The idea that AI will get this right (for interesting / realistic use cases) sufficiently often to take the human out of the loop seems optimistic pre-AGI. Most

The idea that AI will get this right (for interesting / realistic use cases) sufficiently often to take the human out of the loop seems optimistic pre-AGI. Most humans I’ve tried to help with data analysis don’t know what they want, don’t know how to get it, don’t know how to interpret it, and don’t know how to check if their analysis is correct. (Fortunately, mostly they only want to use the data to affirm their preexisting decisions, so it’s actually fine).