NeurIPS 2026 launches inaugural workshop on real-time conversational AI

The Real-Time Conversational Agents (RTCA) Workshop is being launched as the inaugural offering at NeurIPS 2026, the premier annual machine learning conference, which will take place in Sydney, Australia on 11 or 12 December 2026. The Call for Papers and Demos is now open, with submissions welcome through the workshop website at https://rtcaneurips26.github.io/.

The workshop is centered on real-time multimodal conversational agents—systems that must handle streaming speech and video input while generating language output in real time. The focus areas include naturalness in interaction and evaluation methodologies for live systems. This represents a recognition that conversational AI has evolved significantly beyond text-based chat applications. Modern conversational systems now operate as voice-enabled agents that talk back, embodied avatars, and intelligent assistants that can see and interact with shared screens and tools.

What distinguishes this workshop's focus is the acknowledgment that real-time conversational interaction is fundamentally harder than offline generation. Systems must now operate with low latency while continuously listening and watching for new information, and they must handle conversational phenomena that offline systems can largely ignore: managing turn-taking between speakers, responding to backchannels (brief acknowledgments like "uh-huh"), handling interruptions, and maintaining alignment across multiple modalities. These become first-class design problems rather than secondary concerns.

Conversational AI has undergone a significant evolution from text-only interfaces to multimodal systems that operate in real time. The launch of the RTCA Workshop at NeurIPS 2026 reflects this shift: conversational agents now include voice modes, video feeds, embodied avatars, and screen-sharing tools. Unlike offline text generation, where systems can take time to compute responses, real-time conversational systems face fundamentally different challenges—they must stream outputs while simultaneously listening and watching, manage conversational turn-taking and interruptions, and maintain alignment across multiple modalities. The workshop's focus on these problems suggests the community recognizes that the paradigm for building natural conversational systems has changed. Full-duplex interaction (where agents can speak and listen at the same time, as humans do) and the ability to handle backchannels and unexpected interruptions are now first-class design concerns rather than edge cases. This represents a practical maturation in how conversational AI is being deployed in real-world applications.

No comments yet. Be the first to share your thoughts!