Trend Technology 9: AI that learns physical dynamics from multimodal sensory data

Summary

World models are the report's ninth emerging technology, AI systems that learn how the physical world behaves rather than how people describe it. The report contrasts a baby who knows objects fall before learning the word gravity with large language models that learn from text and quickly show gaps on unfamiliar physical problems. World models take in several sensory streams at once, such as video, depth sensors, pressure readings and motion capture, and compress them into one shared representation. Different recordings of the same event, whether footage, sensor data or a written account, land in the same place in that space. The approach became feasible through larger models, a decade of sensor data from robotics and vehicles, and an architecture that predicts compressed representations. The outlook expects early impact in manufacturing and logistics, with workforce shifts in fields built on physical judgement.