SEA: Interactions between humans and machines

This first SEA in 2021 will be all about interactions between humans and machines. We have two amazing speakers lined up: He He from New York University and Paul Bennett from Microsoft Research.

*** IMPORTANT: Make sure to (1) attend the meetup on the meetup page and (2) ensure you receive emails from Meetup. Shortly before the event we will send you the Zoom link and password to attend, as well as the info you need to log in via the browser (if your organisation does not allow you to install Zoom). You will only receive this if you have done both these steps. ***

** 17:00 - 17:30 CET - He He, NYU **

Title: Collaborative Dialogue between Humans and Machines

Abstract: Intelligent, interactive agents (e.g. home robots, self-driving cars) often need to infer intentions of their human partners and learning strategies for achieving specific goals. In this talk, I will describe our work on building collaborative dialogue agents that work with humans to complete tasks through text-based conversations. Two works will be covered: first, learning dialogue agents over knowledge graphs in a game setting; second, negotiation dialogue grounded in real scenarios. I will conclude with thoughts on challenges and future directions in building collaborative conversational agents.

** 17:30 - 18:00 CET - Paul Bennett, Microsoft Research **

Title: Robust and Transparent AI in Search and Recommendation

Abstract: AI models power many search and recommendation tasks in many different domains including the web, enterprise, and e-commerce applications. Within these scenarios as well as other applications, there are key challenges to providing AI that is robust in the context of changing inputs or policies and in how results are surfaced to the people using the system. In this talk, I will overview recently published advances on three key challenges in these domains made by Microsoft researchers and our collaborators. In particular, how can we effectively customize models using few-shot learning and RL to new tasks or applications given very little labeled data; how can we learn more robust models in the presence of skewed data distributions or underlying toolchain nondeterminism, and how can we design UX interventions that improve the transparency to how the underlying AI system behaves? I will then conclude the talk with thoughts on future directions and opportunities.

After the two talks we leave the Zoom call open for another half an hour, for any remaining questions. This always result in a nice discussion!