AI discovers its own ‘fundamental’ physics and scientists are baffled

AI invents its own 'fundamental' physics and scientists are baffled

Image: MR.Cole_Photographer via Getty Images


ABSTRACT breaks down mind-blowing scientific research, future technology, new discoveries and major breakthroughs.

Physics is one of the more rigorous and rigid disciplines of science, riddled with long equations and complex measurements to be made only right to reveal their secrets. But before even the simplest equation was constructed, scientists first had to puzzle out a crucial precursor to written equations: the variables of a system.

Take Newton’s great fundamental force equation: F=MA. Before such an equation could be constructed, Newton first had to understand the concepts of acceleration, mass, and force. This is a task that does not have a well-defined path to follow, Columbia University professor of engineering and data science Hod Lipson told Motherboard.

“It’s an art, there’s no systematic way,” Lipson says. “It’s almost like, how do you discover the alphabet? It just happens organically.”

At Lipson’s Creative Machines Labhe and colleagues want to better understand how this discovery process takes place and how it can be improved by using machine learning to uncover hidden, alternative physics that human scientists may have missed.

To do this, Lipson and colleagues designed a machine learning algorithm capable of studying physical phenomena by “watching” videos, such as the swing of a double pendulum or the flickering of a flame, and counting the number of variables. produce necessary to explain the action. For known systems, the algorithm was able to predict the correct number of variables within 1 value (e.g. 2.05 variables to describe a single pendulum instead of 2) and even make variable predictions for unknown systems. The findings were published last week in a study titled “Automated discovery of fundamental variables hidden in experimental data” in the news Nature Computational Science.

While this algorithm isn’t the first to study data and try to extract a physical relationship from it, Lipson says this work stands out because it’s the first that the algorithm doesn’t provide information about the number or type of expected variables in a system. As a result, the system is not limited to searching for variables through a human lens alone, which Lipson says could be crucial for uncovering hidden physics within these systems.

“It’s not like people toil day and night looking for these variables and this can speed up the process,” Lipson explains.

“It’s more that we’re probably overlooking a lot of things,” he continues. “But so much depends on those variables that we thought if we could throw some AI power on this, maybe we’d discover things that are super useful and will change the way we think.”

To prime their algorithm for success, Lipson and colleagues, including the paper’s first author and now assistant professor of engineering at Duke University, Boyuan Chen, it fed videos of dynamic movement in various complexities. This included well-known movements such as double pendulums and swing sticks, as well as not yet understood movements such as lava lamps, flickering fires or inflatable air dancers.

After studying these videos, the AI ​​tried to model the phenomena a few steps into the future and list smaller and smaller variables responsible for the action. Finally, the AI ​​would spit out the minimum number of variables the system needs to accurately capture the movement.

While the AI ​​has been quite successful in discovering the right number of variables, there is one big catch that will keep it from entering science labs anytime soon. It can tell scientists that there are a certain number of variables in a system, but it currently lacks language to describe what those variables are — for example, it returned eight variables for the “air dancer” and 24 for the fireplace. Explainability is a long-standing research goal for AI systems, which complex black boxes that make it difficult for scientists to reverse engineer a specific decision.

This is something Chen isn’t too concerned about for the time being.

“What we have now is like a general framework,” Chen says. “One thing that will be really interesting is working with experts who have data and an intuition about what that data does. What we want to do is help them discover what they don’t know about the data yet.”

In the future, this could be like studying systems beyond physics, such as the evolution of disease or climate change, Lipson says. Later, they hope that patterns emerging from the algorithm will make the findings easier to communicate with human collaborators. According to Lipson, this will be the next big step forward in scientific discovery.

“People have been doing this for 300 years and it seems we’ve reached the end of what we can do manually,” Lipson says. “We need something to take us to the next level.”

Leave a Comment

Your email address will not be published.