Artificial Intelligence Finds Over 6000 Lunar Craters Undiscovered By Humans
And like most old fashioned things, we've eventually found a better way to do it, although it was only just figured out. A group of researchers from the University of Toronto and Penn State University developed an artificial intelligence to more accurately map lunar craters, and it worked quickly: within several hours, it had already found about 6000 new ones that no human had ever noticed.
A team of researchers developed an #AI system with @NVIDIA GPUs that can automatically count craters on the moon which will help reveal undiscovered clues about the formation of our solar system. https://t.co/k1NiFUnHqy pic.twitter.com/15o8QgbDr7— NVIDIA AI Developer (@NVIDIAAIDev) March 15, 2018
This is precisely what the researchers were hoping for, and it moved much faster and gave them a much bigger number than they expected. The AI was developed from a convolutional neural network, and was first taught to identify craters by receiving data on two thirds of the Moon's surface.
It was then shown the final third to see if it worked, and that's when the AI eventually responded with a much higher number - a few were identified as false positives, but they made up a small fraction of the 6000 new entries. Ari Silburt, a postdoc at Penn State University who helped create the neural network, said the following in a press release:
Since the Moon has no wind or weather to speak of like Earth does, a crater tends to stay put after an asteroid collision, and the Moon has faced up a large amount of collisions over the 4.5 billion years. It's why this AI can similarly be effective in mapping Mercury, which also has no wind and thus no erosion.
And since some of these craters being discovered are over 4 billion years old, there is a lot we can learn from each one we track down. So in this case, automating the crater-counting process has its advantages.