LogIn
The research shows that the AI successfully interacts with humans 75% of the time.
DeepMind researchers designed AI that can interact naturally with humans using the simplification of a virtual environment and learn by imitating human actions. The multimodal interactive agent (MIA) blends visual perception, language comprehension and production, navigation, and manipulation to engage in extended physical and linguistic interactions with humans. The studies showed that the multimodal interactive agent (MIA) successfully interacts with non-adversarial humans 75% of the time.
The authors of the paper think that it's the foundation for training capable agents for interactive robots or digital assistants.
The researchers created a 3D virtual environment "composed of a randomized set of rooms and a large number of domestic interactable objects, to provide a space and setting for humans and agents to interact together." Humans and AI can interact by controlling virtual robots that locomote, manipulate objects, and communicate via text.
The AI was trained by supervised prediction of human actions (behavioral cloning) and self-supervised learning. To evaluate MIA's performance, human participants had to interact with agents and check whether the agent successfully carried out an instruction. The results showed that MIA achieves over 70% success rate in human-rated online interactions, representing 75% of the success rate that humans themselves achieve when they play as solvers.
"MIA exhibits startlingly rich behavior, including a diversity of behaviors that were not preconceived by researchers, including tidying a room, finding multiple specified objects, and asking clarifying questions when an instruction is ambiguous," concluded the researchers.
Check out the paper here and don't forget to join our new Reddit page, our new Telegram channel, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more.