Using open datasets means using data people have made available publicly, for free, for any purpose. So using an AI based on that seems considerably more ethical.
Except gen AI didn’t exist when those people decided on their license. And besides which, it’s very difficult to specify “free to use, except in ways that undermine free access” in a license.
The responsibility is on the copyright holder to use a license they actually understand.
If you license your work with, say, the BSD 0 Clause, you are very explicitly giving away your right to dictate how other people use your work. Don’t be angry if people do so in ways you don’t like.
This specifically talks about AI data scrapers being an issue, and some general issues that are frankly not exclusive to open access info.
Exploitative companies are always a problem, whether it’s AI or not. But someone who uses the Wikipedia text torrents as a dataset isn’t doing anything of what is described in that article for example.
X’s data crawlers don’t give a shit because all their work is closed source. And they have lawyers to just smash anyone that complains.
X intends to resell and make money off others’ work. My intent is free, transformative work I don’t make a penny off of, which is legally protected.
That’s another thing that worries me. All this is heading in a direction that will outlaw stuff like fanfics, game mods, fan art, anything “transformative” of an original work and used noncommercially, as pretty much any digital tool can be classified as “AI” in court.
Oh, so you deserve to use other people’s data for free, but Musk doesn’t? Fuck off with that one, buddy.
Using open datasets means using data people have made available publicly, for free, for any purpose. So using an AI based on that seems considerably more ethical.
Except gen AI didn’t exist when those people decided on their license. And besides which, it’s very difficult to specify “free to use, except in ways that undermine free access” in a license.
The responsibility is on the copyright holder to use a license they actually understand.
If you license your work with, say, the BSD 0 Clause, you are very explicitly giving away your right to dictate how other people use your work. Don’t be angry if people do so in ways you don’t like.
How does a model that is trained on an open dataset undermine free access? The dataset is still accessible no?
“Wait, not like that”: Free and open access in the age of generative AI
This specifically talks about AI data scrapers being an issue, and some general issues that are frankly not exclusive to open access info.
Exploitative companies are always a problem, whether it’s AI or not. But someone who uses the Wikipedia text torrents as a dataset isn’t doing anything of what is described in that article for example.
To be fair, he did say he “used some open databases for data”
Musk does too, if its openly licensed.
Big difference is:
X’s data crawlers don’t give a shit because all their work is closed source. And they have lawyers to just smash anyone that complains.
X intends to resell and make money off others’ work. My intent is free, transformative work I don’t make a penny off of, which is legally protected.
That’s another thing that worries me. All this is heading in a direction that will outlaw stuff like fanfics, game mods, fan art, anything “transformative” of an original work and used noncommercially, as pretty much any digital tool can be classified as “AI” in court.