Abstract

The next phase of exponential AI growth will not come from building better individual models, but from building better human-AI teams. While AI has already transformed the vast majority of job roles, we are only beginning to understand the scientific foundations needed to make humans and AI agents collaborate effectively. This talk presents five foundational challenges for human-AI collaboration: (1) Understanding Groups: formally modeling how heterogeneous human and AI agents interact, form coalitions, and achieve collective intelligence that exceeds any individual contributor; (2) Uncertainty Estimation: knowing when to involve whom in a collaborative decision, spanning LLMs and vision-language models through methods such as MARS and LARS; (3) Universal Memory: designing shared memory architectures that give teams persistent, contextual knowledge across sessions, enabling true collaborative cognition rather than isolated exchanges; (4) Scale-Out Efficiency: addressing the unsustainable cost of agentic AI systems through an information-theoretic lens, where we introduce R_LLM(D), a rate-distortion framework providing principled trade-offs for context compression, prompt optimization, and model routing; and (5) Trust and Alignment: tackling cascading failures, groupthink, and accountability gaps unique to collaborative human-AI settings.

Prof. Salman Avestimehr is a Dean's Professor, the inaugural director of the USC-Amazon Center on Secure and Trusted Machine Learning (Trusted AI), and the director of the Information Theory and Machine Learning (vITAL) research lab at the Electrical and Computer Engineering Department of University of Southern California. He is also the co-founder of FedML, TensorOpera AI, and ChainOpera AI. He received his Ph.D. in 2008 and M.S. degree in 2005 in Electrical Engineering and Computer Science, both from the University of California, Berkeley. Prior to that, he obtained his B.S. in Electrical Engineering from Sharif University of Technology in 2003. His research interests include information theory, decentralized and federated machine learning, secure and privacy-preserving learning and computing.

Dr. Avestimehr has received a number of awards for his research, including the James L. Massey Research & Teaching Award from IEEE Information Theory Society, an Information Theory Society and Communication Society Joint Paper Award, a Presidential Early Career Award for Scientists and Engineers (PECASE) from the White House (President Obama), a Young Investigator Program (YIP) award from the U. S. Air Force Office of Scientific Research, a National Science Foundation CAREER award, the David J. Sakrison Memorial Prize, and several Best Paper Awards at Conferences and Workshops. He has been an Associate Editor for IEEE Transactions on Information Theory and a general Co-Chair of the 2020 International Symposium on Information Theory (ISIT). He was an Amazon Scholar in Alexa AI in 2021. He is also a fellow of IEEE.

Abstract

As intelligent systems become deeply embedded in everyday environments—from smart homes to assistive robots—their increasing complexity risks undermining user understanding, trust, and control. This talk introduces the concept of responsive artificial intelligence, a shift from systems that simply execute actions to systems that actively interpret context, adapt to users, and support human understanding and control. The talk presents a research journey across three domains: IoT automations, robot-assisted cognitive training, and accessibility validation. It shows how Large Language Models (LLMs) can support semantic interpretation, explain complex behaviours, and generate personalised and meaningful interactions. These capabilities enable a new generation of systems that respond not only to inputs but also to user goals, context, and experience, promoting inclusion, responsible practices, and human oversight. Drawing on multiple research projects—including human control in smart environments, humanoid robots, and accessible support—the talk discusses how AI systems can be exploited to obtain more transparent, trustworthy, and socially sustainable applications. It highlights design approaches for explaining automations, detecting and resolving conflicts, analysing activation chains, and addressing related security and privacy issues, showing how to obtain user understanding and control in smart environments. Building on this perspective, the talk explores the transition to Small Language Models (SLMs), which enable more privacy-aware, locally deployable intelligent support. It concludes by outlining design principles and open challenges for responsive intelligent systems, emphasising the need to balance automation with transparency, personalisation with control, and technological power with human values.

Prof. Fabio Paternò is Research Director at the Italian National Research Council, where he leads the Laboratory on Human Interfaces in Information Systems at the Information Science and Technologies Institute, in Pisa. He has been a principal investigator in several international and national projects, mainly in the area of Human-Computer Interaction. His current research interests are Interactive Smart Spaces, Accessibility, End-User Development, Human-centered Artificial Intelligence, and Humanoid Robots. He is an IFIP Fellow and a member of the ACM SIGCHI Academy.

Keynote Speakers

Meet the Keynotes

Professor Salman Avestimehr

Professor Fabio Paternò

Professor Daniella Tilbury

Be there for the keynotes