“Like putting on glasses for the first time”—how AI improves earthquake detection (arstechnica.com)

Over the past seven years, machine-learning tools that treat seismic records like images have largely automated one of seismology’s core tasks: detecting earthquakes. These AI systems can pick out tiny events—like a 2008 Calipatria quake of magnitude −0.53 that would have gone unnoticed by people—from continuous, noisy data streams, especially in urban settings where human analysts and older algorithms struggle. Researchers describe the effect as transformative—“like putting on glasses for the first time”—because reanalyzing existing datasets with these methods reveals many more small quakes and finer details in seismic activity. The technical and scientific implications are significant: richer, more complete catalogs improve our ability to map subsurface structure and assess seismic hazard, and they free human analysts from routine detection so they can focus on interpretation and higher‑order problems. Yet the disruption has limits. AI has replaced humans for detection tasks but hasn’t yet produced the next leap—reliable forecasting or wholesale automation of all seismic data processing. Experts call it an ongoing revolution: detection is largely solved, but many other seismological workflows remain ripe for innovation.