AIXPERT

Can AI see the world without stereotypes? Why accuracy isn't enough for trustworthy AI [BLOG] It's tempting to answer that question by looking at benchmark scores or accuracy. But as multimodal AI systems become more capable, researchers are finding that those metrics tell only part of the story. A new blog post on the AIXPERT website explores recent research into how multimodal models interpret people, professions and cultures and why strong technical performance does not necessarily imply fair or unbiased reasoning. The work spans text-to-image generation, vision-language models, multilingual AI and audio-video understanding, revealing how demographic bias can persist even in state-of-the-art systems. The article also looks at emerging evaluation frameworks that assess AI systems across dimensions such as fairness, explainability, faithfulness and human-centred reasoning. Together, these benchmarks point towards a broader shift in AI research: moving beyond measuring what models can do, to understanding how they reach their conclusions. As AI finds its way into healthcare, recruitment, manufacturing and other high-impact domains, robust evaluation will become an increasingly important part of building systems that deserve public trust. 👉 Read the full article here: https://lnkd.in/evxJ2MZG This body of research features important contributions led by AIXPERT partner Vector Institute whose work on benchmarking multimodal foundation models is helping advance human-centred and trustworthy AI. Shaina Raza, PhD Aravind N. Ahmed Radwan Christos Emmanouilidis Vahid Reza Khazaie

🔴 📹 (Re)introducing AIXPERT: Watch our new project video How can increasingly autonomous and powerful AI systems remain explainable, accountable and centred on human needs? ▶️ Watch our new project introduction video to discover the vision behind AIXPERT and how we are working towards AI that is not only more capable, but also more explainable, transparent, accountable and trustworthy. 🔗 https://lnkd.in/ekTm26tJ Working on trustworthy AI, agentic systems, multimodal AI, XAI or human–AI collaboration? Watch the video, follow AIXPERT and share it with your network. Athena Research Center Infinitivity Design Labs Barcelona Supercomputing Center Furhat Robotics Vector Institute Amsterdam UMC CNRS Sciences informatiques Kyklos health-tech Workable Orfium Novelcore ITML Sorbonne University Universitat de Barcelona University of Groningen Martel Innovate RobustifAI project TRUMAN - TRUstworthy huMAN-centric artificial intelligence Turing Project HumAIne project AI4REALNET Project TANGO Project THEMIS 5.0 PEER AI

How do we know whether multimodal AI systems are actually trustworthy? #HAICon2026 As Large Multimodal Language Models (MLLMs) become increasingly capable, developing robust evaluation frameworks is becoming just as important as building better models. At HAICon 2026 – Helmholtz AI Conference: AI for Science, AIXPERT partner Vector Institute contributed to this important discussion. During the workshop "Current Status of the Benchmarking Field: Lessons Learned from the First Half of the UNLOCK Initiative," Shaina Raza, PhD presented HumaniBench, a human-centric benchmark designed to evaluate large multimodal models beyond conventional performance metrics. The workshop brought together experts working on benchmarking across AI safety, chemistry, genomics and energy-efficient AI, highlighting how meaningful benchmarks are essential for improving robustness, transparency, reproducibility and trust in AI systems. For AIXPERT, advancing reliable evaluation methodologies is a key step towards building explainable, human-centred and trustworthy AI that can be confidently deployed in real-world applications. Read more about our contribution to HAICon 2026: 🔗 https://lnkd.in/e5g2ivwR

We're delighted to welcome you to the second edition of the #AIXPERT newsletter—this time in a new format! 🗞️ One year after the launch of our project, our multidisciplinary consortium has reached important milestones and continued to expand its impact. This first Summer Pulse edition is dedicated to our #collaboration with fellow projects, our engagement with the wider #AICommunity, and our participation in international conferences, workshops, and events. We hope you enjoy the read! 😎 To stay up to date with the latest AIXPERT news and future editions, don't forget to hit the subscribe button! 👉 https://lnkd.in/eRs-5Ew5 #HaDEA #DigitalEurope #HorizonEurope #AgenticAI

I was honored to present my work on Trace-Aligned Diagnostics for Agentic AI Pipelines in the 32nd ICE/IEEE ITMC conference in Porto, Portugal! As part of the AIXPERT project, we investigate and evaluate trustworthiness and explainability in agentic AI systems. One key factor is alignment with agentic execution traces, and this work presents and analyses a simple trace-aligned diagnostic classifier for debugging errors in agentic pipelines, trying to reverse the trend of always using zero-shot LLMs or classic XAI methods wherever an explainability component is needed. In Porto, I had the opportunity to meet and discuss with many great and like-minded researchers, and broaden my understanding on the field which I hope will help in future works in my doctoral research. I also have to thank my co-authors Christos Emmanouilidis, Laura Maruster and Shaina Raza, PhD for their constant feedback and support!

🔍 How do we ensure AI systems remain trustworthy when deployed in the real world? 🌍 This question brought together experts from leading EU-funded AI projects at the ICE/IEEE ITMC Conference in Porto, where our partner Faculty of Economics and Business - University of Groningen joined the conversation on the future of robust, explainable and accountable #AI. From cross-project collaboration to emerging approaches for evaluating trustworthy AI, the discussions highlighted why building reliable AI is about much more than algorithms. The conference also featured new research from our partner University of Groningen on improving the #reliability and #accountability of #AgenticAI systems. Curious about the key takeaways and our contribution? Read the full story below. 👇 https://lnkd.in/eZY2iVK4 #HaDEA #TrustworthyAI #ResponsibleAI #ExplainableAI #ArtificialIntelligence #HorizonEurope #EUProjects #Innovation #AIResearch #Engineering #DigitalTransformation #IEEE #AIGovernance

Summer may be here, but our partners are not slowing down! 👊 On 18 June, Ahmed Radwan, Associate Applied Machine Learning Specialist at the Vector Institute, took the stage at #TMLS2026 in Toronto. Ahmed presented “SONIC-O1: A Real-World Benchmark for Evaluating Multimodal LLMs on Audio-Video Understanding,” as part of the Technical & Engineering Talks track, contributing to discussions on Evaluation Methods & Capability Benchmarking. We're proud to see our Canadian partners actively sharing their expertise and advancing the conversation around applied AI. 🇨🇦🚀 Have a look at our journey to make AI more human-centred 👉 https://lnkd.in/eJTC8RaT #HorizonEurope #HaDEA #XAI #TrustworthyAI #DigitalEurope #AIAgents

🚀 Robust and trustworthy AI is not one-size-fits-all! We are excited to share that #AIXPERT will be presented at the 32nd Engineering, Technology and Innovation Conference (ICE/IEEE ITMC) by Christos Emmanouilidis of the Faculty of Economics and Business of the University of Groningen in an insightful workshop. The session will explore how AI robustness, reliability, transparency, fairness, safety, and user trust must be addressed differently across domains, reflecting diverse constraints, risks, and stakeholder needs. By bringing together multiple AI projects, the workshop will compare approaches and highlight how trustworthy AI principles are translated into real-world design, implementation, and evaluation. Plenty of insights ahead and lessons to learn! 🤝 We are very much looking forward to it, and are happy to showcase first joint efforts with fellow partners from Turing Project, HumAIne project and TRUMAN - TRUstworthy huMAN-centric artificial intelligence! 🤝 🇪🇺 #ArtificialIntelligence #TrustworthyAI #RobustAI #AIResearch #IEEE # #HorizonEurope #HorizonEurope #HaDEA #AIGovernance #Innovation #DigitalEurope More here 👉 https://lnkd.in/eGeWyyRg

AIXPERT meets in Barcelona to shape the next phase of trustworthy AI development This week, the AIXPERT consortium met at the Barcelona Supercomputing Center for its General Assembly and Technical Meeting, bringing together experts in explainable AI, agentic systems, multimodal foundation models, governance, ethics, and human-centred design. As the project enters its second year, partners aligned the technical roadmap, validated implementation plans for the five pilot domains, and prepared the next phase of platform development and integration. From healthcare and recruitment to manufacturing, education, and the creative industries, AIXPERT is advancing a new generation of AI systems designed to be transparent, accountable, and explainable by design. Read our latest news article to discover the key outcomes of the meeting and what's next for the project: https://lnkd.in/eYnHXzY4 Athena Research Center Novelcore Amsterdam UMC University of Groningen Philips Martel Innovate Infinitivity Design Labs Furhat Robotics CNRS Sciences informatiques Sorbonne University Vector Institute Kyklos health-tech Workable Orfium Universitat de Barcelona ITML

A quick throwback to last week, when our partner Barcelona Supercomputing Center participated in the 25th International Conference on Autonomous Agents and Multi-Agent Systems (#AAMAS) in Paphos, Cyprus, and delivered a tutorial titled “Approaches for Explainability in Autonomous Agents.” AAMAS is a leading international conference where researchers present advances in #AutonomousAgents and multi-agent systems; software systems that can act and interact independently in complex environments. Stay tuned for more updates! #HorizonEurope #AIXPERT #HaDEA #XAI #HumanCentredAI