And say I used some standard non-AI scheduling software like...

And say I used some standard non-AI scheduling software like Mindbody or JaneApp to schedule an appointment with my mechanic and asked for an appointment to have my tires changed and rotated. If I ended up having my oil changed because the software simply schedules the most common kind of appointment, this would be a clear sign that the software is buggy and no reasonable person would argue that zero effort should go into fixing this bug. And yet, this is a common argument that people are making with respect to AI (it’s probably the most common defense in comments on this topic). The argument goes a bit further, in that there’s this explanation of why the bug occurs that’s used to justify why the bug should exist and people shouldn’t even attempt to fix it. Such an explanation would read as obviously ridiculous for a “classical” software bug and is no less ridiculous when it comes to ML. Perhaps one can argue that the bug is much more difficult to fix in ML and that it’s not practical to fix the bug, but that’s different from the common argument that it isn’t a bug and that this is the correct way for software to behave.

Comments
www.joshbeckman.org/notes/735436481