DisAI Project

Enhancing Fact-Checking Transparency 🔨 ✅ In our ongoing efforts to improve the explainability and trustworthiness of fact-checking systems, we've developed a systematic approach to collecting annotations for explainability purposes 📊 ✨ These annotations are crucial components that help bridge the gap between claims and their verification by highlighting the specific elements that connect posts to fact-checked claims from a user's perspective📲 The resulting dataset extension will be an important resource for developing comprehensible explanations and more transparent fact-checking systems. By incorporating these explanation labels, we will enable: 🟣 Comprehensively identify "why" particular posts are associated with particular fact-checked claims 🟣 More transparency in automated fact-checking decisions 🟣 A better understanding of the reasoning behind fact-checking matches

These days, machine learning is all about #LLMs 💻 💬 In DisAI, we care about leveraging the latest and greatest in NLP to tackle the challenges of searching for existing #factchecks across different languages 🛡️ LLMs, with their improved contextual understanding, should be able to outperform earlier methods in determining what is and is not relevant. ❓The question remains: does this also hold for smaller, open-source LLMs? And, more importantly, will it work across different languages? To explore this potential, we are evaluating open-source LLMs to see how well they perform in practice, especially for low-resource languages 👉 Stay tuned as we dig into the data and uncover what these models can (and can’t) do for this important task! 😊

🗣️ Do We Want Better Explainable #AI? Then we need to put humans in the loop! 💪 Think about it - what good is AI transparency if humans can't understand it? This question is crucial as we develop tools to detect and counter disinformation 🙅 Shifting the focus from purely technical metrics to approaches that meaningfully empower human fact-checkers and content moderators is essential if we are to build trustworthy tools 🔍 An explainability system for such tools must help users understand why content is flagged, reveal patterns of manipulation, and support human judgment in complex cases ⚠️ Real understanding comes from feedback, testing, and refinement with the people who'll actually use these systems. Therefore: ✔️ The best XAI systems are built iteratively, with continuous human feedback guiding improvements. ✔️ Simply making AI transparent isn't enough - explanations need to be tested and refined with real users. The future of #XAI? 👉 Creating explanations that work for people.

🤖✨ Revolution in AI is constantly going. Almost every day, new sophisticated AI models are released, changing the landscape of IT and data processing 💻 The models are capable of processing text, images, videos and all these modalities simultaneously. We decided to use them in our fight against disinformation 📱💬 In the DisAI project, we used state-of-the-art, powerful multimodal models to obtain high-quality OCR (Optical Character Recognition) of images 🤳We asked a smart multimodal model to do the OCR for us. The results were amazing. The additional information in the form of OCR text will help us create retrieval tools to fight disinformation 💪

😍 It is amazing how fast things are moving in natural language processing! When we began work on the DisAI project, multilingual retrieval models were still teething and the most effective way to search for existing fact-checks across languages was to translate everything into English and fine-tune an English-language model to the task 🪄⌛ Fast forward just over a year, and we can now achieve better results directly in the original languages – and with no fine-tuning required ➡️ The key lies in knowing which models to choose and how to use them effectively. 🤞 Fingers crossed that these breakthroughs will soon significantly enhance the tools available to fact-checkers worldwide! #nlp #project

📢 Have you heard? We are organizing a "Multilingual and Crosslingual Fact-Checked Claim Retrieval" shared task at the well-known SemEval workshop next year! 🤩 In the DisAI project, we are interested in supporting fact-checkers in their mission to improve the quality of information online 📲 When you come across a new factual claim, it is often difficult to find out if and where it has been fact-checked 🧑💻 Fact-checks are scattered across different platforms and written in different languages! We are tackling this problem using the latest and greatest in the field of language technologies 🫵 If you care about the quality of information online and think you could make a difference, join the shared task and help us push the envelope of what is possible 🖇️ Read more on our website: https://lnkd.in/d7iHPSew

⚖️ We take ethics as an integral part of advancing AI technology! That's why we were thrilled to welcome Prof Dr Maximilian Kiener at the DisAI Summer School with his keynote on Deep Ethics 🎤 👉 Prof. Kiener presented a framework for bridging the gap between the qualitative world of ethics and the quantitative realm of machine learning. His approach to integrating ethical principles like autonomy and justice directly into ML reward functions provided valuable insights for the summer school participants. We believe that by networking with international experts in #AI ethics, we can ensure that ethical considerations are a driver for innovation rather than an afterthought 🫶

🎤 Marián Šimko and Michal Gregor represented Kempelen Institute of Intelligent Technologies and our DisAI Project at the #LDSCountryWorkshop in Slovakia organized by European Language Data Space and Slovenská akadémia vied / Slovak Academy of Sciences 🤖 Our DisAI project was used as a representative example of why development and maintenance of high-quality language resources is critical for solving important societal issues like disinformation and other information disorders, especially when considering limited-resource languages like Slovak 🧑💻✨ Being a part of the LDS community falls within our efforts to participate in impactful European networks related with language technology. Thanks Simon Ostermann and Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) to guiding us in this endeavor 🤝

📣 AI designers, builders and believers meet next Wednesday at the BETTER_AI_MEETUP at the Kempelen Institute of Intelligent Technologies🌃 Together with experts from the banking industry, we will discuss how LLMs and GenAI are improving customer service in banking 🏦 You can also look forward to a talk by Martin Tamajka, Research engineer at KInIT 👨💻 ✅ Free registration for this online event: https://lnkd.in/dJdn7M-V 📅 13. November at 17:00 (CET)

✨ We are ready for the next BETTER_AI_MEETUP! Will you join us? 🫵 Our main topic will be finances and AI. How are LLMs and genAI improving customer service in banking? 💳 Find the answers with us during the 13th meetup 👨💼 We will discuss the use of AI with the specialists from the largest Slovak banks: Tatra banka, ČSOB Finančná skupina, VÚB banka, Slovenská sporiteľňa and Národná banka Slovenska 🏦 Also, we will welcome a guest from the banking regulator to shed light on their stance on using AI in financial services. 📅 13. November at 17:00 (CET) 💻 Free registration: https://bit.ly/3YpO5P9