AI systems fail differently. They produce output that's fluent, well-structured and plausible, even when that output is wrong ...