In 1978, the United States Geological Survey (USGS) allocated over half its research budget ($15.76 million) to earthquake prediction, a level of spending that continued for much of the next decade. Scientists deployed hundreds of seismometers and other sensors, hoping to observe telltale signals heralding the arrival of the next big one. They looked for these signs in subterranean fluids, crustal deformations, radon gas emissions, electric currents, even animal behavior. But every avenue they explored led to a dead end. ... Since the early 20th century, scientists have known that large quakes often cluster in time and space: 99 percent of them occur along well-mapped boundaries between plates in Earth’s crust and, in geological time, repeat almost like clockwork. But after decades of failed experiments, most seismologists came to believe that forecasting earthquakes in human time—on the scale of dropping the kids off at school or planning a vacation—was about as scientific as astrology. By the early 1990s, prediction research had disappeared as a line item in the USGS’s budget. ... Defying the skeptics, however, a small cadre of researchers have held onto the faith that, with the right detectors and computational tools, it will be possible to predict earthquakes with the same precision and confidence we do just about any other extreme natural event, including floods, hurricanes, and tornadoes. The USGS may have simply given up too soon. After all, the believers point out, advances in sensor design and data analysis could allow for the detection of subtle precursors that seismologists working a few decades ago might have missed. ... At a time when American companies and institutions are bankrolling “moonshot” projects like self-driving cars, space tourism, and genomics, few problems may be as important—and as neglected—as earthquake prediction.