Researchers from Archetype AI have developed an artificial intelligence foundational model called the "Newton AI Model," which can accurately predict various physical phenomena, including those never encountered during training, by analyzing sensor data alone. This groundbreaking research could revolutionize the way physical research is conducted, opening a new chapter for scientific discovery.
Traditionally, building AI models for physical phenomena required inputting a large amount of physical laws and specialized knowledge as prior information, which limited the model's application scope and made it difficult to generalize to other fields. In contrast, the "Newton AI Model" adopts a novel "phenomenological" approach, completely independent of any physical laws or prior knowledge, instead learning and understanding the workings of the physical world through the analysis of vast amounts of sensor data.
The researchers trained the model using 590 million sensor data samples from 41 public datasets, covering phenomena such as electric current, fluid flow, and optics. The trained "Newton AI Model" can encode and predict various physical behaviors, including mechanical motion, thermodynamics, and even complex non-analytical physical processes like urban-scale meteorological changes.
To validate the model's generalization capability, the researchers conducted a series of experiments, including simulating mechanical oscillations with a spring-mass system and using a thermoelectric device to mimic thermodynamic phenomena. The results showed that the "Newton AI Model" could accurately predict the future behavior of these physical systems, with prediction accuracy even surpassing models specifically trained for particular physical systems.
The emergence of the "Newton AI Model" brings new possibilities to physical research. It can help scientists analyze experimental data more quickly and accurately, discover new physical laws, and even predict and control complex physical systems. Additionally, the model possesses "zero-shot inference" capabilities, meaning it can predict phenomena it has never encountered before, opening new frontiers for scientific discovery.
The researchers stated that the "Newton AI Model" is just the beginning, and they plan to further expand the model's training dataset and explore its applications in other fields. This research brings hope for building a unified AI foundational model to understand and predict various physical world processes.