NEW YORK — Can artificial intelligence discover a new kind of physics? A team of roboticists at Columbia Engineering decided to put this question to the test. They developed an AI program that detected physical phenomena and discovered relevant variables, which are a necessary precursor to any physics theory.
To conduct the study, researchers gave the AI program raw video footage of phenomena for which they already knew the answer. The video showed a swinging double-pendulum which has four “state variables” — the angle and angular velocity of each of the two arms. The AI system gave a nearly correct answer of 4.7 variables after several hours of analysis.
“We thought this answer was close enough,” says Hod Lipson, director of the Creative Machines Lab in Columbia’s Department of Mechanical Engineering, in a university release. “Especially since all the AI had access to was raw video footage, without any knowledge of physics or geometry. But we wanted to know what the variables actually were, not just their number.”
Researchers report that two of the variables the AI program chose roughly corresponded to the angles of the arms, but the other two remain 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 various combinations of known qualities,” notes Dr. Boyuan Chen, an assistant professor at Duke University. “But nothing seemed to match perfectly.”
Dr. Chen believes the program found a valid set of four variables, since it was making good predictions, “but we don’t yet understand the mathematical language it is speaking.”
The AI system discovered new ways to describe the universe
Researchers began feeding the computer program videos of systems which scientists don’t know the explicit variable answer. The first video showed an “air dancer” rising and falling in front of a used car lot. The AI program returned eight variables from that video. The program gave the same number of variables after being fed a video of a lava lamp. The program then returned 24 variables after analyzing a video clip of flames from a fireplace loop.
The study found that the number of variables were the same each time the AI program restarted. However, the specific variables were different. This proves that there are alternative ways to describe the universe.
“I always wondered, 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?” notes Lipson. “Perhaps some phenomena seem enigmatically complex because we are trying to understand them using the wrong set of variables.”
Researchers believe this AI system can help scientists uncover complex phenomena in areas ranging from biology to cosmology.
“While we used video data in this work, any kind of array data source could be used — radar arrays, or DNA arrays, for example,” explains study co-author Dr. Kuang Huang.
Scientists still don’t have all the answers
Since scientists might not have a decent set of variables to describe the phenomena, Lipson says, scientists may be misinterpreting or failing to understand many phenomena.
“For millennia, people knew about objects moving quickly or slowly, but it was only when the notion of velocity and acceleration was formally quantified that Newton could discover his famous law of motion F=MA,” says Lipson, proving that variables are a precursor to any theory.
The study is published in the journal Nature Computational Science.