Though I don’t often comment on court cases, this one, as reported in the L.A. Times and in TechSpot, involved fiction authors and artificial intelligence (AI), so I couldn’t resist.
The Prosecution
More than 500,000 authors—not including me—sued the AI company Anthropic, contending the company illegally obtained the authors’ books and used them without permission.
The authors claimed the company downloaded the books from pirate websites, then fed the electronic text into the company’s AI to train it. However, the authors wrote the books to be read by humans, not to be used as training for AI.
The Defense
I’m not sure what Anthropic said about how they obtained the books. However, they contended that feeding text into an AI to train it is “fair use,” a term used in copyright law.
In the U.S., you can’t copy someone’s work and use large chunks of it without citing the source. However, if what you do is “transformative,” and adds value to the original work, then it’s fair use. Anthropic claimed the training of its AI was transformative.
The Settlement
From what I can gather, the case involved two issues. Was the company guilty of:
1. Obtaining the books illegally? and
2. Using the books illegally?
If I understand correctly, this case never went to trial. The two sides agreed to a settlement, but the court had to approve it, and did so.
For the first issue, the court ordered Anthropic to pay $1.5 Billion to the authors, who will receive about $3000 per pirated book.
Regarding the second issue, the court found no guilt. Anthropic may use digital copies of books to train their AI.
The Impact
Get set for more related cases. About 350 authors opted out of this settlement, so they may seek compensation via an alternate path. This settlement only affects Anthropic, not the many other companies with AI, who are also training theirs in similar ways. Further, a major newspaper is suing some AI developers with regard to news articles. Movie studios are suing AI developers for training AI using copyrighted movie characters.
My Take
As a science fiction enthusiast, I’m fascinated by AI I’m curious what the real-world future holds. I’ve read enough fiction about AI to know things can turn out well, or lead to our demise.
For most human-invented tools, people can use them for good or evil purposes. At present, AI does what a human tells it to do. It doesn’t initiate actions on its own. When it becomes sentient and takes action of its own accord, it will start to seem less tool-like and more human. At that point, we’ll have to re-think things. We may have to start regarding AI as a fellow sapient species, with legal rights, susceptible to punishment for illegal acts.
In this particular Anthropic case, I don’t know if the monetary payout is reasonable, but the overall judgements in the settlement strike me as fair. My books have appeared on pirate websites at cut-rate prices, with no money going to the publisher. That’s not fair or legal. As fly-by-night operations, these websites disappear when challenged and spring up under other names, so they’re tough to prosecute. Anthropic shouldn’t have purchased training materials from pirate sites.
AI companies wishing to provide optimal training to their AIs are free to do so after paying full price to purchase books by—
Poseidon’s Scribe