With EU elections looming in 2024, reforming social media recommender systems is crucial to protect democratic discourse and the future of democracy, write Jan Nicola Beyer and Sofia Calabrese.
Jan Nicola Beyer is the research coordinator for digital democracy at Democracy Reporting International. Sofia Calabrese is digital policy manager at the European Partnership for Democracy.
In 2023, the generative AI boom has taken centre stage in the tech policy discussion and it has widely overshadowed another critical AI issue: the pressing need to reform social media’s recommender systems.
This does not come as a surprise as policymakers’ interest in the issue has progressively waned, after a moment of peak interest in 2021, following a leak of Meta’s internal documents.
As half the world gears up for elections in 2024, however, addressing the flaws in these algorithms is now becoming even more vital for safeguarding democratic discourse and democracy’s future.
Reforming recommender systems cannot therefore be left to large social media platforms alone but should require a more thorough consideration across a wide range of stakeholders, including policymakers.
In 2021, Francis Haugen, a former Meta employee, brought critical attention to the opaque operations of social media platforms, particularly their engagement-based recommender systems.
Her revelations, supported by internal documents, uncovered the damaging impact of these algorithms.
In selecting what users see in their feeds, they prioritise content that maximises user engagement, often at the cost of exacerbating mental health issues, fueling divisiveness and hate speech, promoting ‘polarisation by design‘ and creating ‘filter bubbles’ that reinforce users’ pre-existing beliefs, especially among individuals who are uninterested in politics.
These systems also inadvertently foster environments ripe for mis- and disinformation, with false information arguably generating more engagement and spreading faster and further than true information. Despite the initial widespread discussions Haugen’s insights sparked, the intensity of the conversation has now significantly waned.
Legally, there have been efforts to address these issues. The European Union’s Digital Services Act (DSA) provides a legal framework to tackle illegal content and mandates greater transparency in algorithmic design.
Yet, this regulation may not fully ensure the mitigation of all risks highlighted above. The DSA’s broad understanding of risks, encompassing everything from distorting civic dialogue to jeopardising public safety, raises concerns about its effectiveness.
Furthermore, a lot of discretion is left in the hands of platforms when it comes to risk identification and mitigation, with no indication on how to balance different risks against each other when adjusting the design of recommender systems.
Alternatives to engagement-based systems do exist, but they are not without their challenges. For instance, some platforms offer chronological feeds as an optional moderation strategy, in which users simply see what was last posted.
This approach does mitigate some biases associated with engagement-driven models, but it also has its own set of complications. For example, chronological feeds can lead to overcrowded timelines and make it easy to miss relevant content.
More critically, they don’t effectively address the issue of harmful content as users might still encounter misinformation or offensive content, which, due to its recency, remains visible and prevalent. In this context, simply posting frequently can ensure prominence in a user’s feed, which can be exploited by malicious actors.
Stanford University researchers have attempted to innovate in recommender systems with a ‘democratic attitudes’ model, incorporating social values into algorithm design. Using machine learning, they have suggested a method that identifies content promoting anti-democratic attitudes and de-emphasizes (or even deletes) it.
However, this approach is not without its challenges. There’s a risk that such models could inherit biases from their creators, leading to the suppression of critical voices under the guise of upholding democratic principles. Additionally, given that social media platforms operate in diverse cultural contexts, such approaches could impose a ‘Western lens’.
Another innovative approach is the ‘bridging-based ranking’ system, tested by Meta and X (formerly known as Twitter). This algorithm promotes content that garners positive reactions from diverse audiences, focusing on cross-demographic resonance rather than predefined content themes.
While this method aims to counter extremism, it might also dampen socially innovative ideas that initially appear extreme or objectionable. Think of the historical evolution of LGBTQ+ rights and how they were also initially perceived as extreme.
The key question is whether such a system could inadvertently stifle novel ideas at the peripheries of democratic discourse, thus impacting the natural evolution of social innovation.
Despite their inherent limitations, these alternative algorithms could represent a meaningful improvement over the current status quo. They might even be models worth aspiring to in our quest for better digital ecosystems.
However, the decision to adopt one model over another should not be solely in the hands of executives of large social media platforms given that even well-intentioned alterations in social media recommender systems can lead to unintended negative outcomes for our democracy.
The potential impacts of these algorithms are too significant for our democratic systems’ health and integrity to be decided in corporate boardrooms. It is therefore crucial for our democratic society to engage in a thorough debate about which consequences we are prepared to accept.
For these reasons, pivotal decisions to balance risks and advantages when choosing a recommender system model demand broader, more inclusive deliberation by policymakers in the immediate future, reflecting the diverse perspectives and stakes of the wider society they aim to serve.
This post was originally published on this site be sure to check out more of their content
