if someone seriously suggests relying on machine learning models for safety-critical systems, never take them seriously again. If someone cannot understand why life and death should not be decided by unnecessarily complex and unauditable software, I'm not sure you can trust them with safety in general.