Researchers that study earthquakes have been trying to model earthquake aftershocks for many years, and now, Google’s new AI can do it for them. Thanks to the new neural network, researchers can fairly precisely predict earthquake aftershocks, as well as scale how strong they would be. Nevertheless, there needs to be much more research done before the algorithm reaches perfection.
Rapid development of artificial intelligence has allowed scientists to use it for various things, like the World Cup predictions or planet exploration. However, Google and a Harvard team collaborated to focus AI toward something that is extremely difficult to be predicted by humans – earthquake aftershocks.
The team trained its newly developed neural network, similar to the one that helps run Facebook photo tagging, as well as Amazon Alexa’s voice transcription, using a database containing more than 131,000 earthquakes and the locations of their respective aftershocks.
Doing so allowed them to determine where the future aftershocks would take place. The network itself is rather interesting, it takes the data, regardless of whether it is pictures of someone’s face or locations of earthquake aftershocks, and the algorithm will attempt to find the underlying pattern. The network uses pixel arrangements of a person’s face in order to attempt facial recognition. In terms of earthquakes, Google’s AI can use that to explain why an aftershock would occur in a certain area.
The findings of the team were published in a paper in the scientific journal Nature on Aug. 29. In the paper, researchers explain that one of the reasons for algorithm accuracy is that they use two complex metrics that were previously thought to be associated with aftershocks. Those are called maximum shear stress change and the von-Mises yield criterion. The two metrics are often used in studying bendable materials like copper or aluminum. However, the metrics are not used in earthquake aftershocks predictions. Given the new discovery, scientists may start using it.
More time to pass
Unfortunately, the neural network can’t start working immediately to predict earthquakes, and will take more time to develop.
“We’re quite far away from having this be useful in any operational sense at all. We view this as a very motivating first step,” Harvard researcher Phoebe DeVries, coauthor of the paper told the BBC.
Nevertheless, even though the predictions are not 100% accurate and need refinement, scientists are pleased with the outcome of Google’s new AI as no one else has come so far when it comes to these predictions. Perhaps more accurate algorithms or other systems are just around the corner.
“Aftershock forecasting in particular is a challenge that’s well-suited to machine learning because there are so many physical phenomena that could influence aftershock behavior and machine learning is extremely good at teasing out those relationships,” DeVries told Science Daily. “I think we’ve really just scratched the surface of what could be done with aftershock forecasting…and that’s really exciting.”
Anyhow, Google’s new AI is a firm step forwards to a better recognition of earthquake aftershocks, and being able to prepare ourselves before they strike, knowing how disastrous and scary they can be.