Researchers from the Columbia School of Engineering tested whether artificial intelligence can independently determine the variables needed to describe a physical phenomenon. For example, in Einstein’s most famous formula E=MC2, there are three such variables: energy, mass, and speed. Determining the relevant variables is the first step to understanding physics, and this is the task put in front of AI.

The program had to observe physical phenomena through a video camera and search for a minimal set of fundamental variables that would fully describe the observed dynamics.

The experiment began with the observation of already known phenomena. The AI was ​​shown a recording of a double pendulum that oscillates and has 4 state variables: the angle and the angular velocity of each of the two parts. The AI ​​gave an answer close to the one known to scientists – 4.7.

The researchers then tried to decipher which variables the artificial intelligence had identified. This turned out to be difficult, since the program cannot describe them in an intuitively understandable way for people. It turned out that the two variables were close to those used by physicists. However, the other two remained a mystery.

“We tried correlating the other variables with anything and everything we could think of: angular and linear velocities, kinetic and potential energy and and various combinations of known quantities. But nothing seemed to match perfectly,” says Dr. Boyuan Chen, who led the project.

Scientists are confident that the artificial intelligence has correctly identified all the variables, but they do not yet understand the “mathematical language it speaks”.

In subsequent experiments, the AI ​​showed videos of phenomena for which the researchers did not know the exact number of variables. For the lava lamp and the “air dancer” – an inflatable figure in the shape of a person that swayed near the parking lot – the program defined 8 variables each. For the flame from the fireplace, there were 24 variables. A particularly interesting question was, does the program invent new variables each time, or are some of them repeated?

“If we ever met an intelligent alien race, would they have discovered the same physics laws as we have, or might they describe the universe in a different way? Perhaps some phenomena seem enigmatically complex because we are trying to understand them using the wrong set of variables,” says Hod Lipson, professor of engineering and data science.

AI proves that there are alternative ways to describe the universe. Researchers believe that this type of artificial intelligence can help explain those phenomena that still remain unclear. Although video was used in the experiments, any data set can be chosen for this purpose. For example, radar or DNA.

Lipson claims that some physical phenomena are not understood or misunderstood by people because they do not have an appropriate set of variables to describe them. So you can expect help in this direction from artificial intelligence.