This PhD course explores the emerging field of Agentic AI: systems designed to perceive, decide, and act responsibly within human and ecological contexts. In Part 1, students explore the foundations of human-centered and trustworthy agentic intelligence, examining how autonomy, adaptivity, and proactivity can align with human oversight and empowerment. Through the concept of augmented judgment, learners design AI tools that enhance, rather than replace, human reasoning by surfacing alternatives, expressing uncertainty, and fostering collaboration between humans and machines. The course then extends beyond anthropocentric design to consider more-than-human Agentic AI, where systems are built to respect ecological interdependence and nonhuman stakeholders such as soils, species, and future generations. In Part 2, the focus shifts to Agentic AI safety, addressing how these systems can operate ethically and robustly in complex social environments. Students study the biases inherent in toxicity and sentiment analysis, design agentic moderation and fact-checking architectures, and analyze information diffusion in fringe online communities. Finally, they investigate agentic systems for political messaging ecosystems, exploring coordination, misinformation, and multimodal propaganda across closed networks. Together, these modules equip students to design AI systems that are intelligent, trustworthy, and responsive to both human and planetary well-being.

Guest lectures: Yvonne Rogers Internationally recognized leader in Human–Computer Interaction, ubiquitous computing, and human-centred AI. Elected International Member of the U.S. National Academy of Engineering, Fellow of the Royal Society, and recipient of major international honors including the Royal Society Robin Milner Award, the ACM SIGCHI Lifetime Research Award, and an EPSRC Dream Fellowship. He is also a Fellow of the ACM, BCS, and the ACM CHI Academy, and holds an honorary doctorate from the University of St Gallen. His research focuses on designing interactive and intelligent technologies that augment human cognition, learning, and everyday work. He has been instrumental in shaping HCI through foundational theories, innovative methodologies, and influential research agendas. He has published over 300 works, including 12 monographs, with more than 51,000 citations (h-index 89), co-authored the leading textbook on Interaction Design, and secured over £30 million in research funding in collaboration with academia and industry. Fabricio Benevenuto de Souza He is an Associate Professor in the Department of Computer Science at the Federal University of Minas Gerais (UFMG), Brazil. He holds a PhD, MSc, and BSc in Computer Science from UFMG. His research career includes visiting and postdoctoral positions at the Max Planck Institute for Software Systems and a research internship at HP Labs (USA). He has received several prestigious international awards, including the CNIL–INRIA Privacy Protection Prize, Google Latin American Research Awards, a Humboldt Fellowship, and the CAPES Thesis Award for the best PhD thesis in Computer Science in Brazil. He also serves as a consultant to TikTok on its Safety and Content Advisory Council. Part 1: Introduction to Agentic AI and their Design Foundations of Human-Centered and Trustworthy Agentic AI Introduction of the principles of agentic intelligence: AI systems that perceive, decide, and act within human-defined contexts. Students explore how autonomy, adaptivity, and proactivity can coexist with human oversight, trust, and empowerment. Augmented Judgment: Designing AI that Complements Human Reasoning How to build decision-support tools that surface alternatives, convey uncertainty, and strengthen human judgment, so the system thinks with people, not for them. Foundations of more-than-human Agentic AI How AI systems can be designed to restore ecological health and acknowledge more-than-human stakeholders: soils, watersheds, species, and future generations. Designing more-than-human Agentic AI Frame AI projects using a more-than-human stakeholder map and metrics beyond human utility. Part 2: Focus on Agentic AI Safety A Comprehensive View of Toxicity and Sentiment Analysis Methods in Agentic AI Examine how language models encode and reproduce social and dialectal biases in toxicity and sentiment analysis, and how to design fairer, context-aware NLP systems. Agentic AI for Online Safety: Architectures for Moderation and Fact-Checking Agents Design and evaluate agent-based systems that proactively detect, verify, and contextualize harmful or misleading content, complementing human moderators and fact-checkers. Information Diffusion and Coordination in Fringe Online Communities Analyze how information and misinformation spread in fringe or alternative online ecosystems, and identify the dynamics, network structures, and coordination strategies that sustain them. Agentic AI for Political Messaging Ecosystems: Coordination, Stickers, and Forwarding Chains Develop sensing and analytic agents for closed-messaging environments (e.g., WhatsApp) to study political communication, coordinated behavior, and multimodal propaganda flows. Students conduct an interdisciplinary research or design project that applies the principles of trustworthy, human-centered agentic AI to a chosen domain (e.g., health, education, public service, or creative industries). Expected outcomes: • A research paper or design documentation demonstrating the integration of ethical, technical, and social dimensions • Reflection on trust, alignment, and human collaboration within agentic AI

In presenza

Sviluppo di project work in team

P.D.2-2 - Maggio