance according to agreed principles and based on a network of various influences, claims and demands (McQuail, 2003, p. 17) from various stakeholders and their divergent interests.

Most of the manipulative sources which are the subject of this paper are not issued by journalists but by anonymous persons or organizations which deliberately spread partisan content. This is why the well-known concept of media accountability with its rules and sanctions (Fengler, 2012) does not well apply to social media platforms. It also does not apply to users who post their opinions on social media. These are no professional journalists, but individuals who are neither trained nor interested in explaining or justifying their online posts. According to Barbie Zelizer (2010), the notion of media accountability is outdated, and not appropriate for an emerging environment of prosumer news, continuous global news flows, and messy news.

But who is responsible for the publication of content on a platform that claims to be only a carrier?

The user who voices an unsubstantiated opinion or belief,

Users who forward and recommend the statement,

The platform which provides the channel for this type of publishing,

The platform's paying customers, the advertisers who are using the data of all those who read, forward, like or comment the incident.

Like everyone who speaks to a larger audience, the users of social media platforms and the forwarders of messages carry a responsibility. They can report disinformation to the platform or to independent fact-check organizations. But sometimes users make up fake news themselves to win attention, as several youtubers or instagramers did during the corona crisis. Others forward thrilling news for similar reasons, or because the disinformation is not recognized as such, or the message is perceived as fun or easy explanation of a subject which otherwise is quite complex. Amidst the dense flow of information, the platforms have a hard time identifying which pieces of information are harmful, dangerous or manipulative. And whatever content it is, the broader it is spread, the better for advertising. The role of advertisers became visible in June 2020, when more than 1000 companies boycotted Facebook through pausing their advertising. They were responding to civil-rights groups’ call for a boycott over what they saw as “a lack of progress in preventing hate speech and misinformation” (Ives, 2020). Facebook announced it would expand its policies around hate speech and prohibit a wider category of hateful language in ads on the site. But even if a post violates their rules, it will remain, if it is from an important political figure, because it is deemed “newsworthy” (Isaak & Frenkel, 2020). Several companies were not satisfied with this procedure (ibid.) and seem to look for alternative ways of online marketing. Others hope that Facebook makes progress through conversations with marketers and civil rights organizations about how, together, they can be a “force for good" (Vranica & Seetharaman, 2020).

A variety of actors have caused a problem of distributed agency and morality, due to non-individual responsibilities and autonomous agents such as AI (Floridi, 2013; Taddeo & Floridi, 2018). To overcome such kinds of issues, the Future of Life's Asilomar AI Principles (2017), one of the first prominent ethics codes for AI, declared: “Designers and builders of advanced AI systems are stakeholders in the moral implications of their use, misuse, and actions, with a responsibility and opportunity to shape those implications”. Corporate responsibilities can be applied similarly to legal responsibilities of institutions: In order to avoid internal shortcomings, the leading persons need to design an internal governance and training to make sure that subordinates follow the guidelines. The owners and designers of the business models are thus responsible because they decide about the aims, strategies, methods and tools for designers and builders of advanced AI systems. Thierry Breton, European Commissioner for the Internal Market, underlined that the CEO and board of directors, together with the team of developers and managers of the media platforms and with their shareholders and customers (advertisers), cannot deny their overall responsibility (Centre on Regulation in Europe, 2020). Breton's view takes the perspective of the European market: It does not have similarly huge social networking companies and follows the ethical and (potentially future) legal norms of Europe. The European interest is “… to ensure transparency of political and issue-based advertising and to tackle fake accounts and malicious use of bots” (European Commission, 2019). A further goal is achieving progress, after representatives of online platforms, leading social networks, advertisers and advertising industry had voluntarily agreed to a first self-regulatory Code of Practice which addresses the spread of online disinformation and fake news in 2018 (European Commission 2018a; European Commission 2018b).

In their summary of “Media responsibility and accountability: New conceptualizations and practices” Bardoel and d'Haenens (2004, p. 22) conclude: “Given the structural changes in the media context such as commercialization, increasing competition, decreasing public spirit and more self-conscious and demanding citizens, a shift in media ethics and accountability systems from the level of the individual professional to media organizations and institutions seems imperative.” The concept of Corporate Responsibility is therefore transferred to social media platforms: Only those platforms have the means to detect signals generated by inauthentic content and toxic behaviour, they may use AI to combat abuse and dis/misinformation. Albright (2018, pp. 55–56) sees platforms in a position to measure news exposure and audience impact, as well as the means to trace manipulation efforts back to specific technologies, actors, and sources. Those capabilities reflect power and enable the social media platforms to be responsible, much more than publishers of newspapers, or as governments and non-governmental fact-checker organizations could do. If they wanted, social media platforms had several means to act: they could flag disputed content, impose blanket algorithmic penalties (i.e., downranking), and block the redistribution of content suspected of fraudulent, spam-like activity, e.g. through the removal of a post's re-share function (Albright, 2018, p. 56). Doing this would mean efforts and costs and probably also evoke discussions, but those discussions would be valuable to restore an unbiased electoral information.

Global responsibility standards and strategic corporate responsibility

The big social media platforms such as Facebook, YouTube, Instagram, Twitter, Pinterest and TikTok have expanded their services to many countries all over the world. As a consequence, cross-border information and disinformation flows arrive in places with different cultural, political and legal systems. Their governments and courts may have different views regarding online comments and targeted political manipulation. Different perceptions and regulations on social media may cause legal issues and demands for filtering.

The European Commission (2011) published a definition of CSR as “… a concept whereby companies integrate social and environmental concerns in their business operations and in their interactions with their stakeholders on a voluntary basis”. Strategic CSR addresses societal challenges and creates added value for business and society at the same time (Schmidpeter, 2013). CSR is used as a tool for managerial reflection and better decision making, and, at the same time, it serves the enlightened self-interest of the firms (Masoud, 2017, p. 2 and 5). Those self-interests include, among others, the stability and prosperity of the conditions around them (European Commission, 2001b).

The definitions and understandings of Corporate Social Responsibility (CSR), and/or Societal Responsibility (which is close to Corporate Citizenship) mirror the differences in the respective historical (cultural, political), socio-economic and legal contexts, and they reveal the powers of the respective actors, such as companies, employees or governments (Berthoin, Oppen & Sobczak, 2009). Similar differentiations were discussed by Gerner (2019) in his analysis of corporate sustainability and by Fifka (2011) in his comparison of Corporate Citizenship in the USA and Germany. Since the term Corporate Citizenship might be mainly associated with a specific nation, this paper uses the term Societal Responsibility in order to make clear that global networks and businesses have a cross-border responsibility for the respective societies. In a nutshell, CSR indicates the positive impacts of businesses on their stakeholders (Turker, 2008).

The United Nations (2000) Global Compact Principles, and OECD (2011) Guidelines for Multinational Enterprises were introduced in order to foster responsibility for the companies’ international business. Those guidelines are not binding, since the United Nations has no means of enforcement. Nevertheless, global players need to take responsibilities for people and the societal and political environment in the countries where they are operating. They do so because their customers and the local and home country governments expect it and to avoid losses in brand image and reputation (Šontaitė-Petkevičienė, 2015). For social media platforms, such a “shame” factor is factually not relevant: the more people are already using a platform, the more attractive it becomes for new users. Billions of users of the social media platforms remain loyal to the platforms although they disagree with some of their decisions. Users can hardly choose alternative platform providers as long as there is no data-portability which would allow them to take along all their contacts, pictures and conversations. This is why the European Commission (2019) tries to break up oligopoly structures of social media platforms and fosters data-portability.

Nevertheless, the European Market is big enough to be considered, as the GDPR and the preparedness of several platforms to sign the Code of Practice on Disinformation (European Commission, 2018c) shows. Inside and outside the companies, the societal and political consequences of disinformation are leading to a debate about the usage of one-to-many channels as political propaganda. A further subject of debate are the side-effects of insisting on the freedom of speech, whether it supports the truth or not. The US-elections in 2020, with the Corona crisis and the “Black lives matter” campaign, had further impact on the social media platform's decisions: Twitter has suggested fact-checks for questionable statements and has banned political ads (The Economist, 2020).

Reflective CSR for social media platforms?

For social media, a reflective CSR concept is a novelty. Typically, social media see their platform as the place for others to show their contents, opinions and activities.

For example, Facebook publishes no explicit CSR report, but reports about initiatives of others on their “CSRglobally” page (Facebook, 2019). Additionally, environmental friendliness is mentioned as well as their “fair pay” and diversity policy. WhatsApp and Instagram, the other two platforms of the Facebook group, also show only the initiatives of their users. A critical reflection of their own actions, and of the (side) effects of the usage of their platforms is not apparent.

“Twitter for good” claims “We reflect the power of Twitter through civic engagement, volunteerism, and charitable partnerships.” (Twitter 2019). On its website they list the following points: Internet Safety and Education, Freedom of Expression and Civil Liberties, Universal Access and Adoption, Equality, Emergency Response and Disaster Recovery, Community Engagement, Employee Engagement. Some of those are wishes, some values, others action examples.

Even though the above-mentioned social media platforms are global players and rank among the biggest and most profitable companies in the world, there are no global responsibility standards for this kind of companies. This vacuum is surprising, considering the cases of (unintended) problems, as explained in Part 3. Potential reasons for the lack of explicit responsible actions are:

Their self-definition as a “facilitator” for presumably good purposes,

The strong focus on shareholder value, and

A lack of long-term societal focus on potential risks which do not yet directly affect the company and their revenue until such risks become a public issue in the press and media.

Anticipating, recognizing and openly discussing the positive and negative societal impacts of content and processes, requires people, budgets and a long-term strategy. If shareholders do not appreciate these, because of quarterly profit goals and quarterly reporting, the companies’ directors probably have a hard time addressing long-term responsibility issues. Different opinions regarding the impacts for their own and other societies, or certain political orientations, might additionally influence the willingness of board members to decide between societal harmony and profit maximization.

But corporate foresight is a responsibility for powerful and globally active companies. If business revenues are earned through the provision of services in foreign countries, the responsibilities of social media companies need to be extended to those societies. Responsibility comprises foresight and creative imagination, to anticipate long-term societal impacts, contextual changes and stakeholders’ reactions. An analysis and reflection (Forrester, 1961) of the local circumstances is essential, especially before and during the implementation process, and then regularly to observe the developments in social media usage. This is the basis for an adaption of their systems and societal engagement.

Since global regulation and enforcement is unlikely to come soon, social media companies remain the ones who can act responsibly in different countries. Closer contact between social media directors and local stakeholders, such as local media, local scientists, users, non-governmental and governmental institutions would be helpful for mutual learning and trustful co-operation. Reflecting the complexity of the societal interactions, the only way to achieve an agreement or compromise is through interactive discourse (Habermas, 2009), an open dialogue between the platform providers, their users and the communities and regulators. It remains to be seen whether this may result in global or regionally diversified processes of improvement. Responsible design, usage and communication of social media benefits and risks could contribute positively to digital and technical literacy, as well as civilized media usage and public debate. This could be seen as a valuable contribution to tackling global issues such as injustice, inequality, poverty and societal destabilisation.

Cultural, historical and political framework conditions need to be considered, including education levels and political instabilities. Anticipation and consideration of potential consequences, especially for developing countries are essential. In order to find out how their platforms might be (mis)used, social media platforms could engage a council of independent local and international scientist to monitor and discuss the situation and (ethical) usage in different societies. Local or international non-governmental organizations caring for development and (digital) ethics could also be involved. This way, social media companies could open up and learn from a dialogue with stakeholders that have a different perspective. Scenarios for potential (mis)use of the new platform need to be discussed, in order to avoid any harm caused by rumours or agitation, such as in India or Myanmar in 2018 (Hogan & Safi, 2018). Therefore, supporting digital literacy and awareness for potential risks could be one of the responsibilities in lesser developed countries.

Ethics codes for social media platforms

Ethics for digital media are on the point of intersection of several academic disciplines such as applied ethics and philosophy, social science, especially communication and media science, journalism, sociology and psychology, business science, computer science, mathematics, statistics, data science and Artificial Intelligence (AI). The respective initiatives and norms will be explained in this paragraph with a focus on ethics codes to avoid political manipulation. Many countries have introduced regulations in order to protect users from content such as violence, terrorism, and child abuse. Social media companies need to remove harmful content within a short time. Since their AI is useful but limited in identifying potentially harmful content, social media companies have employees all over the world monitoring and removing illegal content. YouTube has 10,000 and Facebook more than 35,000 content moderators (Reality Check team 2020). Nevertheless, political disinformation seems to be very difficult to identify and to remove, as investigative journalists found out when following traces of radical right-wing groups on Facebook (Basl, Riedel, Pittelkow & Altland, 2020). Code-words and seemingly friendly terminologies, combined with irony make it difficult for AI to recognize the real meaning of a text. But if humans such as journalists help with decoding, social media platforms can improve the effectivity of their filtering AI for hate speech.

In early 2018, a group of leading AI researchers launched a paper on “The potentially malicious use of AI”. Beyond severe risks such as remote control across frontiers, they also mentioned the risk of more sophisticated social engineering (Brundage et al., 2018, p. 24): They forecasted the next step of manipulation: “Victims’ online information is used to automatically generate custom malicious websites/emails/links they would be likely to click on, sent from addresses that impersonate their real contacts, using a writing style that mimics those contacts. As AI develops further, convincing chatbots may elicit human trust by engaging people in longer dialogues, and perhaps eventually masquerade visually as another person in a video chat.” This paper was an appeal to designers, companies and governments to think more deeply on what to allow, what to regulate, where to negotiate and how to protect humans.

The EU, based on the fundamental right of data protection, was first to implement data protection regulations (EU, 1995). They were replaced by a much more detailed European General Data Protection Regulation (GDPR) which was issued in 2016 (EU, 2016) and applies since 25 may 2018. This aligned the previously different implementation modes in the European member states, and applies to every company that collects and uses personal data in the EU and abroad, if European citizens are influenced. The mostly US-based social media companies adopted their terms of use and prolonged their customer information, asking their users to agree with the existing scope of data collection and their usage for advertising; otherwise their services wouldn't work properly. European users and the GDPR sought an active management of the data, profile content and usage. But this is still an aspiration in most social media. Meanwhile, the California Consumer Privacy Act released a similar legislation on data protection which became effective on January 1, 2020 (California Legislative Information, 2018).

Several principles and guidelines for AI were elaborated by international and trans-national organizations between 2017 and 2019. The most prominent are listed below along with the date of their publication.

Further ethics codes originate from national associations, e.g. the Chinese Academy of Engineering's “Beijing AI Principles” (Beijing Academy of Artificial Intelligence, 2019), and from Information and Communication Technology (ICT) corporations themselves. A comparison has been provided and updated by Linking AI principles.org (Zeng, Lu & Huangfu, 2019a and Zeng, 2019b). The timeline shows that the recent guidelines are more holistic and try to give suggestions for practical implementation. The “Ethics Guidelines for Trustworthy AI” of The High Level Task Force (2019) of the European Commission is one of the most comprehensive and detailed guidelines for a responsible operational use.

They address four ethical principles:

Respect for human autonomy, Prevention of harm, Fairness and Explicability.

They suggest to implement seven key requirements that AI systems should meet:

Human agency and oversight, Technical robustness and safety, Privacy and data governance, Transparency, Diversity, Non-discrimination and fairness, Societal and environmental wellbeing, Accountability.

They operationalise key requirements through an assessment list, tailored to the particular system's application.

In all of the 41 pages of the guidelines, social media platforms are not directly mentioned. However, the principles and guidelines offer a good template for developing their specific Corporate Responsibility. The question whether a sectoral approach is needed, given the context-specificity of AI systems, is kept open by the EC's High-Level Task Force, as well as the question of future policymaking or regulation. In the EC's Task Force and in the working groups of all the other initiatives, independent experts met representatives of social media, both contributed their experience and views. This mixture might explain the prudent and careful approach of giving suggestions and leaving the real decisions to the companies and governments.

Societal responsibility of social media platforms

Social media platforms are commercial businesses. Since social media plays a leading role in influencing societies, their directors and shareholders need to become aware that good corporate governance is necessary to help in tackling the ethical and societal issues caused by the development and use of their platform.

Research on the effects of digitalization seems to prove that pre-existing inequalities are being increased (Cohen, 2018, and Didier et al., 2015). One of the reasons is money: high quality sources are costly, and who doesn't pay gets free, but often lesser quality or partisan information. Engagement does therefore not mean showing pictures and videos of social engagements of the users, but the companies’ decision makers’ acting and aligning their business models to the challenges given in the respective societies. Sometimes it helps to change the settings: One of the responsible actions was implementing the “forwarded message” notification in WhatsApp. It informs the user that a message is forwarded and is not an original text of a friend. After broad debates in India after mob-lynchings, WhatsApp reduced the forwarding of information to a maximum of five recipients in order to limit the virality of toxic rumours (Gopal Jayal, 2018). A further limitation of the group size in WhatsApp groups (256 max., as of Rennó, 2018) could be a further step – but this also limits the revenues earned through advertising. Responsibility and adaption to societal needs contradict the short-term profit interests of the platforms along their international expansion.

Digital media platform services address many different societies worldwide, which makes their responsibilities manifold and interwoven. Diverging ethical and legal standards as well as education and wealth levels influence the opportunities and risks regarding the provision and usage of online media. With regard to political risk minimizing, responsible societal action of social media companies could comprise the following fields, listed in order of their importance:

Human agency: “AI systems should support human autonomy and decision-making … This requires that AI systems should both act as enablers to a democratic, flourishing and equitable society by supporting the user's agency and foster fundamental rights, and allow for human oversight” (European Commission 2019).

Privacy and data governance/Access to data: Personal data, opinions and information paths need to be protected in order to safeguard the freedom and confidentiality of the citizens’ political opinions.

Societal and environmental well-being / Society and Democracy: “The use of AI systems should be given careful consideration particularly in situations relating to the democratic process, including not only political decision-making but also electoral contexts” (European Commission 2019). For social media platforms this also means independence from political advertising to avoid their platforms becoming multipliers of partisan propaganda wherever they operate.

Stakeholder participation and social dialogue: Companies should engage in an open discussion with the involvement of social partners and stakeholders, including the general public (European Commission 2019) in all the countries where they are active.

Actions against harmful content: Companies should develop strategies to avoid becoming multipliers of violence or hate and actively engage in the protection of persons and groups who need support. Clear criteria, AI (Arnold & Scheutz, 2018), and teams of independent experts can help to identify potentially harmful content and ban it from platforms. By anomaly detection, AI can be used to detect the start of viral processes. If those turn out to be toxic, such as the video of the massacre in Christ Church in March 2019, social media platforms could “freeze” suspicious activities until further investigation is made and a release has been given by a council of independent advisors. This type of action can be institutionalized, similar to stock markets which close dealing if there are strange movements in the market.

Tech and media literacy (Entsminger, Esposito, Tse & Goh, 2018) can support citizens’ short- and long-term ability to spot untrustworthy news sources and information (Jacob, Sapienza & Lister, 2019). For example, courses on specific fields of AI could be provided in schools, public learning centres, senior citizens’ homes. This helps to integrate different parts of the population and gain experience and a better understanding of the benefits and risks of online media and AI.

Educational support for vulnerable populations with low or inadequate education levels or resources, e.g. through providing teaching programs for digital literacy (in general, not merely as advertising for the usage of their own app). Users need to be shown how to identify illegal and illegitimate content, and to avoid creating, using or forwarding such content. To learn about the quality of sources could be one of the main teaching aims in digital literacy.

Education and awareness to foster an ethical mind-set: “A prerequisite for educating the public is to ensure the proper skills and training of ethicists in this space” (European Commission, 2019).

Taxes as a fair contribution to the local societies

Taxes are not mentioned in the principles and guidelines, probably because corporate representatives have a strong saying in the respective task forces. But taxes are needed to provide high quality ICT infrastructure throughout a country, and are necessary to run states and to implement and execute good governance. Taxes are also needed to increase the citizens’ education level which is crucial to be able to differentiate between high-quality sources and disinformation. Knuutinen (2014) points out that strongly increasing profits of companies for many years, combined with decreasing taxes for states, can undermine the foundations of democracies. Taxes are part of each company's societal responsibility, are a matter of fairness and expected by stakeholders, for social media companies as well as for any other type of business. If companies choose low tax havens or loopholes to evade payment of adequate taxes in the countries where they earn revenues, they are failing in their ethical duties and should not be proud of any donations which they might later give.

Since profiling is supported by AI, the process of classifying human online behaviour needs to be made transparent. Algorithm audits (Zweig et al., 2017) and some external accreditation for critical algorithms, could be helpful to assess the status quo and progress of responsibility of the respective companies. Possibly start with the ones that have a broad societal impact and engage independent certified accreditation bodies. Algorithms could be analysed through laboratory testing and reverse engineering (Diakopoulos, 2015), which means reconstructing the algorithm to identify the functional principles. Additionally, more cooperation with independent researchers would be helpful in order to research the dissemination of misinformation (Lazer et al., 2018; Torabi Asr & Taboada, 2019).

Regarding algorithm transparency, three points need to be considered:

The details of the algorithms are usually business secrets, because of curious competitors or hackers. But a certified auditor or accreditation agency could be entitled to check, similarly to how they check the company's financial data or processes. Even though algorithms-in-action are usually inscrutable, a blueprint-algorithm can be comprehensive (Kavanagh, McGarraghy & Séamas, 2015).

The generalized depictions of the technology, and the claims about how they work or are used, may differ a lot from how they are embedded in specific social or organizational contexts (Kavanagh et al., 2015). Responsibility also requires a continuous monitoring and tracing through empirical studies to ensure the reliability. This should happen with the companies’ own algorithms, and with the services connected to affiliated customers. In the end, a regular and credibly performed external evaluation could avoid legal initiatives in this field.

Reflection and limitations

Media platforms hold unique power in societies. The temptation is strong for political and economic elites in developing or immature democracies to capture and pervert the media to promote their own interests (Rothman, 2015). Any type of regulation has two sides: it might prevent harm, but decreases the degree of freedom. The debate on “freedom versus regulation” has a long history. and Immanuel Kant tried to establish an ethical concept in order that decision-making be in the hands of the free, but responsible individual. Freedom and responsibility are thus seen as a pair, and corporate responsibility is a complementary concept to entrepreneurial freedom.

If social media companies choose to act responsibly, they have the scope for their own thematic focusses and claims. They can decide whether to do a bottom-up (recommended) or top-down approach, whether they integrate employees and other stakeholders (from different nations) or whether they “think for them”. An interactive approach would allow them to learn from the stakeholders’ expectations, to be more innovative, through a diversity of experiences and thoughts. All actors in this debate need situation-centred critical (self-) reflectivity (Mead, 1910), empathy, care and compassion for others in order to responsibly balance divergent interests. As a method, multi-criteria decision making (MCDM) could facilitate decision making related to complex problems (Brugha, 2004).

Responsibly governed, social media platforms can be a tool for better informed decisions and for leveraging their societies upwards, socially and economically. As Bell (2018, p. 5) says: “It's becoming obvious that platforms cannot scale advertising revenue in tandem with the robust intervention needed to ameliorate the type of material they publish.” Therefore, the likelihood that social media platforms take this responsibility is high – especially if more competition and data-portability would create more choices for users who then might chose the more responsible offer. If companies find their own ways of dealing constructively with the ethical challenges of their operations, they could find multiple and adaptive ways of improvement, rather than being bound by maybe rigid or impractical regulations.

Since social media platforms have different principles (Zeng et al., 2019a), challenges and user and stakeholder groups, the ways to act responsibly will be different and the companies could be in competition with each other in order to achieve the highest degree of responsible action, credibility and reputation. Data portability is a necessary prerequisite to revitalize competition, otherwise it is almost impossible to change a social media provider without losing contact data, conversations, photos and videos in a long timeline of usage. This creates a so called “lock-in” situation for users (Zweig et al., 2017). The network-effect, reinforced by the acquisitions of WhatsApp and Instagram, strengthen the current “the winner takes it all” position of Facebook. A change in the current oligopolistic structure requires competition, anti-monopoly laws and their ideas of entangling conglomerates and/or creating data-portability between different providers (Entsminger et al., 2018). The EC (2020b) is following this path along their new European Data Strategy but it remains open when and how they can achieve their goal. The EU is not the only region which requires changes: The different political systems and interests are leading to a trend to install specific regulation which ends up in a split internet (“splinternet”). Whether and how far such new “borderlines” between regional cyber systems systematically follow certain societal/political values and norms, and what a splinternet practically means for global communication respectively social media, is a subject for future research.

Ethical norms could be implemented into technology or innovation assessment practices which see ethics and multi-stakeholder involvement as a “design” factor of technology. Following the EU's goals, programmes and sample cases, von Schomberg (2013) defines “Responsible Research and Innovation as a transparent, interactive process by which societal actors and innovators become mutually responsive to each other with a view to the (ethical) acceptability, sustainability and societal desirability of the innovation process and its marketable products in order to allow a proper embedding of scientific and technological advances in our society”. How fast this will happen, whether companies or regulators will take the first steps, and whether different approaches will be chosen for different cultures, remains to be seen. More efforts in developing and discussing realistic scenarios for a desirable future of our societies would be helpful (Cath, Wachter, Mittelstadt, Taddeo & Floridi, 2018).

Conclusion

Analysing, clustering, tracking, profiling, prediction and recommendations are powerful tools for supporting decision making. Profiling and targeted information created by machine learning algorithms have been growing exponentially, and so has their significance for citizen's political decision making. In digital media, AI can potentially be (mis-)used as an instrument of power through information filtering and targeted (mis-)information and/or manipulation. This can happen in different formats such as text, audio and video, and be multiplied by social bots and viral effects. Since the US-American social media platforms tend to understand themselves as tools for others, they show a few CR activities, but do not assume comprehensive societal responsibility. This would be necessary to contribute positively to a more balanced communication amongst competing parties and interest groups in societies, to safeguard trust in their democratic and legal system and to help stabilize fragile democracies.

Suggestions for responsible actions have been derived from the European concept of societal cohesion and the Guidelines for Trustworthy AI. They extend CSR to a responsibility for all societies in which social media services are used and create revenues for the companies. The complexity and diversity of local framework conditions and societies in terms of wealth, education, values, and political systems turned out to be a challenge for social media platform providers. They can master those challenges through transparency and opening up for dialogue and cooperation with local experts and stakeholder groups. Ensuring benefits and preventing harm will depend on defining the common interests regarding digital media platforms and on forecasting/anticipating potential benefits and risks. The transparency of processes and interdisciplinary thinking are as necessary as situation centred consciousness. This would help guiding and bundling technical, ethical, and societal research efforts and help hedging the risks and harvest the benefits of social media and their AI for generations to come.

Table 1

Principles and guidelines for AI elaborated by international organizations.

Published in	Organization	Name of the Principles and/or Guidelines
08 Jan. 2017	Future of Life Institute (FLI)	Asilomar AI Principles
Feb. 2018 (V1) and 2019 (V2)	The IEEE Global Initiative on Ethics of Autonomous and Intelligent SystemsInstitute of Electrical and Electronics Engineers (IEEE)	Global Initiative for Ethical Considerations in Artificial Intelligence and Autonomous Systems
14 Nov. 2019	European Commission (EC)	Ethics Guidelines for Trustworthy AI
03 May 2019	Organization for Economic Cooperation and Development (OECD)	Principles on AI
22 May 2019	World Economic Forum (WEF)	AI Governance: A Holistic Approach to Implement Ethics into AI