Artificial intelligence can diagnose cancer, forecast cyclones, translate languages and even help spacecraft navigate millions of kilometres through space. Yet when the Earth suddenly shakes, it confronts humanity with a question AI still cannot answer: When will the next big earthquake strike?
The answer, geophysicists say, lies deep beneath our feet. Unlike weather, which is continuously monitored by satellites, radar and weather stations, earthquakes originate several kilometres underground, hidden from direct observation. There, tectonic plates creep at little more than the speed of growing fingernails, slowly building stress over decades or even centuries before releasing it in seconds as powerful seismic waves. Scientists understand why earthquakes occur. Predicting exactly when, where and how intense the next one remains one of science’s greatest challenges.
“An earthquake is not an isolated event. It is the outcome of geological processes that may evolve over hundreds or even thousands of years,” said Dr Sreenagesh, former chief scientist at the National Geophysical Research Institute (NGRI), Hyderabad.
That immense timescale also explains why AI struggles. Artificial intelligence excels when it can learn from vast amounts of high-quality data, but earthquake science simply doesn’t have enough of it. Instrumental earthquake records extend back only about 150 years, while major earthquakes on the same fault may recur only once every few hundred, or even several thousand, years.
“We simply don’t have enough examples for AI to identify dependable patterns. It is a classic case of ‘garbage in, garbage out,” Dr Sreenagesh explained.
Even if more data were available, the Earth itself remains extraordinarily difficult to decode. Rock composition, underground fluids, pressure, temperature, fault geometry and friction all influence when a fault may rupture, yet many of these processes cannot be observed directly. Scientists are effectively trying to understand a machine whose most important components are buried kilometres beneath the surface.
That does not mean AI has no role in earthquake science. Machine-learning algorithms now detect tiny earthquakes hidden within enormous seismic datasets, analyse satellite imagery to measure subtle ground deformation, identify previously unknown faults and improve long-term seismic hazard assessments. These advances are transforming how scientists study earthquakes, even if they stop short of predicting them.
One study, however, has offered a glimpse of what the future might hold. Researchers at the University of Texas at Austin developed an AI model that reportedly forecast nearly 70% of earthquakes up to one week in advance during a seven-month trial in China. The findings generated international interest because the algorithm appeared to recognise subtle statistical signals missed by conventional methods.
Scientists, however, remain cautious. The model has yet to demonstrate the same level of success across different tectonic regions or over longer periods.
“One encouraging experiment does not solve the earthquake prediction problem. A prediction system must perform consistently across different regions and over many years before it can be trusted,” said a seismology expert at IIT Hyderabad.
The study also highlights an important distinction that is often overlooked. Earthquake prediction is not the same as earthquake early warning. Prediction means forecasting an earthquake before it begins, a capability science still does not possess.
Early warning systems, unlike prediction, detect an earthquake only after it begins and issue alerts before the strongest shaking arrives. Depending on the distance from the epicentre, these warnings can provide a few seconds to several tens of seconds to stop trains, shut down industrial facilities, pause surgeries or take protective action. Countries such as Japan have demonstrated how effective these systems can be, while India has strengthened its seismic monitoring network, particularly across the Himalayan region.
For now, however, scientists say preparedness remains humanity’s best defence. Earthquake-resistant buildings, strict building codes, public awareness and efficient disaster response can significantly reduce casualties. According to Dr Sreenagesh, incorporating earthquake-resistant features typically adds less than 10% to a building’s structural cost, far less than the losses caused by a major earthquake.
Perhaps, the race to predict earthquakes is as much about understanding the Earth as it is about advancing technology. Until researchers unlock the mysteries hidden beneath the planet’s surface, no algorithm can say with certainty when the ground will shake next.