Don’t über-blame the algorithms — lessons from Europe

Guest Blogger

From the Workshop on  “News and Information Disorder in the 2020 US Presidential Election.” 

Amélie P. Heldt 

One can hardly talk about disinformation and Europe without mentioning Brexit and the Cambridge Analytica scandal. However, the last four years have shown that blaming only technology for the information disorder is illusory. After Brexit, Europe feared disinformation campaigns would affect the French and the German elections in 2017. Until today there’s no evidence of such, either from foreign actors (e.g., so-called troll armies) or as a result of content curation systems on platforms. For instance, the 2017 German elections saw hardly any manipulation from abroad, and the influence of personalization by algorithms in search engine results was negligible, as Cornelius Puschmann wrote in “Beyond the Bubble: Assessing the Diversity of Political Search Results.” Similarly, false news sites had no significant influence on the 2017 French elections, researchers found. On the contrary, research shows that most of the issues are man-made and partly based on a failure of traditional media. Nevertheless, rumors concerning then-presidential candidate Emmanuel Macron perturbed the election campaign period and sparked off a regulatory reaction. 

Indeed, France was the first European country to pass a law against “information manipulation,” targeting the dissemination of false information for the purpose of election rigging. Under this law, “any allegation or inaccurate or misleading attribution of a fact which could affect the truthfulness of the forthcoming elections, which is intentionally, artificially or automatically disseminated on a large scale via an online public communication service” is prohibited. The law thus targets false statements of facts that could reduce the credibility of the election results. The application of the law is limited in time to the three months preceding an election, including the day of the election. Against statements that violate the law, urgent preliminary ruling procedures can be initiated in court and receive a preliminary ruling within 48 hours. The court may decide on all proportionate and necessary measures to prevent the dissemination of false information, including blocking, deleting, and/or not disseminating the content. The law also includes higher transparency requirements for political advertising on social media platforms. 

While this law might prevent information manipulation during future French elections, it remains unclear how to assess this kind of regulatory initiative with regard to its efficiency and its proportionality. Taking regulatory action without disproportionately restricting fundamental rights such as freedom of opinion or freedom of the press is extremely complicated, precisely because the phenomenon is so diffuse and complex. Digital services can be used to try to influence people, and automation can simplify and enhance this process. But the actual effect of automation on election results, whether real or fake, cannot be measured. In fact, a recent report by the UK Information Commissioner’s Office invalidates many accusations made against Cambridge Analytica two years ago. Cambridge Analytica’s prediction model wasn’t as efficient as presumed and, as researchers pointed out before, political microtargeting existed before social media platforms. 

At the supranational level, the EU so far chose self-regulatory forms of governance to address the issue. In September 2018, the EU and leading tech companies agreed upon the Code of Practice on Disinformation. The CPD, as it’s known, defines disinformation as “verifiably false or misleading information” which, cumulatively, (a) “is created, presented and disseminated for economic gain or to intentionally deceive the public”; and (b) “may cause public harm,” intended as “threats to democratic political and policymaking processes as well as public goods such as the protection of EU citizens’ health, the environment or security.” Tech companies can choose if and how they comply with the CPD — hence, to which extent they will go beyond their own set of rules. Ultimately, the platforms still govern the evaluation and interpretation of what type of false information might be harmful to democracy. As one might expect, the CPD was no game changer in this area. 

Currently, the EU Commission is preparing several drafts of regulatory projects which could potentially include measures against disinformation on social media platforms. The Digital Services Act, a draft of which will be revealed on Dec. 15, will most probably include new rules regarding content curation and transparency obligations. 

The EU Democracy Action Plan, published on Dec. 3, 2020, ought “to ensure that citizens are able to participate in the democratic system through informed decision-making free from unlawful interference and manipulation.” With regard to the role of online platforms, the DAP includes six objectives:

“1. monitoring the impact of disinformation and the effectiveness of platforms’ policies, 2. supporting adequate visibility of reliable information of public interest and maintaining a plurality of views, 3. reducing the monetization of disinformation linked to sponsored content, 4. stepping up fact-checking, 5. developing appropriate measures to limit the artificial amplification of disinformation campaigns, and 6. ensuring an effective data disclosure for research on disinformation.”

It remains to be seen what concrete measures will be taken to achieve these objectives, but it is already apparent that it could have a significant impact, including beyond EU borders. Last but not least, the EU Data Governance Act will likely introduce a fiduciary duty of the data intermediary with regard to the data subject. 

In sum, the EU now focuses on two types of countermeasures: a significant shift to procedural measures instead of targeting specific types of speech and, in parallel, a stronger protection of the digital public sphere by strengthening both the users as individuals and traditional media outlets as trusted conveyors of news. 

The latter goal is also pursued by Germany with its new State Treaty on Media, under which search engines and social networks have to mark social bots and must clarify the basic principles of their selection and sorting, in a user-friendly and understandable way. Given the complexity of the algorithmic selection procedures used, it remains unclear whether this duty will be practicable. According to section 94 of the State Treaty on Media, “to ensure the diversity of opinion, media intermediaries must not discriminate against journalistically and editorially designed contents on whose perceptibility they have a particularly high influence.” This provision raises many questions as to its practicability and its plausibility, mainly because intermediaries convey a wide variety of very diverse content, each of which is of varying relevance to the formation of opinion. 

Regulating automated speech raises questions: To what extent is the law required to observe proxy freedoms for automated agents such as social bots? Over the past months, we’ve seen a rise of misinformation and conspiracy theories amplified by recommender systems, sometimes leading to real-life violence. But the other part of the story is that most conspiracy theorists succumb to human pied pipers, even if distributed in YouTube or Telegram channels. In theory, algorithms in recommender systems could contribute to a more diverse and balanced online environment and therefore be part of the solution, provided they meet substantial transparency requirements. 

All in all, it’s an ongoing iterative process, and regulators on both sides of the Atlantic are struggling to find adequate responses to the issues raised. Eventually, they need to take action because the stakes are high, but there’s no single-sized solution because the roots are many-sided (as in the U.S.). It is therefore not sufficient to target only “the algorithms,” instead of taking a broader view, including institutional and political dynamics. Moreover, algorithmic content curation poses many challenges because of the risk of violating freedom of expression. Content-based approaches are most likely unconstitutional because it seems impossible to identify manipulative or dangerous political speech only. In the end, higher exposure to pluralist and diverse media sources at the recipients’ level might reduce the risk of misinformation and disinformation, and be the most moderate approach with regard to autonomy and democracy considerations.

Amélie P. Heldt is a junior legal researcher and doctoral candidate at the Leibniz-Institute for Media Research | Hans-Bredow-Institut.