Invest in artificial intelligence to predict earthquakes

Seismology is an incredibly complex field of science. Earthquakes and aftershocks are still unpredictable in their timing, location and intensity. Despite strenuous research, seismologists have yet to identify a reliable precursor to earthquakes. Such a thing would be any kind of geological phenomenon that consistently precedes every tremor. Science has revealed much about our world, but understanding the tectonic plates seems to be just outside of our reach.

Earthquake.jpg

Nonetheless, public pressure to find the answer is growing. Earthquakes can be destructive and terrifying. The Nov. 30 earthquake here in Alaska caused extensive damage to UAA, resulting in a four-day closure. Sometimes survivors are eager to point fingers. In Italy, six scientists were sentenced to six years in prison for manslaughter after they failed to predict a 2009 earthquake that killed more than 300 people.

We know the process of earthquakes. They are sudden releases of stress from the Earth’s crust, built up as tectonic plates move against each other. However, the geological variability of different locations that earthquakes occur makes it difficult to build simulations of them. Any laboratory simulation will only apply to the geographic inputs that seismologists type into the computer. For example, the earthquake that struck Alaska this year involved different geographic features than anywhere else on Earth. Seismologists cannot practically create new simulations for every possible variable in any given earthquake.

Fortunately, there is one type of seismologist that has the ability to do this, but it isn’t human. Artificial intelligence is gradually being accepted by the scientific community for its potential role in earthquake prediction. Only AI possesses the mass-computational prowess to arrange and sort hundreds of thousands of geographic variables. Only AI can generate enough virtual simulations to accurately represent earthquakes as they occur in different places, depths and intensities. From these simulations, only AI can identify patterns and possibly a precursor to every earthquake.

Research has already begun to yield results on this. Paul Johnson, a geophysicist at Los Alamos National Laboratory, submitted a detailed report in 2017 that recorded his team’s progress on machine learning for earthquakes. Johnson’s team fed the computer raw data from earthquake measurements. This differs from how scientists have tried to predict earthquakes in the past, primarily by using the United States Geological Survey’s Earthquake Catalog. The USGS catalog is limited because it only includes magnitudes, locations and times. Johnson’s machine was able to process raw data, which includes all measurements taken, even if they don’t seem relevant, and produce a pattern out of it. Patterns can be thought of as the foundation for discovering an earthquake precursor.

Since Alaska is the most earthquake prone state in the U.S., it is only fitting that we get a stake in this AI research. UAA hosted its own AI system for the purpose of Arctic research. This university has also demonstrated its ability to secure federal support, as evidenced by the Arctic Domain Awareness Center with the Department of Homeland Security. Given this history, UAA is well-positioned to pursue access to seismic research AIs in the Lower 48. UAA’s Department of Geological Sciences and Computer Science and Engineering should cooperate to locate, petition, acquire and utilize these AIs.

- Advertisement -

The end-goal here isn’t about academics, though. UAA only serves to provide the intellectual foundation for a future where Alaska can employ an earthquake AI permanently. Eventually, this fully operational AI will probably move to the Alaska Earthquake Center at UAF. The good thing about a robot seismologist is that it doesn’t matter where it is located. As long as it is maintained and fed a continuous supply of raw data, it can work around the clock to observe seismic activity in Alaska.

This AI will require a multitude of measurements and sensors. There are currently 280 sensors in place in Alaska. The State of Alaska and the USGS should expand efforts to increase that number, because no amount of raw data is too much for an AI. The AI can process a virtual mapping of Alaska’s unique geophysical variables. Human operators will input data on previous earthquakes in the state. Then the AI will work to identify Alaska-specific seismic patterns and, hopefully, a precursor.

Predicting earthquakes would be the most ideal employment for this AI. But we should always be temperate when it comes to estimating future technology. At the very least, an Alaskan seismic AI can work to predict aftershocks. This would be easier since the AI will already have the initial earthquake’s data to work off of. Either way, Alaskans need to reconcile with the fact that a tremor rivaling the intensity and duration of the 1964 Good Friday Earthquake is inevitable. It would be good to make use of cutting-edge technology to prepare for that day.