Your description of how "archive" becomes "dataset" is incisive & needed!
Others have noted this shift, though with less critique:
Andrej Karpathy describes as moving from "Software 1.0", where humans write the logic for the computer to execute, to "Software 2.0", where humans just feed the machine data and it decides the logic. In other words, the software is written by machines, in the weights of a neural net. He describes the benefits of moving to Software 2.0, and (briefly) the dangers, here: https://karpathy.medium.com/software-2-0-a64152b37c35
Your description of how "archive" becomes "dataset" is incisive & needed!
Others have noted this shift, though with less critique:
Andrej Karpathy describes as moving from "Software 1.0", where humans write the logic for the computer to execute, to "Software 2.0", where humans just feed the machine data and it decides the logic. In other words, the software is written by machines, in the weights of a neural net. He describes the benefits of moving to Software 2.0, and (briefly) the dangers, here: https://karpathy.medium.com/software-2-0-a64152b37c35
He also talks about it with Lex Fridman: https://www.youtube.com/watch?v=cdiD-9MMpb0&t=3944s