Researchers at Nanjing University have developed a method that improves how humans and artificial intelligence (AI) work together. Their new benchmark and training framework enable AI to collaborate effectively with humans, even in unexpected situations where both have different skills and ways of understanding the world—a common challenge in everyday life.
New AI Playbook: Speaking in Multiple “Languages”
Imagine a basketball team where every player speaks a different language and brings unique strengths. Traditional AI systems are like a teammate who only knows one playbook and struggles to communicate with others on the court. With this new approach, AI learns to understand and use several “languages,” allowing it to share important information with human teammates and adjust strategies quickly and smoothly—much like an experienced player who adapts mid-game.
Empowering Industries with Enhanced Human-AI Collaboration
This advancement is significant as AI becomes an increasingly common part of our daily lives, workplaces, and industries. Enhanced human-AI collaboration can boost productivity and improve safety in key areas such as automated manufacturing, emergency rescue operations, and healthcare assistance. Policymakers, businesses, and technology developers can use these insights to set higher standards for human and AI interaction.
Real-World Testing and Advanced Training Approach
To evaluate their new approach, the researchers introduced the Open and Real-world Coordination Benchmark (ORCBench). This testing environment realistically measures how well AI can work with human partners with a wide range of capabilities. They also developed a novel training method called HeteC (Heterogeneous Training with Communication), significantly improving AI’s ability to communicate and coordinate with people.
AI Outperforms Traditional Methods in Teamwork
Key findings from the study indicate that AI agents trained with this new method outperformed those using older approaches (such as SP, PP, FCP, MEP, and MAZE) in tasks that require effective teamwork between humans and AI. The research further revealed that clear and effective communication helped the AI adapt and respond much better to unpredictable, real-world situations. In addition, the AI showed strong adaptability to new and unforeseen challenges, teaming up successfully with new human partners. These results highlight the robustness and versatility of the innovative approach.
Training Together: Breaking Down Communication Barriers
The researchers designed their method by training AI agents alongside simulated human partners in various conditions. By integrating a specialized communication module, the AI learned to share critical information in a clear and timely manner, overcoming previous barriers to effective teamwork.
Ultimately, this research paves the way for more practical, effective, and dynamic human-AI teamwork. It sets the stage for future innovations across multiple sectors, promising a new era of collaboration that could transform industries and everyday interactions. Reflecting on the achievement, lead researcher Prof. Yang Yu stated, “Our work bridges the communication gap between humans and AI, leading to a future where technology and human ingenuity combine to tackle real-world challenges like never before.” The complete study is accessible via DOI: 10.1007/s11704-024-3797-6.
MEDIA CONTACT
Register for reporter access to contact detailsArticle Multimedia

Credit: Cong GUAN
Caption: Illustration of the Overcooked game

Credit: Cong GUAN
Caption: Illustration of different layouts on Overcooked

Credit: Cong GUAN
Caption: Illustration of the emergency rescue environment

Credit: Cong GUAN
Caption: Illustration of the high-level workflow of HeteC

Credit: Cong GUAN
Caption: Illustration of the policy network and communication module

Credit: Cong GUAN
Caption: Illustration of the HeteC on FC layout of Overcooked
CITATIONS