Alternate Physics Has Just Been Discovered by an AI

Alternate Physics Has Just Been Discovered by an AI ...

You'll find formula after formula describing how things move, fly, swerve, and stop. The formulas describe actions we can observe, but behind each could be a collection of unknowns.

Now, a new AI program developed by Columbia University students has appeared to have discovered its own alternative physics.

The AI didn't discover the current ones we're using; instead, it invented new ones to explain what it saw.

To be clear, this doesn't mean that current physics is flawed, or that there is a better fit model to explain the world around us. (Einstein's laws have proved extremely robust.)But those laws could only exist because they were established through centuries of tradition.

Would we still frame the mechanics that explain our Universe in the same way if we had an alternate timeline in which other individuals approached the same problems from a slightly different viewpoint?

Even with advanced equipment for measuring black holes and detecting strange, distant worlds, these laws have remained constant (side note: quantum mechanics is a whole other story, but let's stick to the visible world here).

This new AI just examined videos of a handful of physical phenomena, so it is in no position to propose new physics to explain the Universe or to surpass Einstein. This was not the intention here.

"I always wondered, if we ever encountered 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?" says roboticist Hod Lipson from Columbia's Creative Machines Lab.

"The number of variables was the same each time the AI restarted, but the specific variables were different each time."

Beyond that, the team wanted to know whether AI might be able to identify new variables and therefore help us explain complex new phenomena emerging in our current data overload that we don't currently have the theoretical knowledge to keep up with.

New data about new physics are emerging from massive experiments such as the Large Hadron Collider.

"What other laws are we missing simply because we don't have the variables," says Columbia University mathematician Qiang Du.

So how does an AI discover new physics? To start, the team fed the system raw video footage of phenomena they already understood, and asked the program a simple question: What are the minimum fundamental variables needed to describe what's going on?

The first video depicted a swinging double pendulum that is thought to have four state variables in play: the angle and angle of the two pendulums.

The AI mulled over the footage and the question for a few hours before spouting a reply: This phenomenon would require 4.7 variables to explain it, according to the company.

It's close enough to the four we know of, but it didn't explain what the AI thought the variables were.

The AI could then manipulate known variables to the ones it had chosen. Two of them were fairly similar to the angles of the arms, but the other two variables remained unsolved, so the team concluded the AI must have discovered something.

"We tried to correlate the other variables with whatever and everything we could think of, including angular and linear velocity, kinetic and potential energy, and various combinations of known quantities," says software researcher Boyuan Chen, who is now an assistant professor at Duke University.

"But nothing appeared to match perfectly, we don't yet understand the mathematical language it is speaking."

The AI said the first featured a wavy arm 'air dancer' blowing in the wind (the AI said this had eight variables). A video clip of flames returned with 24 variables.

The variables were unique every time.

"Our algorithm discovers the intrinsic dimension of the observed dynamics without any prior knowledge of the underlying physics," the researchers conclude in their paper.

This suggests that artificial intelligence (AI) might be useful in the future.

The work has been published in Nature Computational Science.

You may also like: