Are Human Experts Less Prone to Catastrophic Errors than Machine-Learned Models?