With its extensive equations and intricate measurements, physics is one of science’s most demanding and rigid sciences. Its secrets can only be revealed if everything is done just right. A system’s variables, a key precursor to written equations, must be discovered first before even the simplest equation can be put together.
F=MA is Newton’s great fundamental equation of force. Newton had to first grasp the ideas of acceleration, mass, and force in order to construct such an equation. Professor of engineering and data science Hod Lipson tells Motherboard that this is a challenge with no clear path to pursue.
Lipson describes it as “an art; there is no methodical way.” There is a resemblance to how one discovers the alphabet. There is nothing forced about it; it just happens naturally.” With the help of machine learning, researchers at Lipson’s Creative Machines Lab are trying to better understand how this process of discovery occurs and how it may be improved upon.
A machine learning algorithm developed by Lipson and his colleagues is capable of analyzing physical phenomena by “viewing” films, such as the swing of a double pendulum or the flicker of a flame, and producing the number of variables required to describe the motion. A single pendulum can be described by 2.05 different variables instead of 2, which is the correct number predicted by the method for known systems, and it can even predict variables for systems it isn’t aware of. Published in Nature Computational Science, a paper titled “Automated discovery of fundamental variables buried in experimental data” revealed the findings just last week.
Although this algorithm is not the first to analyse data and attempt to derive a physical relationship from it, Lipson believes that this work stands out because it is the first to not provide the algorithm with any information on the number or kind of predicted variables in a system. Lipson. Since the system is not limited to looking for variables through merely a human lens, Lipson claims that this could be key to finding hidden physics in these systems.
In order to speed up the process, Lipson argues, “It’s not that people are toiling away day and night to seek for these factors and this can expedite it.” According to him, “We’re probably overlooking a lot of things,” he continues. As a result, the researchers reasoned that applying artificial intelligence to the problem might provide results that are both extremely beneficial and fundamentally altering the way we think.
The algorithm was fed movies of dynamic motion in a variety of complexities by Lipson and colleagues, including the paper’s original author, Boyuan Chen, who is currently an assistant professor of engineering at Duke University. Both well-known and yet mysterious motions, such as those produced by lava lamps, flickering fires, and inflatable air dancers, were included.
After watching these movies, the AI tried to predict the future and generate a list of smaller and smaller variables that were responsible for the behavior. Finally, the AI would output the bare minimum of variables required by the system to accurately capture motion.
While the AI was able to identify the proper amount of variables, there is a major snag that will prevent it from being used in scientific research. There are currently no words to define what the variables in a system are; for instance, it returned eight variables for the “air dancer” and 24 variables for the fireplace. As AI systems become increasingly complicated and opaque, it becomes increasingly difficult for researchers to decipher how they make any particular choice.
For the time being, Chen isn’t too concerned about this. “We have a general framework at the moment,” Chen explains. There’s a lot of value in working with professionals who have data and an understanding of what that data is doing. Help people uncover what they don’t know about data” is our goal.
According to Lipson, in the future, this could look like studying disease evolution or climate change. They anticipate that the patterns generated by the computer will make it easier for human colleagues to understand the findings in the future. He believes this will be the next big scientific breakthrough. According to Lipson, “It looks to me as if we’ve come to a point where we’ve exhausted our ability to accomplish this manually.” “We need a boost in order to move forward.”