Simulations for the future of democracy
When you take the underground or tram, you can be sure of one thing — you will see most people glued to their phones. Besides watching a seemingly endless stream of cat videos and viral dance clips, we also scroll through the news, follow political debates and share ironic memes or hateful rants on social media. We like. There is no doubt that mobile devices are not merely for entertainment, but serve as gateways to an always-on digital information environment. "By digital information environment, we mean the full range of information sources available to users online. This concept includes online social networks, but also traditional mass media which we now primarily consume digitally," explains Annie Waldherr.
As part of the international research consortium WHAT-IF, the Professor of Computational Communication Science at the Department of Communication at the University of Vienna studies how the dynamics of our digital information environments affect democracy. It is clear that online media as well as the possibilities and limits of social media platforms shape the ways in which we stay informed, exchange views and organise politically.
Digital ambivalence
These possibilities prove to be both a blessing and a curse for liberal democracy. On the one hand, social media enable users to move beyond the passive role of the consumer, allowing them to engage more actively in public discourses and raise the visibility of issues that are overlooked by traditional media or suppressed in authoritarian regimes. On the other hand, social media gives rise to new problems for the democratic discourse, including disinformation or viral hate speeches that algorithms keep pushing onto our screens over and over again.
The research consortium examines three aspects of the problem in this context. "First, we assess the quality of information available to users, then we check whether the information sources use offensive language. Finally, we explore how digital mechanisms affect polarisation, in particular whether people are still willing to talk to people from other social groups."
Policymakers respond to the downsides of our online worlds with regulation, while platforms ban accounts or enhance moderation – unless they prioritise profit over these measures. But how can we evaluate the impact of such interventions on digital information environments and thus on our democracy? It is clear that any change to the current system is essentially an experiment, with users serving as unwitting test subjects.
Simulated networks
The answer may lie in the WHAT-IF machine, a computational model-based ‘digital twin’ of our online information environment, allowing researchers to safely test possible interventions such as regulations against disinformation or changes to algorithms. Waldherr, whose team at the Computational Communication Science Lab of the University of Vienna is responsible for the simulations, explains, "We use two different methods. One of them relies on an agent-based model, in which we assign opinions and intentions to agents and place them in a predefined environment. These agents can exchange information and connect with each other."
Postdoc Aytalina Kulichkina who is currently developing the core modeling framework for the Horizon Europe WHAT-IF project adds, "We also model strategic agents representing highly active content creators and amplifiers who seek to influence ordinary users. The computational model allows us to ethically study how their intentional behavior shifts political opinions over time and to evaluate interventions that reduce harmful effects, including risks of radicalization and extremism."
"WHAT-IF Machine"
The "WHAT-IF Machine" is a simulation platform designed to act as a digital twin of the political information landscape. It combines state-of-the-art technologies such as agent-based modelling (ABM) and large language models (LLMs) to simulate the interaction of individuals and groups with (social) media.
Surprisingly, even this simple model can capture complex realities, although each agent consists only of mathematical formulas that represent beliefs or interests and determine their reactions to other agents. "We can also incorporate emotions," says the expert on communication studies. "A similar model showed in a previous project that anger is a major driver of interactions."
Agent-based models also take into account that users are not all the same. "There are typical social media platforms such as Facebook or Twitter where well-connected users can gain new followers more easily. As a result, some accounts develop large reach, while many others have only a small number of contacts," explains Waldherr. "In addition to such network structures, we also simulate conventional forums, where everyone can interact with each other."
AI support
In cooperation with David Garcia and the Social Data Science Lab at the University of Konstanz, the Vienna research team complements its agent-based models with artificial intelligence systems that have recently entered our digital information environments in the form of AI-based chatbots. In this approach, individual agents are connected to AI systems. “Even today, AI-generated posts on platforms such as Reddit are barely distinguishable from those written by humans.” Although this is concerning, it also means that we can use AI to better simulate human behaviour.”
The validity of the models is based on empirical input. While the computational model is still under development, the project partners of the Vienna team are conducting representative surveys in several European countries. The users' views on topics such as migration and the climate crisis will then be fed into the model to help ensure more realistic behaviour. "In the next step, we will also check whether the simulation produces patterns similar to those observed in reality," explains the expert on communication studies.
What if platforms no longer rewarded hate and hate speech with attention? What if calm, solution-oriented discourses gained more attention on social media? What if conspiracy theories were consistently flagged? If the empirical tests are successful, we could use the models to explore these questions. This is ultimately the goal of the WHAT-IF consortium.
Digital self-empowerment
"Our main focus is on moderation interventions, including algorithmic ones, such as restricting or suspending accounts that are spreading large volumes of hate speech or disinformation," says Waldherr. "We want to understand where regulation is beneficial and where it may be harmful to democracy. We also explore how different platform logics shape our digital information environments." The latter could be relevant for European social media projects, especially in the context of digital sovereignty.
And it is also useful for the users. "Our simulations show that it often takes fewer people than expected to bring an issue to wider attention. This way, we can challenge the perception that individual concerns go unheard and we can enhance the users’ self-efficacy." In general, Waldherr adds, the models are also ideal tools for explaining how digital platforms work.
The models developed by Waldherr and her team help us understand how the digital world influences our real-world democracies. At a time of multiple crises, rapid technological advancements, remilitarisation and demographic change, democracy is walking a fine line around the world. In this situation, research can help us choose the right path forward. Waldherr: "Which levers will help us bring people back together? How can we stop users from retreating into silence?" These are the questions the academic will continue to investigate.