Overpreparing for artificial general intelligence is low risk compared to the high risk of under-preparing.