Over the past few years, methods for learning from interaction have become a crucial paradigm in Artificial Intelligence, and we are now witnessing a growing interest in learning from grounded interaction, in particular through dialogue games. In the Dagstuhl Perspectives Workshop 24492, "Human-in-the-Loop Learning through Grounded Interaction in Games", we discussed these new developments, and identified a few crucial directions for this research. These directions were considering agent behavior in complex interaction; ensuring that games properly tested all aspects of an agent’s cognitive and communicative ability; considering the types of grounding required at all levels of interaction; and developing new training methods that could fully leverage these richer types of context and communication.
@Article{bernardi_et_al:DagRep.14.12.28, author = {Bernardi, Raffaella and Hockenmaier, Julia and Kruschwitz, Udo and Jayannavar, Prashant and Poesio, Massimo}, title = {{Human in the Loop Learning through Grounded Interaction in Games (Dagstuhl Perspectives Workshop 24492)}}, pages = {28--45}, journal = {Dagstuhl Reports}, ISSN = {2192-5283}, year = {2025}, volume = {14}, number = {12}, editor = {Bernardi, Raffaella and Hockenmaier, Julia and Kruschwitz, Udo and Jayannavar, Prashant and Poesio, Massimo}, publisher = {Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik}, address = {Dagstuhl, Germany}, URL = {https://drops.dagstuhl.de/entities/document/10.4230/DagRep.14.12.28}, URN = {urn:nbn:de:0030-drops-230487}, doi = {10.4230/DagRep.14.12.28}, annote = {Keywords: artificial intelligence, conversational agents in games, grounded dialogue and interaction, human-in-the-loop learning} }
Feedback for Dagstuhl Publishing