Identifying, deciphering and combating misinformation stories

This interview for the European Journalism Observatory is available in German, French, and Portuguese.

What are the objectives of PROMPT?

PROMPT is a ‘European Narrative Observatory’ funded by the European Commission which seeks to identify and analyze the emergence of disinformation narratives, using AI, on three topics considered as a priority for the European Commission: the war in Ukraine; LGBTQIA+ rights and freedoms; and the 2024 European elections. 

For the next 18 months, we’ll work to decipher the origins of these narratives and their emergence both offline and online in 6 countries (France, Baltic Region, Italy and Romania) and 8 languages (add English and Russian!), their role in shaping people’s minds and the emotional values that support “successful” disinformation narratives. We’ll develop a methodology to better understand how narratives emerge, how they spread across platforms. We’ll focus on key informational nodes in influential media outlets, social platforms and Wikipedia.

Ultimately, we’ll offer journalists and activists new tools to recognise, anticipate and tackle disinformation narratives.

PROMPT is a pilot-project, meaning that we mobilize cutting-edge expertise and AI technology to attempt tackling disinformation. We’re lucky to work across multiple disciplines (mathematics, linguistics, sociology, media studies,etc.) with 11 leading actors in 8 EU countries, in particular with media experts and journalists, such as the Erich-Brost Institute (Germany), Les Surligneurs (France), Re:Baltica (Baltic states) or Asociatia Digital Bridge (Romania).

What inspired the development of this project?

We have a clear purpose: support healthy democratic debate. We’ve worked on monitoring opinion online for the past 10 years; and on communication/rhetorics for the past 20 years. We see how disinformation has evolved - and how nuanced it is (it’s not just about ‘fake’ or ‘true’ news). We’ve worked with journalists and activists, and we see that there is untapped potential with AI tools to support their work.

The project's two main drivers are technological and societal. Technological, because recent advances in AI, which we have been integrating into our work on analyzing large volumes of social data for several years now, have enabled us to make a considerable leap in performance, by allowing advanced processing of online expressions with LLMs and their automated qualitative categorization, "augmenting" the capabilities of analysts.

Societal, because we believe that only a whole-of-society approach will enable us to win battles in the war against disinformation. Professionals engaged in this battle sometimes feel alone or disarmed in the face of the upsurge in destabilization operations. Our humble ambition is to provide them with additional weapons.

What sets this project apart from other (previous) projects on AI, media reporting and training?

PROMPT is the second European Narrative Observatory. We stand on the shoulders of NODES (Narratives Observatory Combatting Disinformation in Europe Systemically), a project which primarily focussed on how online disinformation communities are structured and interact on other topics of interest - migration, climate change and Covid-19. 

Our research builds on this work with additional steps, using for example Large Language Models to deepen the monitoring of the emergence and spread of narratives. 

But PROMPT is not just about building the best “silver-bullet” model - we’re skeptical about that kind of talk. It’s about making the most useful one. 

There are many academic efforts to develop cutting-edge methodologies to analyze narratives. What is also very important, and often missing, is that additional effort to make research operational for those who fight disinformation everyday. Good examples in this field include projects such as vera.ai, which help design and test AI-powered tools to detect disinformation. With fact-checkers and Wikimedia France - Wikipedia being a Very Large Online Platform faced with systemic risks according to the EU - onboard the PROMPT team, we’re confident that we can have a great impact across our audience.  

We also complement work on fact-checking and media literacy, for example of the European Digital Media Observatory (EDMO) or of members of the European Fact-Checkers Standards Network. We are also thrilled to be building on the expertise of the Erich-Brost Institute, which can amplify our reach to a comprehensive EU-wide network of journalism institutes, spanning all member states.

How does the European Narrative Observatory contribute to fighting disinformation?

To answer this question, you first need to understand why narratives matter. Disinformation narratives carry important risks and threats for EU countries. The dissemination of false and misleading information on the war in Ukraine has a strong potential to undermine public support for national and joint EU actions supporting Ukraine. False narratives targeting and attacking the LGBTQIA+ community can cause significant harm in the integration and equality of these groups, as well as influence the exercise of human rights. Disinformation narratives around elections can mislead voters and undermine the foundations of our democracies.

Disinformation actors, including those powered by rogue states attempting to destabilize our democracies, work tirelessly to sow discontent and polarize public debate. We have to address this, with the best evidence and techniques. By naming, uncovering, shaming, counter-attacking, and making disinformation efforts more costly, work on narratives, like PROMPT and others, bring their own contribution to efforts to fight back disinformation online and offline.

There are still many aspects of disinformation to study: how for instance it circulates across social platforms with “universal” rhetorical patterns or with local language and culture idiosyncrasies. This is one of the key questions the PROMPT project will address.

What is the role of Large Language Models in discovering fake news across media and the internet?

Disinformation is not just about “fake” and “true” news. Conceptually, we think that disinformation has preferred and repeated ways of presenting itself. But at the same time, language is complex, and the language structures and meaning may vary across languages and types of discourse (aggressive explicit hate speech versus implicit humor for example). LLMs are very good at detecting and analyzing at large-scale these systematic characteristics and their variations. 

Most people know LLMs as tools that help you write text (or code). But not only can genAI detect disinformation, it can also contextualize it for journalists and fact-checkers in a very simple form. In other words, just as you discuss the best birthday plans for a 4-year old with ChatGPT, with the model we aim to develop, you can discuss the characteristics of the misleading/false claim the AI has detected: where did it come from? Where else have we seen similar claims posted? In what languages? What are the biases this claim exploits, and the rhetorics it uses, to pass on disinformation? 

There are many things LLMs can do, but ultimately, this has to be useful to those at the frontline of the fight against disinformation. That’s why journalists and civil society activists are core, and participate, in this project.

Which topics does the project address, and why did you choose those? 

The European Commission defined the issues. The needle is moving from focussing on Covid-19 in particular, to other (and often related!) major concerns. Unsurprisingly, among those are the war of aggression against Ukraine. There is so much going on there online that there is a lot of data and knowledge that PROMPT can build on. There is also the increasing spread of false, harmful narratives related to gender identity and gender expression (gender-based disinformation) often seen in the context of discriminatory or pejorative speech against women, homophobia and discrimination against the LGBTQIA+ community, examining ways to mitigate the risks posed to societal cohesion, democracy and fundamental human rights. Misleading narratives around elections have of course a huge potential impact on public opinion, voter behavior and democratic processes. 

But as I mentioned, topics - migration, Ukraine - are not just a to-do list of disinformation topics. We know from our previous studies, from other research projects and from the work of journalists and fact-checkers, that accounts sharing disinformation post one day on Covid or vaccines, and the next to promote Putin’s version of what’s happening in Ukraine. Or that similar narratives and claims pop up from one election to the next. This means that PROMPT has a lot to take from, and give to, other initiatives to tackle disinformation.