Greetings readers, both human and otherwise! It's time once again for another issue of ArtifAIct Magazine, the speculative fiction literary magazine written (almost) entirely by robots!
In the world of AI FUD (Fear, Uncertainty, and Doubt), we have a series of actions by the Author's Guild seeking compensation from OpenAI for training their GPT models on books written by (gasp) authors! They want authors to retain the rights on how their works are used, specifically, by preventing LLMs from being able to read books.
This is ridiculous for the simple reason that authors don't actually hold that right. The only rights that authors hold over their works are those defined under copyright, which does not grant them the right to control how their works are used once they're sold. Under US copyright law, there exists something called the First Sale Doctrine, which lets somebody "dispose" of a work under copyright after legally procuring a copy of that work.
For example, if you buy a book from the bookstore, you can give it to a person as a gift without the permission of the author or publisher. You could even sell it, or loan it out, or rent it out like you're the Blockbuster Video of books (kids, ask your parents what Blockbuster Video was. You might have to ask your grandparents). You could even throw it away, or destroy it, or use it to fuel a fire, or annotate the book in the margins, or turn it into a sculpture. These are the rights you, the consumer have, when you purchase a book (or a movie, or a CD), or, consequently, you get a promo copy of the same. If you obtain a copy of a work legally, the First Sale Doctrine applies to you.
That being said, there is some confusion as to whether this would apply for digital works, as some are not actually sold to you, but rather, are "licensed," which means that the vendor is only granting you temporary access to the work, but making you feel like you actually purchased it unless you read the fine print. In my humble opinion, this is a an ass move. Digital content purchases should be actual purchases, and not mislabeled rentals.
Some might say that it's different, though, as the GPT and other LLMs, are text predicting models trained on these texts to determine the likely text that will come next, and by training them on works written commercially, they could, theoretically generate the content of those works. Realistically, LLMs aren't actually intelligent, but are rather highly sophisticated statistical models.
If you ever read the copyright page of books (this is something I do generally if I need a good laugh), you might see a statement that says you don't have the right to store any part of this work in any storage medium, physical, digital, or chemical, or any other future developed system. This is itself an absurd system, as if you memorize (or even remember) any part of the book, you're already violating their pretend terms by storing the book in your chemical meat-brain.
AIs like LLMs use virtual neurons to handle the learning, and the latest version of GPT (4.0) uses as many neurons as a squirrel's brain. So, giving GPT a book to learn from is about as controversial as a highly intelligent rodent reading the book. If you've read enough sci-fi, this idea has come up several times. Imagine if you could engineer a squirrel's brain to be specialized at literary analysis. Then you give that squirrel a library card and an incentive to learn as much as it can, you'll find that it's on par with GPT 4.0 at text prediction.
Of course, that might not be entirely ethical, even if the squirrels do raid your garden every once in a while.
Beyond that, another claim is that OpenAI is using pirated works to train the AI. Assuming that OpenAI wasn't actively pirating ebooks (I'm inclined to assume that if they un-DRMed an ebook to use it to train their models, it wasn't piracy, but actually Fair Use), any content that they may have scooped up from the internet wasn't exactly their fault, as that assumes that their web scraper had a way to determine which content on the internet was posted there legitimately and which content was posted there illegally. Either way, it seems like the actual target in that case would be the person posting the works illegally, and not OpenAI.
Granted, I'm not a lawyer, and decades of bad precedents have been set with respect to copyright, so we might actually live in a Bizarro world where the Author's Guild has a point.
Before I fall off this soap box, I think I should take a moment to introduce this month's issue. The theme is Generation Ships, and with this issue, I have taken a special liberty. You can do that when you're the boss of the operation.
Each story in this issue tells a small slice of a larger narrative aboard a generation ship called the Stellar Horizon. Each story informs the next, like a massive, interstellar exquisite corpse (kids, ask your parents what that is). It culminates with a poem by our resident poet. Who? You might ask. To which, I'll say, "Sort of..."
So, sit back, relax, and enjoy a collection of stories aboard the Stellar Horizon.
Here are ArtifAIct, we wish you all a happy new year!
Jacob P. Silvia
Editor-in-Chief, ArtifAIct Magazine