Ars Technica: At a hearing Monday, US District Judge William Alsup blasted a proposed $1.5 billion settlement over Anthropic's rampant piracy of books to train AI.
The proposed settlement comes in a case where Anthropic could have owed more than $1 trillion in damages after Alsup certified a class that included up to 7 million claimants whose works were illegally downloaded by the AI company.
3 comments:
I understand the want from the Author’s Guild to send a strong, early message that it will cost you to use other’s work to train algorithms on, and it is true that doing this early will help people quickly to ensure their work is properly used, but nonetheless it feels massively too small a price to pay for the infringement at play here. Thankfully the Judge agrees, but $1.5 billion to somewhere around 500,000 authors / works (mentioning a rough $3,000 per work) feels too low a price to pay when millions of works were used to train the dataset and give Anthropic their successful algorithm. While only 500,000 works have been identified as fitting the category of infringement on some definition, it is absurd to think that the algorithm would be nearly as successful just on that smaller dataset. By focusing on an early and albeit meaningful win, we are giving a massive pass to the other work that largely shapes out these algorithms. There ought to be a larger fight for more compensation and a broader definition of infringement here. These algorithms do not get successful by just the subset, but instead how the entire dataset fills out the boundary cases.
I’m glad the Author’s Guild is taking a firm stance on this. Artificial intelligence development over the past few years has accelerated dramatically, and consequently the training data sets have ballooned in size. In creating those data sets, companies have often committed massive copyright infringement upon the original authors of the works. In an ideal world, Anthropic would have to pay the $1 trillion figure originally quoted in damages, as opposed to the pitiful $1.5 billion. The fact that they are trying to evade that is, unfortunately, typical. But, given the fact that the judge appears to be on the authors’ side in this situation is a big win, given the current state of the United States Justice Department. Hopefully, the authors of the original 7 million individual works are able to get a ruling in their favor.
Fines do not make sense for anyone that thinks of that fine as an inconvenience rather than a threat. 1.5 billion to a company that is worth almost two hundred times that is a burning bag of shit on your porch. It is something that when put out immediately leaves a bigger problem than you started with. Fines and settlements should be tied to the money that the company profits or net revenue from the abuse of others. Anthropic’s preliminary settlement is more money than billions of people will ever have in their lifetime but it is barely worth noting on Anthropic’s budget. This is a cost of convenience that Anthropic will profit over a hundred or thousand fold. The judge should be applauded for refusing their preliminary offer because this would not cover the abuse of the copyrights of authors and would likely lead to a similar action by Anthropic or other AI company later down the line.
Post a Comment