Keynote Speaker

Many conversational systems focus on providing a most likely response to a given text chat context. While current large language models allow this technique to result in very fluent responses, for Multimodal Agentic, Collaborative and Socially Aware (ACSA) systems, this is often not enough. In addition to the chat context, some other very important motivations include what is happening while the conversation is going on, and the social and physical goals of the agent as well as the interlocutor. Reacting to the situation is sometimes not enough and a form of planning may innecessary to bring the state closer to one in which goals may be accomplished. Mixed initiative dialogue balances reactive and responsive behavior with goal-directed reasoning and planning to achieve goals.

In this talk I will illustrate some of these ideas with recent work embedding agents within social, physical and simulated worlds and architectures to support mixed initiative interaction.