Get any physics book and you’ll find formula after formula that describes how things swing, fly, swerve and stop. The formulas describe the actions that we can observe, but may be behind each of them combinations of factors that are not immediately obvious.
Now, a new artificial intelligence program developed by researchers at Columbia University appears to have discovered its alternative physics.
After showing videos of physical phenomena on Earth, AI hasn’t rediscovered the current variables we’re using; Instead, he actually came up with new variants to explain what he saw.
To be clear, this does not mean that our current physics is flawed or that there is a better model to explain the world around us. (Einstein’s laws proved incredibly powerful.) But these laws can only exist because they are built on the back of a theoretical “language” and pre-existing principles established by centuries of tradition.
Given an alternate timeline where other minds approached the same problems with a slightly different perspective, would we still put the mechanisms that explain our universe in the same way?
Even with new technology for imaging black holes and discovering alien worlds far away, these laws have held up time and time again (side note: quantum mechanics is another story entirely, but let’s stick to the visible world here).
This new AI has only looked at videos of a few physical phenomena, so it is by no means appropriate to invent new physics to explain the universe or try to improve Einstein. That was not the point here.
“I’ve always wondered, if we ever met an intelligent alien race, would they discover the same laws of physics that we have, or would they describe the universe differently?” says roboticist Hood Lipson of Columbia’s Creative Machines Laboratory.
“In the experiments, the number of variables was the same each time the AI restarted, but the specific variables were different each time. So yes, there are alternative ways to describe the universe and it is entirely possible that our choices are not perfect.”
Beyond that, the team wanted to see if AI could actually find new variables—and thus help us explain new, complex phenomena emerging in our current deluge of data that we don’t currently have the theoretical understanding to keep up with.
For example, new data emerging from giant experiments like the Large Hadron Collider suggest new physics.
“What other laws are we missing simply because we don’t have the variables?” says mathematician Qiang Du of Columbia University.
So how does AI find new physics? First of all, the team fed raw video footage to the system of phenomena they already understood and asked the program a simple question: What are the minimum essential variables needed to describe what’s going on?
The first video showed a swinging double pendulum known to have four state variables in action: the angle and the angular velocity of each of the pendulums.
AI thought about the footage and the question for a few hours and then spat out an answer: This phenomenon requires 4.7 variables to explain, she said.
That’s close enough to the four elements we know… but it still doesn’t explain what AI thinks about the variables.
So the team then tried to match the known variables with the ones chosen by the AI. Two of them fit loosely at the angles of the arms, but the other two variants have remained a mystery. However, the AI can make accurate predictions about what the system will do next, so the team thought the AI must be into something they couldn’t fully understand.
“We tried to relate the other variables to anything and everything we could think of: angular and linear velocities, kinetic and potential energy, and various combinations of known quantities,” says software researcher Boyuan Chen, now an assistant professor at Duke University, who led the work.
“But nothing seems to be quite a match… We don’t yet understand the mathematical language he speaks.”
The team then continued to show other AI videos. The first was a wavy “air dancer” arm blowing in the wind (AI said this had eight variables). The lava lamp footage also produced eight variants. A video of the flame appeared with 24 variants.
Each time, the variables were unique.
“Without any prior knowledge of fundamental physics, our algorithm detects the intrinsic dimension of the observed dynamics and identifies candidate combinations of state variables,” the researchers wrote in their paper.
This suggests that in the future, AI will likely help us identify variables that support new concepts that we are not currently aware of. See this space.
The search was published in Computational natural sciences.