Artificial Intelligence Discovers Alternative Physics

Energy, mass, and speed The famous equation E=MC2 of Albert Einstein has these three elements. However, how did Albert Einstein come to hear about these ideas at all? Comprehending relevant variables is a prerequisite to understanding physics. Without the notions of energy, mass, and velocity, not even Einstein could have discovered relativity. But can these kinds of variables be found automatically? This would speed up scientific research significantly.

Researchers at Columbia Engineering asked a brand-new artificial intelligence software this inquiry. The AI software was created to use a video camera to examine physical occurrences before attempting to identify the smallest possible collection of fundamental variables that may adequately capture the dynamics being witnessed. On July 25, the work was released in the journal Nature Computational Science.

A chaotic swing stick dynamical system is seen moving in the image. Our approach seeks to directly extract from high-dimensional video footage the minimal set of state variables required to directly characterize such a system. Credit: Columbia Engineering/Yinuo Qin

The scientists first fed the system unprocessed video of physics processes for which they already had an answer. For instance, they provided a video of a swinging double-pendulum whose two arms' angles and angular velocities were determined to be its precise four "state variables." After several hours of investigation, the AI produced the following result: 4.7.

Hod Lipson, head of the Creative Machines Lab at the Department of Mechanical Engineering, where the study was principally conducted, stated, "We decided this response was near enough. Particularly considering that the AI was only given access to raw video material and had any understanding of physics or geometry. Instead of merely counting the variables, we wanted to know what they were.

The researchers then went on to visually represent the real variables that the computer had detected. It was challenging to extract the variables since the software is unable to explain them in a form that is logical and clear to people. Following some research, it turned out that two of the variables the software selected about matched the angles of the arms, but the other two are still unknow.

Boyuan Chen PhD '22, currently an associate professor at Duke University, who oversaw the experiment, said: "We tried associating the other variables with anything and everything we could think of: angular and linear velocities, kinetic and potential energy, and other combinations of known values. But nothing didn't seem to fit exactly. Since the AI was producing accurate predictions, the team felt sure that it had discovered a reliable set of four variables, but, as he noted, "we don't yet comprehend the mathematical language it is speaking.

Boyuan Chen describes how an innovative AI system observed physical processes and discovered pertinent variables, which is a crucial step before developing any physics theory. Credit: Columbia Engineering/Boyuan Chen

The scientists entered films of physical systems for which they lacked the explicit solution after evaluating a number of other physical systems with known solutions. In one of these movies, a local used car lot was the backdrop for a "air dancer" who was swaying. The software yielded 8 variables after many hours of investigation. Similarly, a video of a lava lamp yielded 8 8 variables. The algorithm retrieved 24 variables after they gave it a video clip of Christmas fireplace flames.

Whether the collection of variables was the same for every system or if a separate set was generated each time the program was launched was a particularly intriguing subject. If humanity ever encountered an intelligent extraterrestrial culture, Lipson stated, "I often wondered if they would have found the same physical principles that we have or if they may have described the cosmos in a different way." "Perhaps some things look enigmatically difficult because we are utilizing the wrong set of variables to analyze them,"

Every time the AI restarted in the studies, the total number of variables remained constant, but the individual variables changed. So, certainly, there are other ways to explain the cosmos, and it's likely that our choices aren't entirely accurate.

The researchers claim that this kind of AI can assist researchers in revealing complicated phenomena in fields ranging from cosmology to biology where theoretical understanding is not keeping up with the flood of data. Although we employed video data in this work, any form of array data source might be used, such as radar or DNA arrays, according to coauthor Kuang Huang PhD '22.

Qiang Du, a professor of mathematics at Lipson and Fu Foundation, has been interested in developing algorithms that can transform data into scientific rules for many years. Only if the variables were known in advance could earlier software systems, like Lipson and Michael Schmidt's Eureqa program, extract freeform physical principles from experimental data. But what if none of the factors are known yet?

Lipson, who is also the James and Sally Scapa Professor of Innovation, contends that a lack of an adequate collection of variables to define many occurrences may be causing scientists to misinterpret or underappreciate them. "People have known about objects moving swiftly or slowly for millennia, but Newton could only uncover his famous rule of motion F=MA when the idea of velocity and acceleration was explicitly defined," said Lipson. Before rules of thermodynamics could be defined, variables characterizing temperature and pressure had to be found. This process repeated itself for every field of science. A theory's variables come before it in some way. Du, who co-led the effort, questioned, "What other laws are we missing just because we don't have the variables?"

By COLUMBIA UNIVERSITY SCHOOL OF ENGINEERING AND APPLIED SCIENCE