Periodic Reporting for period 2 - AI4Debunk (Participative Assistive AI-powered Tools for Supporting Trustworthy Online Activity of Citizens and Debunking Disinformation)
Período documentado: 2025-01-01 hasta 2025-12-31
To revolutionize the fight against disinformation, a four-year innovation action, AI4Debunk, has been launched January 2024. Funded by the European Union (EU) through the Horizon Europe Programme under the Grant Agreement No. 101135757, the project brings together an interdisciplinary consortium of 13 partner entities from eight countries towards a shared mission: to support trustworthy online activity by providing citizens with human-centered AI-powered tools. This EU project is committed to safeguarding democratic values and fostering a more informed and resilient society in response to the challenges posed by the digital age.
The partnership involves various stakeholders, from media professionals to software developers and AI experts, this new EU project takes a holistic approach to fighting disinformation through Human-AI collaboration. It will be validated with two case studies: Russian propaganda related to the war in Ukraine and manipulated content on climate change.
AI4Debunk aims to develop four human-centered AI-powered interfaces: a web plug-in, a collaborative platform, a smartphone app, and an Augmented Reality/Virtual Reality interface based on the first-of-its-kind open-source debunking API. The plug-in will be designed for seamless integration with web browsers and social media platforms, delivering instant notifications to users when they encounter false content, regardless of the format. As for the collaborative platform (Disinfopedia), it will allow proactive users to report suspicious content, which will then be checked by human experts and potentially removed from circulation. The app, in turn, will enable people to detect disinformation in everyday life directly through their smartphone. Lastly, the AR/VR interface will provide guidelines on how to deal with disinformation in future AR/VR-based social media. AI4Debunk’s interfaces cover different needs – from real-time analysis of multimodal content to community-driven reporting and immersive experiences. These will provide citizens with a comprehensive set of fact-checking resources to navigate the digital media landscape more consciously and make informed decisions. These interfaces will also be innovatively and playfully designed for educational purposes to help students become discerning consumers of information.
Links
- Website: ai4debunk.eu
- LinkedIn: linkedin.com/company/ai4debunk
- Twitter/X: twitter.com/ai4debunk
- Facebook: facebook.com/ai4debunk
- YouTube: youtube.com/@ai4debunk
Contacts
E-mail: info@ai4debunk.eu
The partnership involves various stakeholders, from media professionals to software developers and AI experts, this new EU project takes a holistic approach to fighting disinformation through Human-AI collaboration. It will be validated with two case studies: Russian propaganda related to the war in Ukraine and manipulated content on climate change.
AI4Debunk aims to develop four human-centered AI-powered interfaces: a web plug-in, a collaborative platform, a smartphone app, and an Augmented Reality/Virtual Reality interface based on the first-of-its-kind open-source debunking API. The plug-in will be designed for seamless integration with web browsers and social media platforms, delivering instant notifications to users when they encounter false content, regardless of the format. As for the collaborative platform (Disinfopedia), it will allow proactive users to report suspicious content, which will then be checked by human experts and potentially removed from circulation. The app, in turn, will enable people to detect disinformation in everyday life directly through their smartphone. Lastly, the AR/VR interface will provide guidelines on how to deal with disinformation in future AR/VR-based social media. AI4Debunk’s interfaces cover different needs – from real-time analysis of multimodal content to community-driven reporting and immersive experiences. These will provide citizens with a comprehensive set of fact-checking resources to navigate the digital media landscape more consciously and make informed decisions. These interfaces will also be innovatively and playfully designed for educational purposes to help students become discerning consumers of information.
Links
- Website: ai4debunk.eu
- LinkedIn: linkedin.com/company/ai4debunk
- Twitter/X: twitter.com/ai4debunk
- Facebook: facebook.com/ai4debunk
- YouTube: youtube.com/@ai4debunk
Contacts
E-mail: info@ai4debunk.eu
During the 2nd year of activity the work was continued as planned. All tasks, 20 deliverables and 4 milestones were accomplished on time – updated management documents; two Working papers on disinformation target groups, sources and hosts of propaganda and narratives and foreign interference; starting dataset of fake statements and related multimedia contents; updated release of the dataset containing extracted features; Report the requirements; the building process of the knowledge graphs and their continuous graph adaptation has been ensured and modules developed; the multimodal fake news detection modules and multimodal fake news dataset; the trustworthiness of different modules developed; initial calculation of a score representing the amount of disinformation in the data; Initial explainability module tracing back between the data and the score; Reports on 1) the possible impacts of the tool on the perceptions of the citizens and the social media users; 2) the resilience mechanisms triggered by the tools; 3) the desk review analysis; 4) the multi-stakeholders perspectives and Gender Equality Plan.
In 2025, the project further developed a framework for disinformation research by integrating societal analysis, advanced artificial intelligence, and outcomes related to policy making. The analysis identified population groups particularly vulnerable to disinformation, including elderly people, minorities, immigrant communities, and geographically or educationally disadvantaged groups, showing how exposure, susceptibility, and impact vary across contexts.
The research applied narrative analysis, discourse analysis, forensic linguistics, and case-study methods to examine how foreign interference campaigns exploit polarizing themes linked to geopolitics, climate change, and identity. Profiling of state and private actors provided context within European strategic communication and cybersecurity frameworks. Research outputs were translated into working papers and policy briefs to support technological work and policy making.
From a technological perspective, the work combines curated, fact-checked multimodal datasets with machine learning and semantic knowledge representation. Linguistic, visual, and auditory features are embedded into web-standard-based knowledge graphs, enabling explainable analysis, semantic querying, and human-in-the-loop adaptation. A modular multimodal AI architecture has been integrating similarity analysis, deepfake detection, and cross-modal coherence checking to generate an interpretable probabilistic disinformation score supported by trustworthy AI methods.
User-centered, ethical, and inclusive design principles are being embedded in the work. Multilingual surveys, beta testing with real-world cases, and multi-stakeholder focus groups assess public perception, trust, and resilience, while gender, equity, and bias analyses support tool development.
The research applied narrative analysis, discourse analysis, forensic linguistics, and case-study methods to examine how foreign interference campaigns exploit polarizing themes linked to geopolitics, climate change, and identity. Profiling of state and private actors provided context within European strategic communication and cybersecurity frameworks. Research outputs were translated into working papers and policy briefs to support technological work and policy making.
From a technological perspective, the work combines curated, fact-checked multimodal datasets with machine learning and semantic knowledge representation. Linguistic, visual, and auditory features are embedded into web-standard-based knowledge graphs, enabling explainable analysis, semantic querying, and human-in-the-loop adaptation. A modular multimodal AI architecture has been integrating similarity analysis, deepfake detection, and cross-modal coherence checking to generate an interpretable probabilistic disinformation score supported by trustworthy AI methods.
User-centered, ethical, and inclusive design principles are being embedded in the work. Multilingual surveys, beta testing with real-world cases, and multi-stakeholder focus groups assess public perception, trust, and resilience, while gender, equity, and bias analyses support tool development.