We're not currently able to predict the actual earthquakes at all, only probabilities of likelihood on very large timescales. I could see it being useful to predict future damages to certain buildings/areas based on existing data, assuming an earthquake happened.