An AI can figure out the hidden laws of physics by simply watching a video of a swinging pendulum.
April 29, 2026
Original Paper
Information bottleneck for learning the phase space of dynamics from high-dimensional experimental data
arXiv · 2604.24662
The Takeaway
A new method called DySIB can extract the fundamental coordinates of a physical system directly from raw pixel data. It does not need any human labels or prior knowledge of the system it is watching. For a pendulum, it can identify the exact angle and velocity required to describe the motion just by looking at the footage. This means we can discover the hidden variables of complex natural processes that we do not yet have equations for. This technology could allow scientists to automate the discovery of new physical laws by training AI on video of the deep sea or distant stars.
From the abstract
Identifying the dynamical state variables of a system from high-dimensional observations is a central problem across physical sciences. The challenge is that the state variables are not directly observable and must be inferred from raw high-dimensional data without supervision. Here we introduce DySIB (Dynamical Symmetric Information Bottleneck) as a method to learn low-dimensional representations of time-series data by maximizing predictive mutual information between past and future observation