Columbia University's AI program discovers alternative physics
Ever wondered how artificial intelligence can be used to discern the mysteries of physics? Well an AI belonging to Columbia University was made to observe physical phenomena on Earth and search for the fundamental variables that describe them. The program discovered new variables to explain what it saw instead of the ones currently used. This will have massive implications for scientific discovery.
Why does this story matter?
Scientists of the yesteryears did not have the luxury of relying on computer systems for making scientific discoveries. However, times have changed. The rise of AI has begun. By making complex calculations in the blink of an eye, they might find new ways to describe how the universe works. They will become more powerful in the future and accelerate the pace of scientific growth.
How were the tests conducted?
The AI observed raw video footage of physical phenomena using a video camera. Then the scientists questioned the AI about the minimum fundamental variables necessary to explain them. When a swinging double-pendulum with four known variables—the angle and angular velocity of each of the two arms—was shown, it gave the answer as '4.7.' The researchers failed to decipher its mathematical language.
What was its response to other phenomena?
The AI also saw other videos. They include an 'air dancer' blowing in the wind (it had eight variables according to the AI), a holiday fireplace loop with flames had 24 variables, and a clip of a lava lamp footage produced eight variables. Each time the AI restarted, the number of variables was the same. However, the specific variables were different each time.
What are its future applications?
As per researchers involved with this project, AI similar to the one at Columbia University will aid scientists in understanding complex topics ranging from biology to cosmology, where theoretical understanding is not keeping pace with the available data. Also, to do the work the AI can be fed not only videos but any array-based data source, including DNA and radar arrays.
