Monday, February 6, 2023, 10am

Natural language communication has long been considered a defining characteristic of human intelligence; as such, it acts as a north star in the pursuit of artificially intelligent agents. However, despite the recent explosion of interest and progress in the broad field of Artificial Intelligence, agents that can communicate both with humans and each other remain rather primitive. Many autonomous agents, such as Siri-like voice assistants, lack the ability to learn from their failures or draw effectively from both immediately relevant context and worldly commonsense knowledge. Unless addressed, these flaws will continue to hamstring future attempts to create practically usable natural language agents.

This talk thus focuses on the research question of how to imbue learning agents with the ability to understand and generate contextually relevant natural language in service of achieving a goal. It will focus on creating reinforcement learning agents inspired by communication strategies in humans that can: (1) interactively gather and align to human preferences via feedback to shape “how” a given task is executed, and (2) build and use neuro symbolic world models that tell an agent “what” action to take in the current context and “why” to take that action given the dynamics of a grounded environment.

—

Prithviraj Ammanabrolu is a postdoctoral researcher at the Allen Institute for AI advised by Professors Yejin Choi and Hannaneh Hajishirzi. He received his PhD in Computer Science from the School of Interactive Computing at the Georgia Institute of Technology, advised by Professor Mark Riedl. His research focuses on creating trustworthy and practically usable language agents via interaction and environment grounding. This work spans the fields of reinforcement learning, natural language processing, computational creativity, and knowledge graphs. It has been recognized at a broad array of conferences such as NeurIPS, ICLR, *ACL, EMNLP, AAAI, IJCAI, AIIDE, and ICCC.  Learn more

Event Type: Talks
Room Number: In Person and Virtual - ET
Building: Newell-Simon 4305 and Zoom
Speaker's Name: PRITHVIRAJ AMMANABROLU
Speaker Website: prithvirajva.com
Speaker's Professional Title: Postdoctoral Researcher, Allen Institute for AI
Talk Title: Interactive and Grounded Language Learning
For More Information: astowers@andrew.cmu.edu
Affiliations: Language Technologies Institute (LTI)
Organization(s): SCS