Over the last few years, there has been rapid growth in research activity on ML applications in seismology, thanks to the increasing size of seismic data, enhancement in computational power, and the availability of easy‐to‐use ML frameworks. The contributions can be generally grouped in three categories: engineering automation, modeling/inversion, scientific discovery. In this talk, I will go through our recent studies that happen to fall into each of these categories: 1) Automatic earthquake/noise discrimination in earthquake early warning; 2) Modeling of realistic high-frequency ground motions in hazard mitigation; 3) Discovery of rupture complexity patterns of global moderate-large earthquakes. These examples hopefully give the audience a sense of how to steer the power of ML tools in seismological research.