AI discovers new physics all by itself

In the shadowy world between science fiction and technological reality, artificial intelligence has just taken a leap that's turning heads in research labs worldwide. A neural network has autonomously uncovered a new way to describe physical systems—without being specifically programmed to do so. This development signals not just a milestone in AI capabilities, but potentially a new approach to scientific discovery itself.

Key revelations from this breakthrough

• AI independently discovered physics equations – Researchers at Duke and Harvard created an AI system that, when fed videos of physical phenomena like pendulums or planetary motion, autonomously derived mathematical equations describing these systems with remarkable accuracy.

• The AI uses a "coordinate-free" approach – Unlike traditional physics which relies on arbitrary human-chosen coordinate systems, this AI identifies inherent properties in the systems themselves, creating what scientists call "meaningful latent variables" to describe motion.

• It outperforms traditional methods – In tests against conventional approaches, the AI demonstrated superior predictive power, particularly with complex systems where traditional coordinate-based methods struggle.

Why this matters: AI as scientific collaborator, not just tool

What's particularly striking about this research isn't just that AI can learn physics—it's that it seems to have discovered a fundamentally different way of describing physics than humans have traditionally used. This represents a profound shift in how we might approach scientific discovery.

For centuries, physics has relied on coordinate systems—artificial frameworks we impose on the world to make calculations manageable. Think of the x, y, and z axes we use to plot positions. These coordinates are useful but ultimately arbitrary human inventions. The AI doesn't need these artificial constructs. Instead, it identifies what the researchers call "meaningful latent variables"—properties inherent to the system itself.

This coordinate-free approach has proven remarkably effective. When tested against traditional methods in predicting how physical systems would evolve over time, the AI consistently demonstrated superior accuracy. For complex systems with many interacting parts—the kind that often confound traditional physics—the difference was even more pronounced.

"It's like the AI is seeing the world as it actually is, rather than through the mathematical filters we've been using," explains one of the researchers. This isn't just a computational shortcut—it potentially represents a fundamentally different way of understanding reality.

Beyond the h