Anthropic wins key AI training court battle

The courts just handed down an important ruling about the training of AI from internet data. If this interests you, you should read this article:

Anthropic wins key US ruling on AI training in authors' copyright lawsuit | Reuters

When reading that article, don’t get misled by the fact that it covers two separate issues.

As I have been saying, the court has ruled unequivocally that training AI off of publicly available media is fair use. In my experience people are often confused by Copyrights and Trademarks. The key thing to understand about copyright is that it says exactly what it is: a right to control the copying of a work. Thats it. In the US an artist has no control over the use a copy of their work is put to. In the European union there is something that is somewhat deceptively called “moral rights.” This does not mean the author has a right to make others subscribe to their morals. Moral rights are only about attribution and are 2 specific rights: The right to attribution, and the right to not be mispresented. Thats it.

The court decision hinged on a key question when it comes to fair use which is how transformative is that use? The more transformative it is, the more likely it is to be fair use. Converting a body of works to statistical information about those works the court ruled to be “highly transformative.”

In a separate, second ruling about a companies right to keep a library of copies of works for the purpose of transformation, the court was more inclined to call that copyright infringement and I would personally agree. But the illegality is in making your own non-transient copies and NOT the use to which they are put.

This was a bit silly of anthropic as local copies of publicly accessible works are simply unnecessary. Anything that is publicly available from the internet can be read right off the internet, no copies needed. (And no, there isn’t somehow a magical copy of the work as itself inside the data model of an AI. That would be impossible due to the storage requirements. You can read more about that in information presented by the EFF here. (How We Think About Copyright and AI Art | Electronic Frontier Foundation)
In the end, this is the law. An artist does not have any right to control the use of a legally obtained copy of their work. Only to control the distribution of their content. And the more transformative the usage, the less it becomes something that they can control.

If you don’t like that, then lobby your congress creature for a change to the laws.