Self-organised criticality of the crust appears to make deterministic earthquake prediction of time, place and magnitude of individual large earthquakes inherently impossible. This closes one line of approach to mitigating earthquake hazards. This paper suggests that a viable alternative to earthquake prediction is monitoring the build-up of stress before a large earthquake can occur. A new understanding of rock deformation allows stress changes to be monitored with seismic shear-wave splitting (seismic birefringence). With a suitable monitoring installation, this would allow the stochastic proximity of impending earthquakes to be recognised so that earthquakes could be forecast in the sense of recognising that crustal deformation was preparing for a large earthquake. Such stress-forecasting is not prediction, but, in many circumstances, a possible forecast crescendo of increasing urgency is exactly what is needed to best mitigate hazard to life and property.
