At the 2026 Inclusion·Bund Conference, held from September 9 to 12 at the Shanghai Huangpu Expo Park under the theme "Co-Creating the New AI Economy," Shen Yujun, Chief Scientist at Ant Lingbo Technology, offered a forward-looking perspective on the evolving dynamics between embodied intelligence and high-level AI frameworks. According to Shen, the rise of embodied intelligence will, in turn, exert a significant influence on the developmental pathways of large language models and cloud-based systems.

Shen highlighted a fundamental distinction in how embodied models operate on robotic platforms compared to conventional conversational AI. Unlike digital systems that rely on a query-response cycle, robots must continuously ingest data from onboard sensors. Even while executing physical tasks, the sensor data stream cannot be interrupted, meaning the inference process is constantly fed with new, real-time information that demands immediate adjustments in decision-making. This persistent, non-stop input requirement presents a core challenge uniquely tied to robot-deployed embodied models.

Given these distinct operational demands, Shen asserted that the tasks handled by embodied models diverge sharply from those managed by high-level cloud services or foundational architectures like Astra. He articulated a clear vision for the future hierarchy: "I believe embodied models will ultimately become tools invoked by higher-level large models. As a tool, it requires its own independent training paradigm tailored to its specific scenarios, which is designed to process continuous sensor input and output action strategies or policies."

Looking ahead, Shen argued that once robots become a routine part of daily life, they will generate vast quantities of physical-world interaction data, which will serve as a novel and rich corpus for large model developers. Leveraging this new data type, models could transcend their current status as cold, screen-bound intelligences. Instead, they would gain the ability to perceive their surroundings like humans, dynamically adapting their dialogue strategies based on environmental shifts. This evolution would move AI beyond passive question-answering, enabling it to proactively offer recommendations and interactions grounded in a richer, context-aware understanding of the physical world.