This article examines the feasibility of a user-led moderation model (hereafter referred to as the ‘Model’) in tackling disinformation on platforms, and how it can be augmented to become the predominant tool in combating fake news. It is proposed that the model will serve as a preferable alternative to a future that is heavily reliant on algorithmic moderation. Firstly, the area of ‘fake news’ of interest to the discussion will be identified, followed by a discussion of the tensions in regulating ‘fake news’ and the current tools available. Thereafter, the benefits and limitations of the Model will be analysed. Lastly, lessons will be drawn from parallel models: firstly, from the notice and takedown model implemented by copyright law; and secondly, the crowd-sourced editing model used by Wikipedia.
Introduction
‘Falsehood flies, and the truth comes limping after it’1
Centuries later, Swift’s words remain as relevant as ever in our modern age.2 The propagation of falsehoods is hardly a new phenomenon, but the advent of the internet has enabled unparalleled exchange of information (both real and fake) and unprecedented ability to broadcast to the world instantaneously.3 While this has ‘democratised the power of mass communication’, it also facilitates the increase and spread of ‘fake news’ which can endanger public health,4 destroy reputations and careers5 and even distort political discourse, as seen from the 2016 US presidential elections and Brexit in the UK.6 Despite the detrimental effects of ‘fake news’, attempts to regulate the phenomenon inevitably face strong resistance or criticism from stakeholders,7 given that such proposals could inadvertently result in the suppression of freedom of speech as collateral damage. Indeed, the key criticism of any model that attempts to confer the responsibility of regulating fake news upon the platforms is that already powerful platforms are given yet more power. Instead of a top-down approach to regulation, what if users were given tools to moderate the content themselves, with a view to quelling the fake news pandemic?
This article examines the feasibility of a user-led moderation model (‘the Model’) in tackling disinformation on platforms, and how it can be augmented to become the predominant tool in combating fake news. It is proposed that the model will serve as a preferable alternative to a future that is heavily reliant on algorithmic moderation. Firstly, the area of ‘fake news’ of interest to the discussion will be identified, followed by a discussion of the tensions in regulating ‘fake news’ and the current tools available. Thereafter, the benefits and limitations of the Model will be analysed. Lastly, lessons will be drawn from parallel models: firstly, the notice and takedown model implemented by copyright law; and secondly, the crowd-sourced editing model used by Wikipedia.
Defining ‘Fake News’
While it is tempting to view ‘fake news’ as a singular, monolithic concept, the reality is that the deceptively simple term encompasses a multitude of meanings. Its implications even differ from person to person, depending on where one stands on the political spectrum. For instance, while the term ‘fake news’ initially gained traction with political figures using it to discredit mainstream (and often, reputable) news providers that criticised them,8 objectively deceptive content spread online for the benefit of these same political figures has also been given the same label.9
For the purposes of this article, the author elects to borrow from Leiser’s ‘typologies of a deceptive campaign’ to pin-point which exact ‘fake news’ scenario the Model seeks to tackle. Leiser identifies four types: (1) pure disinformation: content known by both the author (who creates the content) and propagator (who shares the content) to be false; (2) misinformation through disinformation: content perceived as true by the author but deliberately shared by propagator who knows it to be false; (3) disinformation through misinformation: content known by the author to be false but perceived as true by the propagator; and (4) pure error: content that both the author and propagator perceived as true.10
The Model targets Types 1 and 3.11 While ‘disinformation’ has itself been defined as ‘false, inaccurate, or misleading information designed, presented and promoted to intentionally cause public harm or for profit’,12 Types 1 and 3 encompass the additional context where disinformation is shared further by propagators, whether maliciously or innocently.13 Such deception campaigns are likely to cause substantial societal harm, particularly because they are often complemented by tactics that advance their profiteering and political agendas.14 Indeed, the European Association for Viewer’s Interests identified ‘bogus content’ (which is ‘entirely fabricated’ and ‘spread intentionally to disinform’) as one of ten different types of misleading news, and deemed it to have ‘high impact’ on viewers.15 Hereinafter, this particular type of content will be referred to as ‘Deceptive Content’. Bona fide content that does not fall under any of the four categories shall be referred to as ‘Legitimate Content’.
I. Tensions in Regulating Deceptive Content
The regulation of Deceptive Content has been a difficult issue for governments, legislatures and internet intermediaries (such as search engines and social media platforms) alike.16 On the one hand, the right to freedom of expression, as protected under Article 10 of the ECHR,17 is crucial to a democratic society.18 On the other hand, the past few years have proven it crucial to mitigate the spread of Deceptive Content on the internet so as to prevent the public from being misled and political discourse from being tainted at its source.19 Indeed, equally important to freedom of expression is the availability of accurate information about the political and socio-economic climate on which the public can depend to form opinions and make informed choices, particularly when it comes to electoral votes.20 As such, in developing the optimal way to regulate Deceptive Content, a precise balance must be found: too little regulation allows fake news to run rampant and manipulate political discourse; too much regulation may cause speech to become governed by private corporations or censored by the state.
II. Current tools in content moderation toolbox
The issue is further complicated because developing the ideal method to detect Deceptive Content is difficult and establishing the content’s veracity is a complex task.21 This section will set out the existing technical and human-based solutions used to combat Deceptive Content and their respective shortcomings, the analysis of which will reveal the need for a more democratic, ground-up regulatory approach to resolve the issue.
A. Technical solutions
One option is code-based regulation, particularly the use of algorithms to detect and process Deceptive Content. Currently, a combination of pre-, post- and reactive moderation is being used in concert by platforms to improve the efficacy of content moderation.22 While the increasing use of algorithms to assist in content moderation may enable quicker identification and assessment as compared to human moderators, it also involves a transition towards ex ante forms of moderation.23 As Cobbe warns, it is the prospect of such fully automated ex ante moderation (which identifies and suppresses content instantaneously upon publication) that is concerning.24
Such use of algorithmic moderation by platforms (either voluntarily, or possibly prescribed by law) is likely to block and filter out more than just Deceptive Content, potentially encroaching onto Legitimate Content and resulting in the suppression of free speech.25 Indeed, there are concerns that at the current level of technology, algorithms are still inaccurate, ‘scarcely able to deal with complex linguistic phenomena connected to the pragmatics of communication, such as irony, satire, humour, and absurdity’,26 and thus are ill-equipped to deal with the complexities of ‘fake news’. Furthermore, already powerful platforms are likely to become more powerful through the implementation of such surveillance-based governmentality over private and public communications.27 Their power is enhanced in a novel way – algorithmic moderation would enable profit-motivated platforms to introduce commercial considerations further into all aspects of a user’s communications.28 This mimics Foucault’s description of power that is capillary, which extends further and deeper into the discourse of society and daily life compared to other structures of power.29
As such, with regards to the complex and often equivocal issue of Deceptive Content, this author argues that a future that is heavily reliant on algorithmic moderation is not desirable.30
B. Human-Based Solutions
A number of countries have sought to use state-imposed regulation to target ‘fake news’.31 For example, Egypt has introduced legislation enabling authorities to suspend or restrict access to sites and prosecute certain user accounts for publishing ‘fake news’.32 However, as McGonagle argues, ‘fake news’ legislation is often ‘susceptible to misuse and abuse through arbitrary interpretation and enforcement’, owing to the broad and fluid definition of ‘fake news’.33 Indeed, there remains a real possibility that states might utilise the issue of ‘fake news’ to ‘introduce censorship and to silence dissent or opposition’.34
In terms of determining the veracity of content, a norms-based tool35 currently used is fact-checking initiatives conducted by independent fact-checkers engaged by platforms.36 However, given the sheer volume of information online, such endeavours are likely to face difficulty in processing all content of interest at the speed it needs to be identified before the harm it causes manifests.37
Where the aforementioned solutions all fall short in one aspect or another, this author argues that the balance required in the regulation of ‘fake news’ might be found in the concept of community governance, where the democratic debate is returned to the people. Back in its early days, the internet seemed to represent boundless possibilities for a democratic culture, where every individual had a chance to take part in meaningful discourse on an equal footing, and ordinary people could circumvent traditional media gatekeepers and speak for themselves.38 While it might seem that the halcyon days of the internet are over, the principles of those simpler times are arguably still applicable now. Another potential option is therefore using the Model as the predominant tool to combat Deceptive Content.
III. Framing the Model
Currently, users already take part in certain aspects of content moderation, most notably through the buttons that enable users to report content that goes against the platform’s community standards. This includes content that is inappropriate, abusive, offensive or (on some platforms such as Facebook and Twitter) false.39 Users have even been labelled as the ‘unpaid workforce that polices the platform’ as they carry out the crucial function of identifying possible breaches of community standards.40 However, rather than viewing the task as onerous labour, this author argues that it should instead be seen as a valuable opportunity for the individuals to have their voices heard and participate in the public debate. As Balkin argues, it constitutes the ‘most important form of voice in online platforms’.41 The sheer number and outreach of users renders them an omnipresent policing force on the platform that is, crucially, ‘constitutive of (the) online community’.42 The resulting model of community governance43 forms a sort of policing of the people, by the people. With such measures in place, users take up an ‘editorial function’,44 thereby reducing the need for platforms to turn to ex ante algorithmic moderation.
Where questions arose previously regarding the legitimacy of platforms in regulating speech from the top-down as private commercial actors,45 legitimacy under the Model can be said to be drawn from the fact that it is led by the grassroots. This is likely to form a better, more democratic model within the cyberspace where the platform exists. While this Model still takes on the form of self-regulation by the platform, the Model leaves much (but not all) of the substantive deliberation to the governed themselves, while the platform merely provides the infrastructure, resources, and sanctions to support such a Model.
To imagine how the Model may take shape, it is helpful to examine instances of ‘fake news’ moderation by users that currently occur on two platforms, Facebook and Twitter.
A. Facebook
Based on Facebook’s reporting system at the time of writing, users can report a post which they deem to be ‘false information’.46 Facebook reviews the report and notifies the author of its decision. The author can accept the decision or provide feedback on why they disagree. Owing to a reduction in reviewers due to COVID-19, there is currently no option to request a review of the decision.47 The user will also be notified of the decision but may request another review if they disagree. In general, while the aim is to reduce the spread of false news, Facebook’s policy is not removal but to ‘significantly reduce its distribution by showing it lower in News Feed’.48 The only exception is if the content falls under three categories: (1) potential to cause imminent physical harm (e.g. COVID misinformation), (2) potential to interfere with or suppress voting and (3) doctored content.49
B. Twitter
While Twitter has only introduced a similar function in some countries recently in August 2021 for users to report posts as ‘misleading’,50 the platform has also introduced a novel community-driven pilot programme in certain countries called ‘Birdwatch’. It invites users to identify misleading tweets and compose ‘notes’ below that provide context, which other participants can also rate.51
IV. Pitfalls of the Model
While these developments present promising steps forward for the development of a full-fledged Model, there remain certain pitfalls that potentially undermine it.
Given the politically charged nature of certain types of Deceptive Content, unintended consequences might result from malicious or careless reporting of objectively true content due to a difference of opinion. This is particularly so where strong procedural safeguards are not in place to ensure the reporting process remains in good faith. By way of an analogous example, following the introduction of Twitter’s new ‘private media’ policy which allows individuals to seek the removal of photographs containing their image, extremist users began abusing the system by reporting anti-extremists accounts. In doing so, they were successful in getting those accounts suspended and in removing photographs of themselves at hate rallies.52
On the flip side of the coin, there may equally be a lack of incentive on the part of users to take part in the moderation process at all. This could stem from a sense of indifference from uninterested users, or because they are personally convinced that the post does not constitute Deceptive Content. Since algorithms on platforms are programmed to push personalised content based on past data collected on a user, their newsfeed is often precisely what they want to read, creating what is known as a filter bubble – an internet-based echo chamber.53 Indeed, as Sunstein argues, echo chambers cause people to believe in the false information to such an extent that correcting their positions might prove impossible.54 Furthermore, users may simply seek to use social media to reinforce their stances, rather than for the sake of truth-finding.55
Finally, it could be argued that there is a lack of incentive for platforms to provide expanded infrastructure to implement a Model that empowers users, rather than utilising algorithmic moderation as the predominant tool. In general, platforms are incentivised to implement some form of governance to regulate content. This could be due to the need to create an environment where users feel safe enough to remain and participate, or to enforce their own terms of use – but in any case, both are in furtherance of their own commercial interests.56 However, there might not be an obvious commercial incentive to switch to what might potentially be a resource-intensive Model, not only because implementation itself is likely to bring up costs, but because the platform might see no reason to alter or undermine its current profitable business model (which inevitably and unintentionally encourages the spread of salacious and sensationalist Deceptive Content).57
Nevertheless, where social media platforms have become increasingly like ‘modern public squares’, platforms become ‘wardens’ who have a duty to implement mechanisms to ‘ensure the integrity of users’ interactions’.58 Here, the question of how such a duty should be exercised with regards to the grey area of Deceptive Content is key. This author argues that any incentive to implement a system of community governance would be indirect. The introduction of the Model would benefit the platform’s public image and perception, and demonstrate its commitment to involving the community rather than relinquishing all power to technocratic decision-makers (or algorithms) to make unilateral decisions on speech. By giving users a voice, this may appeal to the public on either side of the spectrum – those who think that platforms are doing too much and those who think they are doing too little. Indeed, Twitter has shown continued dedication to develop its community-based moderation initiatives by expanding the ‘Birdwatch’ pilot from the initial three countries to six countries at the time of writing.59
To explore how the Model can be improved, this author examines the lessons to be learnt from parallel models under copyright law and Wikipedia.60
V. Lessons from Copyright law
While it may seem that Deceptive Content and copyright infringing material are completely different issues, the two are similar in that they both constitute forms of speech that stakeholders have an interest in removing from online platforms.
Backed by copyright law, copyright infringing content constitutes illegal content. Under the EU’s e-Commerce Directive, the host provider is obliged to remove such content or block access once it is made aware of its presence, usually by copyright holders.61 The United States’ Digital Millennium Copyright Act (DMCA) implements a similar notice and takedown model, but the difference is that the EU leaves the process up to member states, while the US has a prescribed procedure enshrined in federal law.62 Similar to how the Model will be driven by user input, copyright owners initiate takedown of infringing material through notifications (which must fulfil certain criteria to be effective) to platforms.63 To rely on the DMCA’s safe-harbour provisions, platforms would have to take down the infringing material expeditiously.64
A. Safeguards in the reporting process
The key lesson from the copyright model is that safeguards should be implemented into the process to ensure user reporting is done on a good faith basis. Otherwise, malicious reporting would render the Model counterproductive by inadvertently suppressing free speech.
Firstly, under the DMCA, an effective notification requires a statement assuring the good faith of the claimant.65 The claimant may also be liable for damages for submitting misrepresentative takedown notices knowingly.66 As Goodyear argues, this ensures that takedown notices are not submitted in bad faith, and prevents situations where individuals who do not own any rights make false complaints.67 Transposing these principles, the issue of malicious reporting can be mitigated if users have to make such reports in good faith and risk facing penalties if they make a knowing misrepresentation that the content amounts to Deceptive Content. However, if similarly heavy penalties are attached to online reporting under the Model, users may be deterred from participating in reporting of material. Therefore, a balance needs to be struck – alternative penalties may involve platform-based sanctions, which can be a valuable extra-legal tool that can keep the online community in check.68
The DMCA also provides for a counter notice, which permits a rebuttal from the author.69 Such counter notices can be transposed into the Model as vital checks and balances to ensure a right of reply. Indeed, the introduction of structured and fixed procedures (as opposed to Facebook’s removal of the request to review feature following COVID-related labour shortages) solves the issue of ‘digital prior restraint’ raised by Balkin.70 Decisions on online speech tend to be made by technocrats and algorithms, with speakers being deprived of procedure and transparency71 and often facing takedowns completed summarily, without being notified or having the opportunity to vindicate themselves.72 In the current pluralistic system of public and private governance of speech, ‘due process becomes an increasingly important value’.73 If platforms are perceived to be, or are indeed becoming, modern public squares, it is crucial that a ‘reflexive framework’ exists to ensure accountability.74
B. Repeat infringers policy
To target future infringement and to qualify for safe harbour protections under the DMCA, intermediaries are required to adopt, implement and inform users of its policy against repeat infringers.75 This includes ensuring its policy for terminating repeat infringers is implemented ‘reasonably’ and ‘in appropriate circumstances’,76 but also consistently and meaningfully.77 With regards to Deceptive Content, where platforms have to make a decision to ban certain users from posting or from the platform altogether, similar principles and standards should be considered before meting out sanctions to ensure uniform and fair decision making. A unique proportionality test tweaked for the online space (known as the ‘internet balancing formula’) can be used for resolving issues of colliding rights in the ‘modern public square’, improving transparency, and deterring possible abuse and censorship by the platform.78
VI. Lessons from Wikipedia
While doubts initially arose as to its viability due to the absence of quality control mechanisms traditionally used by encyclopaedias,79 Wikipedia has surfaced as a surprising and unlikely example of reliability.80 Heralded as ‘one of the few places on the internet dedicated to combating problematic information’,81 it is used by Google and Amazon as the key database that buttresses the knowledge element of their respective products.82 This is despite the crowdsourced and constantly revised content contributed by mostly amateur users and the sizeable number of ‘troll’ contributors that seek to misinform through the insertion of falsehoods or modification of legitimate content.83
A. Transparency
It is possible that Wikipedia achieves its reliability precisely due to this open nature, where the nature and extent of every change (and its responsible editor) is open to public scrutiny, and where transparency is key.84 Similarly, the Model would only be as effective in protecting freedom of expression if the processes which underlie it are transparent and open to public oversight.
B. Policy
An examination of Wikipedia’s policies would also assist in drawing out important principles for the Model.
Firstly, central to Wikipedia’s success is perhaps its policy to focus on verifiability, rather than truth.85 Given the impossibility of pinning down what precisely is the truth, this author argues that it may be more productive for both users and platforms to focus on any inability to provide reliable sources to back up what may be presented as facts. As McGonagle argues, if ‘truth itself proves elusive, (the) process that leads towards it is a goal in itself’.86 In the deliberation process, a post would be classified as Deceptive Content only where its author is unable to provide a reliable source for the post. This would resolve the conundrum of needing to permit someone to be the final arbiter of ‘truth’.87
Secondly, essential to Wikipedia’s ability to provide a ‘balanced coverage’ of material is its policy that more focus should be placed on presenting content from a neutral point of view – this assists in countering biased information presented as facts and the spread of misleading content.88 Under this policy, Wikipedia sets out the following principles – (1) opinions should not be portrayed as facts; (2) ‘seriously contested assertions’ should not be stated as facts; (3) facts should not be stated as opinions; (4) the ‘relative prominence’ of opposing views should be clearly indicated.89 This author argues that the danger of Deceptive Content tends to be that opinions are often masked as facts. Instead of complete takedowns or reducing distribution, platforms could incorporate such policies by introducing user-suggested flags or inputs on reported Deceptive Content to notify other users in the interim that the content may be an opinion or an unverified fact, while the deliberation is still pending.
Another aspect key to Wikipedia’s reliability is the ‘revert’ function, which permits users to return an entry to a previous version following vandalism by disinformers, without needing to rewrite the entry to correct the falsehood.90 While a revert function might not translate directly onto the Model, the key principle behind the function (i.e. to rebalance the effort asymmetry between good and bad actors91) can be applied. For example, where a function enables users to create rebuttals or contextualise posts (as with Twitter’s ‘Birdwatch’) containing information from a verified fact-checked source, the user can be given the option to replicate this addition across all posts containing the same Deceptive Content, as identified by URLs or hashes. This provides the most reliable and complete picture of the current state of the matter across the platform, while avoiding duplication of effort on the part of users.
Conclusion
In conclusion, solving the issue of ‘fake news’ is not an easy task. However, it is perhaps time to return to the original principles of the internet as a community-based environment and make the most out of the incredible ability to crowd-source as a tool for collective problem solving, as opposed to moving towards a future of purely ex ante algorithmic moderation. Only then will this ensure that democratic speech is not suppressed and left to the hands of the technocrats and profit-minded corporations, but monitored, guided and secured by the democratic public that has the most interest in ensuring its freedom.
I would like to thank Professor Andrew Murray, as well as editors of the LSE Law Review, for their valuable feedback and comments on this article.
[1] Jonathan Swift, ‘The Works of the Rev. Jonathan Swift’ The Examiner No. 15 (London, 9 November 1710).
[2] Alex Hern, ‘Scientists prove that truth is no match for fiction on Twitter’ The Guardian (London, 8 March 2018) <https://www.theguardian.com/technology/2018/mar/08/scientists-truth-fiction-twitter-bots> accessed 7 April 2018.
[3] Amy Kristin Sanders, ‘Defining Defamation: Evaluating Harm in the Age of the Internet’ (2012) 3 University of Baltimore Journal of Media Law & Ethics 112, 130.
[4] See World Health Organization, ‘Coronavirus disease (COVID-19) advice for the public: Mythbusters’ (who.int, 23 November 2020) <https://www.who.int/emergencies/diseases/novel-coronavirus-2019/advice- for-public/myth-busters> accessed 10 December 2021.
[5] Alex Mills, ‘Choice of law in defamation and the regulation of free speech on social media: nineteenth-century law meets twenty-first-century problems’ in David Mangan and Lorna E. Gillies (eds), The Legal Challenges of Social Media (Edward Elgar 2017) 270.
[6] Tarlach McGonagle, ‘‘Fake News’: False fears or real concerns?’ (2017) 35(4) Netherlands Quarterly of Human Rights 203.
[7] Mark R. Leiser, ‘Regulating Fake News’ (2017) Leiden University Scholarly Publications <https://openaccess.leidenuniv.nl/bitstream/handle/1887/72154/Regulating_Fake_News_-Leiser–> accessed 10 December 2021.
[8] McGonagle (n 6) 209.
[9] Damian Tambini, ‘Fake News: Public Policy Responses’ (2017) LSE Media Policy Project Series 3 – 4 <https://eprints.lse.ac.uk/73015/> accessed 20 March 2022.
[10] Leiser (n 7).
[11] It should be noted that Types 1 and 3 also inadvertently encompass disinformation that is deliberately produced for the purposes of satire or parody, such as news stories from The Onion, that is thereafter propagated either by actors who understand the content as it is intended to be consumed (Type 1) or by actors who misunderstand it as a genuine news article and share it as such (Type 3). The Model is not intended to cover such parodical or satirical content, which should be allowed as an exception because it neither causes the same societal harm nor has the same malicious intentions. On the contrary, it often adds to the quality of online discourse.
[12] Directorate-General for Communication Networks, Content and Technology, ‘A Multi-Dimensional Approach to Disinformation. Report of the Independent High Level Group on Fake News and Online Disinformation’ (European Commission, March 2018) 10 <https://op.europa.eu/en/publication-detail/-/publication/6ef4df8b-4cea-11e8-be1d-01aa75ed71a1> accessed 20 March 2022.
[13] Leiser (n 7).
[14] Carme Colomina, Héctor Sanchez Margalef and Richard Youngs, ‘The Impact of Disinformation on Democratic Processes and Human Rights in the World’ (European Parliament Director-General for External Policies 2021) 5.
[15] The European Association for Viewers Interests, ‘Infographic: Beyond Fake News – 10 Types Of Misleading News – Seventeen Languages’ <https://eavi.eu/beyond-fake-news-10-types-misleading-info/> accessed 13 January 2022.
[16] McGonagle (n 6) 203.
[17] Convention for the Protection of Human Rights and Fundamental Freedoms (European Convention on Human Rights, as amended) (ECHR) art 10(1); Andrew Murray, Information Technology Law: The Law and Society (4th edn, Oxford University Press 2019) 94.
[18] Jack M. Balkin, ‘Free Speech in the Algorithmic Society: Big Data, Private Governance, and New School Speech Regulation’ (2018) 51 UC Davis Law Review 1149, 1151.
[19] McGonagle (n 6) 204.
[20] ibid.
[21] Murray (n 17) 119.
[22] Cambridge Consultants, ’Use of AI in online content moderation’ (Ofcom, 18 July 2019) 30 <https://www.ofcom.org.uk/__data/assets/pdf_file/0028/157249/cambridge-consultants-ai-content-moderation.pdf> accessed 20 March 2022.
[23] Jennifer Cobbe, ‘Algorithmic Censorship by Social Platforms: Power and Resistance’ (2021) 34(3) Philosophy & Technology 739, 741.
[24] ibid.
[25] Alessio Sardo, ‘Categories, Balancing, and Fake News: The Jurisprudence of the European Court of Human Rights’ (2020) 33(2) Canadian Journal of Law & Jurisprudence 435, 457 – 458; Balkin (n 18) 1176.
[26] Sardo (n 25) 457.
[27] Cobbe (n 23) 744.
[28] ibid.
[29] ibid 748-749.
[30] While the author has argued against the use of predominant use of algorithmic moderation in this specific category of Deceptive Content, the author acknowledges that algorithmic moderation currently plays a large and critical role in the detection and removal of certain forms of unlawful content which tend to be more easily and accurately identified by algorithms. See Cobbe (n 23) 742; Hannah Bloch-Wehba, Automation in Moderation, (2020) 53(1) Cornell International Law Journal 42, 57.
[31] Flavia Durach, Alina Bârgăoanu and Cătălina Nastasiu, ‘Tackling Disinformation: EU Regulation of the Digital Space’ (2020) 20(1) Romanian Journal of European Affairs 5, 11-12.
[32] ibid; ‘Egypt Targets Social Media with New Law’ Reuters (Cairo, 17 July 2018) <https://www.reuters.com/article/us-egypt-politics-idUSKBN1K722C> accessed 15 March 2022.
[33] McGonagle (n 6) 204.
[34] Paul Bernal, ‘Fakebook: Why Facebook Makes the Fake News Problem Inevitable’ (2018) 69 Northern Ireland Legal Quarterly 513, 527; Durach, Bârgăoanu and Nastasiu (n 31) 11-12.
[35] Mark Verstraete, Derek E. Bambauer and Jane R. Bambauer, ‘Identifying and Countering Fake News’ (2021) 73 Hastings Law Journal 30 (forthcoming); Here, Verstraete and others borrow from Lessig’s four regulatory modalities (namely law, code, markets and norms) and classify fact-checking as a norms-based tool due to the fact that the efficacy of the entire mechanism rests largely on the social conventions and behaviour of a community on a particular platform. As much as platforms strive to incorporate greater fact-checking into the interface of a user’s feed, it is ultimately dependent on whether the user subscribes to the practice and utilises the resource to verify the information.
[36] Meta, ‘How Fact-Checking Works | Transparency Center’ (Facebook Transparency Center, July 21 2021) <https://transparency.fb.com/en-gb/features/how-fact-checking-works/> accessed 13 January 2022.
[37] Harrison Mantas and Susan Benkelman, ‘We Asked 19 Fact-Checkers What They Think of Their Partnership with Facebook. Here’s What They Told Us.’ (Poynter, 14 December 2018) <https://www.poynter.org/fact-checking/2018/we-asked-19-fact-checkers-what-they-think-of-their-partnership-with-facebook-heres-what-they-told-us/> accessed 13 January 2022.
[38] Balkin (n 18) 1151, 1185.
[39] ‘How Do I Report Inappropriate or Abusive Things on Facebook (Example: Nudity, Hate Speech, Threats)? | Facebook Help Center’ (Facebook) <https://www.facebook.com/help/212722115425932/?helpref=popular_articles> accessed 13 January 2022.
[40] Balkin (n 18) 1200; Catherine Buni and Soraya Chemaly, ‘The Secret Rules of the Internet’ (The Verge, April 2016) <https://www.theverge.com/2016/4/13/11387934/internet-moderator-history-youtube-facebook-reddit-censorship-free-speech> accessed 13 January 2022.
[41] Balkin (n 18) 1200.
[42] ibid.
[43] ibid, 1208.
[44] Leiser (n 7).
[45] Balkin (n 18) 1197 – 1198.
[46] Meta, ‘False News | Transparency Center |User Experiences’ (Facebook Transparency Center) <https://transparency.fb.com/en-gb/policies/community-standards/false-news/#user-experiences> accessed 13 January 2022.
[47] ‘I Don’t Think Facebook Should Have Taken down My Post. | Facebook Help Center’ (Facebook) <https://www.facebook.com/help/2090856331203011> accessed 13 January 2022.
[48] Meta, ‘False News | Transparency Center’ (Facebook Transparency Center) <https://transparency.fb.com/en-gb/policies/community-standards/false-news/> accessed 14 January 2022.
[49] ‘Our Approach to Misinformation | Transparency Center’ (Facebook Transparency Center) <https://transparency.fb.com/en-gb/features/approach-to-misinformation/> accessed 13 January 2022.
[50] ‘Twitter Tests ‘Misleading’ Post Report Button for First Time’ BBC (18 August 2021) <https://www.bbc.co.uk/news/technology-58258377> accessed 10 December 2021.
[51] Keith Coleman, ‘Introducing Birdwatch, a Community-Based Approach to Misinformation’ (blog.twitter.com, 25 January 2021 <https://blog.twitter.com/en_us/topics/product/2021/introducing-birdwatch-a-community-based-approach-to-misinformation> accessed 13 January 2022.
[52] Emma Roth, ‘Twitter Reportedly Suspended Accounts by Mistake after Extremists Abused New Private Media Policy’ (The Verge, 4 December 2021) <https://www.theverge.com/2021/12/4/22817386/twitter-suspended-accounts-extremists> accessed 13 January 2022.
[53] Cass R. Sunstein, ‘A Prison of Our Own Design: Divided Democracy in the Age of Social Media’ (Democratic Audit 2017) https://www.democraticaudit.com/2017/04/03/a-prison-of-our-own-design-divided-democracy-in-the-age-of-social-media/ accessed January 13, 2022.
[54] ibid.
[55] Bernal (n 34) 523.
[56] Balkin (n 18) 1181.
[57] Bernal (n 34) 525.
[58] Enguerrand Marique and Yseult Marique, ‘Sanctions on Digital Platforms: Balancing Proportionality in a Modern Public Square’ (2020) 36 Computer Law & Security Review 105372, 105376.
[59] Reuters, ‘Twitter Expands Feature Allowing Users to Flag Misleading Tweets’ Reuters (17 January 2022) <https://www.reuters.com/technology/twitter-expands-feature-allowing-users-flag-misleading-tweets-2022-01-17/> accessed 6 February 2022.
[60] The author acknowledges that the current models used by copyright law and Wikipedia are not perfect regimes even in their own respective spheres and have shortcomings of their own. However, a critique of those regimes is beyond the scope of this article and the author submits that such flaws do not preclude the fact that there remain valuable principles and lessons to be drawn from them.
[61] Leiser (n 7); European Parliament and Council Directive 2000/31/EC of 8 June 2000 on certain legal aspects of information society services, in particular electronic commerce, in the Internal Market (Directive on electronic commerce) [2000] OJ L178/1 (e-Commerce Directive), art 14; For accuracy, the author notes that Directive (EU) 2019/790 on Copyright in the Digital Single Market (DCSM) has effectively introduced a new regime that disapplies the safe harbour approach for online content-sharing service providers (OCSSPs) in the EU. However, the focus here for the purposes of this article is on the mechanism of copyright law’s Notice and Take down regime, which is still applicable to entities that are not considered to be OCSSPs.
[62] Martin Husovec, ‘The Promises of Algorithmic Copyright Enforcement: Takedown or Staydown: Which Is Superior: And Why’ (2018) 42 Columbia Journal of Law & the Arts 53, 57.
[63] Digital Millennium Copyright Act (DMCA), 17 USC, s 512(c)(3); (c)(1)(A).
[64] ibid s 512(c)(1).
[65] ibid s 512(c)(3)(a).
[66] ibid s 512(f).
[67] Michael Goodyear, ‘Is There No Way to the Truth? Copyright Liability as a Model for Restricting Fake News’ (2020) 34 Harvard Journal of Law and Technology 279, 303.
[68] Marique and Marique (n 58).
[69] DMCA (n 63) s 512(g)(2)(B), (g)(3).
[70] Balkin (n 18) 1177.
[71] ibid.
[72] ibid 1197.
[73] ibid 1193.
[74] Marique and Marique (n 58) 4. Here, the ‘reflexive framework’, as described by Marique and Marique, relates to a framework equipped with accountability mechanisms that enable users to state their case, be heard and obtain a proper reply in response to their arguments or stance.
[75] DMCA (n 63) s 512(i)(1)(A).
[76] Ventura Content Ltd v Motherless Inc, 885 F.3d 597, 614 (9th Cir. 2018) (United States Court of Appeals).
[77] BMG Rights Mgmt LLC v Cox Commc’ns Inc, 881 F.3d 293 (4th Cir. 2018) (United States Court of Appeals); Amanda Reid, ‘Considering Fair Use: DMCA’s Take down & Repeat Infringers Policies’ (2019) 24 Communication Law and Policy 101, 119.
[78] Marique and Marique (n 58) 7.
[79] Don Fallis, ‘Toward an Epistemology Of Wikipedia’ (2008) 59(1) Journal of the American Society for Information Science and Technology 1662, 1663.
[80] Valentin Lageard and Cédric Paternotte, ‘Trolls, Bans and Reverts: Simulating Wikipedia’ (2018) 198(1) Synthese 451, 452.
[81] Zachary J. McDowell and Matthew A. Vetter, ‘It Takes a Village to Combat a Fake News Army: Wikipedia’s Community and Policies for Information Literacy’ (2020) 6 Social Media + Society 1, 2.
[82] Noam Cohen, ‘Wikipedia Is Finally Asking Big Tech to Pay Up’ (Wired, 16 March 2021) <https://www.wired.com/story/wikipedia-finally-asking-big-tech-to-pay-up/> accessed 9 February 2022.
[83] Lageard and Paternotte (n 80) 453 – 454.
[84] ibid 456.
[85] McDowell and Vetter (n 81) 7.
[86] McGonagle (n 6) 208.
[87] Irini Katsirea, ‘”Fake News”: Reconsidering the Value of Untruthful Expression in the Face of Regulatory Uncertainty’ (2018) 10(2) Journal of Media Law 159.
[88] McDowell and Vetter (n 81) 7.
[89] ‘Wikipedia: Neutral Point of View’ (Wikipedia, 9 February 2020) <https://en.wikipedia.org/wiki/Wikipedia:Neutral_point_of_view> accessed 10 December 2021.
[90] Lageard and Paternotte (n 80) 458.
[91] ibid 455.
Genevieve Heng
LLB (University of Bristol) ’18, LLM (LSE) ’22 and IT and Intellectual Property Law Notes Editor of the LSE Law Review 2021-22