Roboticists Developed an AI Program That May Have Discovered an 'Alternative Physics'

Joaquin Victor Tacla, Tech Times 28 July 2022, 05:07 am

Artificial Intelligence has ushered the advancement of several disciplines throughout the years. But could it ever discover a new form of physics?

A group of roboticists from Columbia University wanted to exploit the vast potential of AI and find out if it can ever find an "alternative physics."

Hence, they created an AI tool that could recognize physical occurrences and identify pertinent variables, essential building blocks for every physics theory.

Feeding The AI

As reported first by SciTechDaily, the AI algorithm was fed with raw video footage of events for which the researchers already knew the solution.

The angle and angular velocity of each of the two arms of the swinging double-pendulum, which has four "state variables," were displayed in the video. After several hours of investigation, the AI system provided a virtually accurate response of 4.7 variables.

According to Hod Lipson, head of the Creative Machines Lab in the Department of Mechanical Engineering, where the work was principally conducted, the team believed that the answer provided by the AI was close enough.

It is worth noting that the AI program had no prior knowledge about physics or geometry and all it could do was access the raw video footage. Hence, Lipson noted that the team wanted to investigate the variables on why the program came up with its answer.

The researchers then went on to visually represent the real variables that the computer had detected. It was challenging to extract the variables since the software was unable to explain them in a form that was logical and understandable to people.

After some research, it became apparent that two of the program's variables roughly matched the angles of the arms, but the other two are still unknown.

Correlating The Variables

According to Boyuan Chen, an assistant professor at Duke University, the roboticists tried correlating the other variables with "anything and everything they could think of", including angular and linear velocities, kinetic and potential energy, and various combinations of known quantities.

But in the end, these correlations fell short. The team is certain that the AI had identified a reliable set of four variables because it produced accurate predictions, even though they don't understand the mathematical language it is articulating.

The scientists entered videos of physical systems for which they lacked the explicit solution after verifying several physical systems with known solutions.

In one of these videos, a local used car lot was the backdrop for an "air dancer" who was swaying. The program yielded eight variables after several hours of analysis. Similarly, a video of a lava lamp yielded 8.8 variables.

Additionally, the algorithm retrieved 24 variables after they gave it a video clip of holiday fireplace flames.

Alternative Ways To Explain The Universe

Every time the AI restarted, the total number of variables remained constant, but the individual variables changed. According to SciTechDaily, this could indicate alternative ways to explain the universe, and that our current models are not entirely accurate.

The researchers claim that this kind of AI can assist researchers in revealing complicated phenomena in fields ranging from cosmology to biology, where theoretical understanding is not keeping up with the flood of data.

Although the team used video data in the study, any array data source can also be employed, such as radar arrays or DNA arrays, according to Kuang Huang, coauthor of the paper.

This article is owned by Tech Times

Written by Joaquin Victor Tacla