AI subverts physics research! Without the laws of physics, this model predicts physical phenomena on its own

Researchers from Archetype AI have developed a basic artificial intelligence model called the "Newton AI Model" that can accurately predict various physical phenomena, even those that have never been detected during training, simply by analyzing sensor data. contact with the phenomenon. This breakthrough research result may completely change the way physics research is conducted and open a new chapter in scientific discovery.

Traditionally, building AI models for physical phenomena requires inputting a large amount of physical laws and professional knowledge into the model as prior information, which limits the application scope of the model and makes it difficult to generalize to other fields. The "Newton AI Model" adopts a brand-new "phenomenology" method, which does not rely on any physical laws or prior knowledge. Instead, it learns and understands the operating rules of the physical world by analyzing massive sensor data.

The researchers trained the model using 590 million sensor data samples from 41 public datasets, covering a variety of physical 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, etc., and can even predict complex non-analytical physical processes such as city-scale meteorological changes.

To verify the generalization ability of the model, the researchers conducted a series of experiments, including simulating mechanical oscillations using a spring-mass system and simulating thermodynamic phenomena using a thermoelectric power generation device. Experimental results show that the "Newton AI model" can accurately predict the future behavior of these physical systems, and its prediction accuracy even exceeds that of models specifically trained for specific physical systems.

The emergence of "Newton AI model" brings new possibilities to physics research. It can help scientists analyze experimental data faster and more accurately, discover new physical laws, and can even be used to predict and control complex physical systems. In addition, the model also has "zero-sample inference" capabilities, which means it can predict physical phenomena that have never been touched before, which opens up new areas for scientific discovery.

The researchers said that the "Newton AI Model" is just the beginning. In the future, they will further expand the model's training data set and explore its applications in other fields. This research result brings hope for building a unified AI basic model to understand and predict various physical world processes.

Paper: https://cdn.prod.website-files.com/669fb9b0365257a2d64b9744/671062d53917e78989931495_Phenomenological%20AI%20Foundation%20Model%202024.pdf