An AI-driven physicist may have uncovered a new form of physics

The study was published on July 25, 2022, in Nature Computational Science.

Using the power of AI to help improve the speed of physics discoveries

The algorithm was first fed raw video footage of occurrences for which the researchers already had the solution. For instance, they fed a video of a swinging double-pendulum whose two arms' angles and angular velocities were determined to be its exact four "state variables." The AI produced the following result after several hours of analysis: 4.7.

Hod Lipson, director of the Creative Machines Lab in the Department of Mechanical Engineering, said, “we thought this answer was close enough.”

“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,” he added.

After that, the researchers started visualizing the actual variables that the program had discovered. Since the algorithm cannot express the variables in any intuitive language that would be accessible to humans, extracting the variables themselves was challenging. After considerable investigation, it turned out that two of the variables the computer selected matched the angles of the arms, but the other two are still unknown.

Boyuan Chen Ph.D. '22, currently an assistant professor at Duke University, who oversaw the experiment, said, "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 quantities," “but nothing seemed to match perfectly,” he added.

The team was confident that the AI had 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,” he explained.

The researchers fed films of physical systems for which they lacked the explicit solution after testing a number of other physical systems with known solutions. In the first videos, a local "air dancer" could be seen swaying in front of a used car sale. Eight variables were returned by the program after several hours of analysis. Additionally, a Lava lamp video also generated eight suggested variables. The program then outputted twenty-four variables after being fed a video clip of flames from a holiday fireplace loop.

Whether the collection of variables was distinct for every machine or the same set was generated each time the program was launched as a particularly intriguing subject for the team.

"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?” said Lipson. “Perhaps some phenomena seem enigmatically complex because we are trying to understand them using the wrong set of variables.” Every time the AI restarted, the total number of variables remained constant, but the individual variables changed.

So, if the AI's results are accurate, there appear to be other ways to explain the universe, and it's likely that our choices aren't entirely accurate.

We may be missing a trick when it comes to defining variables in physics

According to the researchers, this type of AI can aid in the discovery of complicated events in fields ranging from cosmology to biology, where theoretical knowledge is falling behind the flood of data. “While we used video data in this work, any kind of array data source could be used—radar arrays, or DNA arrays, for example,” explained Kuang Huang, Ph.D. ’22, who co-authored the paper.

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 software, extract freeform physical principles from experimental data. But what if none of the variables are not yet known?

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.

“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,” Lipson explained. Variables describing temperature and pressure needed to be identified before laws of thermodynamics could be formalized, and so on for every corner of the scientific world.

These variables, critically, are a precursor to any theory. “What other laws are we missing simply because we don’t have the variables?” asked Du, who co-led the work.

You can view the entire study for yourself in the journal Nature Computational Science.

The paper was also co-authored by Sunand Raghupathi and Ishaan Chandratreya, who helped collect the data for the experiments. Since July 1, 2022, Boyuan Chen has been an assistant professor at Duke University. The work is part of a joint University of Washington, Columbia, and Harvard NSF AI institute for dynamical systems aimed to accelerate scientific discovery using AI.

Abstract:

"All physical laws are described as mathematical relationships between state variables. These variables give a complete and non-redundant description of the relevant system. However, despite the prevalence of computing power and artificial intelligence, the process of identifying the hidden state variables themselves has resisted automation. Most data-driven methods for modelling physical phenomena still rely on the assumption that the relevant state variables are already known. A longstanding question is whether it is possible to identify state variables from only high-dimensional observational data. Here we propose a principle for determining how many state variables an observed system is likely to have, and what these variables might be. We demonstrate the effectiveness of this approach using video recordings of a variety of physical dynamical systems, ranging from elastic double pendulums to fire flames. Without any prior knowledge of the underlying physics, our algorithm discovers the intrinsic dimension of the observed dynamics and identifies candidate sets of state variables."