STATESBORO, Ga. () — Georgia Southern researchers are working on what they said could be a major breakthrough, predicting earthquakes with 98% accuracy.

Scientists can predict if an earthquake can happen somewhere, but pinpointing the exact arrival time has always been impossible. These researchers said that they’re just a few steps away from changing that.

“We never know. It’s not like hurricanes,” said Cemil Yavas, a researcher at Georgia Southern University. “You have no time to evacuate. If your building collapses, you’ll die under rubble.”

Their technology uses past earthquake data to predict future seismic events. By loading datasets filled with recorded characteristics of previous earthquakes, they can teach artificial intelligence to recognize patterns and predict exactly when the ground will shake.

“These features create a pattern for AI, for machine learning to create predictions,” said Yavas.

The researchers said their model could also be applied to other natural disasters.

“We can use this type of technology, this type of research to solve real problems,” said Chris Kadlec, an associate professor with GSU.

They’re already working on a way to predict wildfires in California. But for Yavas, the earthquake project is personal. He survived an earthquake in Istanbul that killed more than 17,000 people.

“If they had a warning—just 30 seconds before—then no one would have died,” Yavas said. “That’s what we are trying to achieve, but we are trying to do that 30 days [before], not 30 seconds.”

AI can be considered controversial due to errors, which Kadlec said could possibly be fixed.

“Yes, AI has some serious problems, but maybe we can solve some of those too,” Kadlec said.

Beyond their work in natural disasters, the researchers are also applying their machine-learning models to another field: wine tasting.

“I think it’s important to about no one but us,” said Yavas.

Kadlec added, “It isn’t necessarily life changing.”

They said their research still needs more computing power to be completed, but they expect to have an early warning system developed within the next two years.