On 9 July, INESC-ID will host two Spotlight Talks. Iolanda Leite, associate professor and researcher at the KTH Royal Institute of Technology will present “Robots (Still) Need Humans in the Loop”, defending that robots need human feedback to learn safely and effectively, showing how interactive robot learning helps align robots with human priorities even in non‑social tasks. André Pereira, researcher at LTH Royal Institute of Technology, will follow with the presentation “Towards Socially Embodied AI: Understanding and Generating the Dynamics of Conversation”, where he shows how socially embodied AI aims to move beyond fluent responses toward systems that can actively engage in meaningful, adaptive conversation by understanding human social dynamics through multimodal learning and neuroimaging.

Date & Time: 9 July, 13h30 – 14h30

Location: FA2, Informatics Pavilion I

Abstracts:

“Robots (Still) Need Humans in the Loop”: As robots become more prevalent in people’s lives, a key challenge remains: ensuring their models align with how humans represent and prioritise information, especially when these robots interact with or operate around people. In this talk, I present research on interactive robot learning (teaching robots from feedback such as preferences, demonstrations, and language) including methods that reduce the cost and ambiguity of human feedback. I argue that human-robot interaction is essential even for non-social tasks: it is a foundation mechanism for ensuring that robots and other AI systems learn to operate safely and effectively in the real world.

“Towards Socially Embodied AI: Understanding and Generating the Dynamics of Conversation”: Artificial intelligence is becoming increasingly capable in language, perception, and action, yet most systems still struggle with the social dynamics that make interaction feel meaningful, engaging, and adaptive. In this talk, I will present my research agenda on socially embodied AI: systems that do not merely process social signals, but actively participate in interaction in ways that are sensitive to people, context, and ongoing conversational dynamics. The central goal is to move beyond AI systems that simply respond fluently, towards systems that can engage socially, adaptively, and responsibly. I will first discuss how multimodal machine learning can be used to measure and predict human experience in interaction, including enjoyment and engagement in human-robot conversations. I will then present work combining neuroimaging and multimodal data to study the brain and behavioural mechanisms associated with naturalistic conversation. Together, these lines of work inform the design of embodied conversational agents and social robots, including systems that integrate scripted interaction, generative AI, and collaborative task environments.

Bios:

Iolanda Leite: “I am an Associate Professor at the Division of Robotics, Perception and Learning. I received my PhD degree from the Technical University of Lisbon. Before joining KTH, I was a Postdoctoral Associate at the Yale Social Robotics Lab and an Associate Research Scientist at Disney Research. The goal of my research is to develop robots that can capture, learn from, and respond appropriately to the subtle dynamics that characterize real-world situations, allowing for truly efficient and engaging interactions with people.” (Source: KTH website)

André Pereira: “I am a researcher at the Division of Speech, Music and Hearing (TMH). I design, implement and evaluate socially intelligent systems, typically robots, that can interact naturally with people in human-centered domains like education, healthcare, and entertainment. My primary research objective is to create autonomous embodied agents that can socially interact, in real-time, with human users throughout extended periods. These systems can establish social relationships, simulate empathy and emotions, generate believable gaze and dialog.” (Source: KTH website)