Human-centered AI systems are intentionally built to align with human behavior and understanding.  When coupled with complex multimedia signals, this design becomes more arduous as the meaning and semantics have to be externalized and communicated in an understandable and actionable manner. In this talk, I will illustrate the design of human-centered solutions in three domains where I built AI to address real-world issues.  First, I will detail how editorial roles can be amplified (not replaced) by AI tools.  Next, I’ll describe an AI instrumented night club that enabled artists to make new experiences.  And finally, I will demonstrate how understanding CO2 literacy is an essential first step to creating AI for sustainable automotive behavior change. I will conclude with a discussion emphasizing the critical role ethnography and human-centered computing play in the future generation of AI research and tools.

Most technology design assumes we know how users interact with systems. But what happens when interactions occur in context, conversations, and around technology experiences difficult to grasp without human studies?

In this keynote, I explore invisible interactions that designers not always consider low-income women entrepreneurs in Brazil using informal networks and cash-stashing for microcredit; open-source contributors' uncompensated documentation and community management labor; call center coordination work that remains hidden; and collaborative AI exhibitions in museum settings where bias emerges as a criterion.

I present interfaces that can consider invisible interactions visible through design. Invisibility is systematically produced through design privileging certain users and formal systems over informal networks. I propose five design principles: map practices through ethnography, include marginalized groups in participatory design, surface background work, support existing practices, and design culturally embedded systems.

This keynote contributes an approach showing HCI must address interactions that remain hidden in context and conversation.

The impressive developments in language and image models have both opened new creative possibilities for artists and surfaced ethical concerns about the impact of generative AI upon the arts. I will discuss AI as a creativity support tool, focusing on live performance and interactive writing with generative AI as well as on the possibility for AI to mentor aspiring writers. I will illustrate the talk with an example of my theater company, Improbotics, that has used AI for improvised comedy since 2016 and recently engaged the wider public at Edinburgh Fringe, as well as my research on the socio-technical evaluation of generative AI tools for visual artists or for co-writing screenplays, theatre plays and comedy, integrating and welcoming critical feedback.

Data shape how we understand the world and make decisions. Yet their meaning is rarely apparent at first glance. Making sense of data requires people to organize, connect, and contextualize, often through representations that bring patterns and relationships into view.  This process is shaped by the interfaces through which people encounter data, influencing what they notice, how they interpret information, and which conclusions they draw. Through an HCI lens, my work examines how people with different levels of data literacy make sense of data in both everyday and professional settings.

In this talk, I will present research on data-centric sensemaking, beginning with work on dataset summaries and conceptual data models as ways to surface the diverse strategies people use to construct meaning, from staying close to the data itself to developing broader narratives. I will then turn to visualizations as a key interface through which people encounter, interpret, and make decisions with data. While widely used for communication, it is often unclear whether audiences understand visualizations as intended. Drawing on four years of research from my WWTF Talking Charts project, I will present studies examining how different audiences interpret messages conveyed through visualizations and how understanding can be assessed beyond measures of simple recall. The talk connects to broader HCI questions around Human-Data Interaction, explainability, and data communication, inviting reflection on how we might design tools for more meaningful encounters with data.