People want brief explanations of reasons for outputs, and a 10gb pile of weights isn't really what they mean. Human explanations meanwhile are often invented

People want brief explanations of reasons for outputs, and a 10gb pile of weights isn't really what they mean. Human explanations meanwhile are often invented to fit evidence to personal biases and beliefs, and are thus typically deeply flawed. But we're more ok with humans making suspect decisions than ML, in many cases.