Well, I often get this counter-arg. Sure, human work is not 100% error-free, but the types of error are vastly different. And this matters.
Types of errors, and then the frequency as well. And then there are so many tools that help humans bring down quantity and seriousness of the errors. These tools were designed for those kinds of human errors (which are somewhat predictable). But you can't use them on the unpredictable AI errors.
At other times, AI is not strictly in error, but the outcome is not what you exactly wanted. Can you imagine this in a medical scenario?
Do you want to fly in a plane that goes 100% on Auto-pilot + AI (the best you have in say, 2045) versus a plane piloted by a competent and experienced human pilot? Imagine you're traveling with all of your kids and extended family members. Would you take that risk knowing that humans also make mistakes as you said? Which risk would you be comfortable taking?
Types of errors, and then the frequency as well. And then there are so many tools that help humans bring down quantity and seriousness of the errors. These tools were designed for those kinds of human errors (which are somewhat predictable). But you can't use them on the unpredictable AI errors.
At other times, AI is not strictly in error, but the outcome is not what you exactly wanted. Can you imagine this in a medical scenario?
Do you want to fly in a plane that goes 100% on Auto-pilot + AI (the best you have in say, 2045) versus a plane piloted by a competent and experienced human pilot? Imagine you're traveling with all of your kids and extended family members. Would you take that risk knowing that humans also make mistakes as you said? Which risk would you be comfortable taking?