🚩 Report: Ethical issue(s)

#3
by Avelina - opened

This dataset contains over 5 million images sourced from an art distribution website, without express permission from the original artists.

Were this to be exclusively used to train tagging/captioning models there would be little ethical issue, but this dataset will almost certainly be used to train diffusion models which most of the original artists will object to and additionally give AI an even worse name than it already has among the art community.

It's legal in Japan, just saying.

Not legal in ALL cases, just saying.
There's no precedent, so I don't know if its illegal.

Agency for Cultural Affairs in Japan (文化庁)
from: https://www.bunka.go.jp/seisaku/bunkashingikai/chosakuken/pdf/94037901_01.pdf

「AI と著作権に関する考え方について」 ページ27

(オ)海賊版等の権利侵害複製物を AI 学習のため複製することについて
「特に、ウェブサイトが海賊版等の権利侵害複製物を掲載していることを知りながら、当該ウェブサイトから学習データの収集を行うといった行為は、厳にこれを慎むべきものである。」

"On AI and the Concept of Copyright." (P.27)

O: Reproduction of pirated or other infringing copies for the purpose of learning AI

Particularly, the act of collecting training data from websites that you know contain pirated or otherwise infringing copies should be strictly avoided.

Sign up or log in to comment