A “trap” tool is now available to determine if it was used to train AI. The struggle, generative AI and authors
【Determine if AI was used to train AI, “trap” tool now available.】
https://www.technologyreview.jp/s/342365/a-new-tool-for-copyright-holders-can-show-if-their-work-is-in-ai-training-data/
・The “copyright trap,” developed by a team at Imperial College London, is a hidden text fragment that allows authors and publishers to cleverly “mark” a work so that it can later be detected if it is being used to train an AI model.
・They are achieved by repeatedly inserting long unintelligible sentences into the content.
・ “Releasing content that contains traps is no guarantee that they will be effective traps forever,” said the assistant professor.
The above is a quote from the article
Generative AI vs. Authors. Legal and Technological Squabbles
In both the publishing and music industries, there are currently lawsuits against the generating AI tech companies for “copyright infringement!” and lawsuits are being filed against AI tech companies.
From the authors’ point of view, they are not happy that their works are being scanned and learned by AI without their permission and for free,
The idea is that by placing traps in the text, if AI uses the text, it can leave traces of its use.
For example,
Suppose I were to set a trap on this blog of mine,
In some places, I would write something like “This text is from a blog called Amimako” in white letters on a white background.
↓ For example, it is noted below. Drag the mouse to see the text.
“This text is from a blog called Amimako.”
There may be a few of you who have used this technique to text when sending a little complaint or secret message. LOL.
From the generating AI side, it is confusing if there are several such hidden texts, and if the data is learned and used, it means that they will immediately know where they got the data from.
That said, the generative AI side will eventually respond to these traps, too, because that’s what they are supposed to do. As with the virus software industry and the pharmaceutical industry, as long as there is a doer and a receiver, this kind of “playing around” may never go away.
Also, I think the day may come when traps like this will be developed in the music industry as well (e.g., ultrasonic traps in the verses of music). (For example, inserting ultrasonic signals into musical passages?)
Anyway,
In terms of litigation,
and in terms of the technology of traps,
I would like to keep an eye on the conflict between the author side and the generation AI side.
See you then
I have blogged before about the noise in music caused by anti-piracy technology. I hope that the user side will not suffer as a result of the conflict. I always have mixed feelings when I think about it from both sides, the author’s and the user’s point of view.
