THE NEW COPYRIGHT RULING IS... MEH
BUT You Don’t Need To Read Part One to Follow Along With “Part Deux”
I think the current copyright ruling has tons of grey areas and solves… basically nothing. And I’m not any more optimistic about the future. What do you think? Read along and then comment.
SUM IT UP FOR ME:
You might have read that the current Copyright Office ruling on AI-generated content was a boon for writers. Not so fast.
There’s an old joke:
The man who created the Thermos takes his new invention to a group of people. He pours them a hot drink that has remained hot. He pours them a cold drink that has remained cold.
One person speaks up: “How DOES IT KNOW?!??!?!”
When it comes to AI-Generated content… How do we know????
So the whole Copyright Office ruling is really just dumb. How do we know what’s been generated by AI and what’s been created by humans? The line is blurry now, and will be indistinguishable later. So much for deciding copyright.
I don’t think anyone in Washington gets this.
WAIT! I DIDN’T READ PART ONE! CATCH ME UP!
For those of you who are wondering, “If I didn’t read part 1 of the Copyright Office’s report on ‘Copyright and Artificial Intelligence,’ will I still enjoy part 2?
The answer is ... YES! I’ll help you out.
The only thing you really need to know about Part One (of THREE scintillating parts) is that the Copyright Office punted on digital replicas, the things actors, singers, performers of all kinds worry about. In essence, they said:
While current laws do NOT protect against the harms of AI-generated digital replicas...
AND there needs to be new federal legislation to protect individuals
It’s not our job to pass laws. So, at this point, a digital replica of a person's image or voice alone does not constitute copyright infringement.
Therefore, if someone creates a digital replica of you without using your copyrighted material (e.g., using an AI trained on publicly available images), copyright law does not provide you with direct recourse.
At which point, the performers in Hollywood have to ask themselves: “Where is our Sonny Bono?” Sonny single-handedly got Congress to extend the copyright term to “death plus 50 years”, something known as the “Mickey Mouse Extension.” (BTW — it’s 50 years after Walt’s death, meaning THE ORIGINAL Mickey is now public domain. Later versions, not so much.)
There’s no emerging Sonny Bono to defend performers in the case of digital replicas. The topic of why Hollywood isn’t standing up to big tech and LLMs in general is a whole other problem, and beyond this post.
AND NOW, PART DEUX
This week, the Copyright Office ruled that “some” works created with AI are copyrightable, some “might be” and some are not.
Here’s how I interpret the Copyright Office ruling, along with some of the questions raised.
It’s all about humans! If a work is entirely generated by AI it is not copyrightable. I wonder: Is that just for when AI tells AI to generate something? Because, I’m pretty sure that, even in the Nick Bostrom never-ending “Paper Clip” example, it BEGINS with a human. Can I copyright a work that begins with me typing the words “Write a novel about a teenage time-traveling vampire in Ancient Rome?” Oh wait… you’re about to answer that…
The mere selection of prompts, even if those prompts are detailed and are the product of some human effort, does not itself yield a copyrightable work, although this determination could change as technology evolves. I still need some help here… What part of the writing do I need to generate to make this copyrightable? And… what evolution of technology will change this?
Where a work includes both human and AI-generated content, only the human contributions are potentially copyrightable. Hmmm… so, if I feed my story summary / outline into AI, that part is copyrightable. But what AI does with it… isn’t? You just told me prompts are enough… so… number of words? Little help here, copyright office.
The use of AI as a tool to enhance the human creative process (e.g., for ideation or to edit an image) does not render the entire work uncopyrightable. Got it. I think. Maybe not.
Alex McFarland of AI Disruptor has broken it down to three scenarios:
Fully AI-Generated (No Human) ... NOT
AI-Assisted with Human Input - PARTIALLY. The human portions, yes. AI-generated portions, no.
AI used as a creative tool... YES... If the human “controlled” AI and actually formed the traditional elements of authorship, the work is copyrightable.
(Alex is also launching a Notion-based research hub for AI copyright research. He’s the real deal, so if this is important to you, get over there and subscribe.)
The grey area comes when you start defining what things like the “traditional elements of authorship” are... or what a prompt is, and how it “forms” the final content.
ZD Net did a nice job of comparing using AI as a tool to what a photographer does with a camera. The actual work of the camera is all light, physics and chemistry. But the photographer puts themselves and their subjects into position, chooses the moment.
BAD NEWS FOR SUNO AND SUDOWRITE?
I suspect this IS bad news for Suno and Sudowrite. Based on my work with Sudowrite, just typing in “Give me a teen coming-of-age story set in Ancient Rome with a time-traveling vampire” might get you a novel in a week... but it’s not copyrightable.
Then again… I really don’t have any faith in this getting resolved.
PART 3: THE FINAL COUNTDOWN
A lot of AI has been trained on copyrighted material. Part 3 is going to address that. Judging by the pattern established by 1 and 2 we should expect … a might punt.
Part 3 will focus on the issues of training AIs using copyrighted works, aspects of licensing, and how liability might be allocated in cases where a spectacular AI failure can be attributed to training (which sometimes results in litigation).
It’s almost impossible to separate the needles of copyrighted content from the haystack of LLM intelligence. My bet is that they’ll say “Okay, we can’t cover what’s already been used... and we encourage laws going forward.”
Which will, I think, leave this to the marketplace. If you’re a writer, you’re going to just have to believe that humans WILL generate better stories and content. Or not.
YOUR TURN
Am I right? Wrong? Any lawyers out there who want to educate me?
If you want to keep following issues around writing and AI, and dig into the process of writing with AI, subscribe here.
And if you know someone who will like diving into this discussion, please share.
Interesting POV. Appreciated it. Whether people agree with the new rules or not, I think one thing is certain: this is just a preview of some serious issues coming soon. I could see the process getting really complicated and specific to each individual case.
While my view is that it's a step forward for trying to create some type of framework, two main issues come to mind:
1. This will probably be irrelevant in like 6 months due to the fast pace of development.
2. Who the hell will want to go through this process of documentation when we don't even know what it looks like. (Screenshots?)
And lastly, just look at the new stuff from Omnihuman: https://omnihuman-1.com/
These are totally uncharted waters..
Hey Fred. I did a deep dive on this over at https://hightechcreative.substack.com/p/copyright-ruling-on-ai-generated recently, covering it from the point of view of people looking to gain copyright on something they had created with A.I. - on that front, the ruling was reasonably good - progressive even. They reasoned some things out far more logically than I had expected.
You were pretty much bang on most of your points there with regards to what is copyrightable and what isn't, which I thought made a lot of sense. There are people who would like to see things generated with just a prompt and no other work copyrightable but honestly I think this makes a decent balancing of it (though there are some issues in their assumptions I work through in the article).
I hadn't considered too closely the idea of people being harmed by digital replica's, could you expand on this? Are you talking primarily about people stealing the "style" of a painting or writing? If so, that's unlikely to be something solved as it was already both possible and perfectly legal to do that - copyright intentionally doesn't protect ideas or abstract things like styles, only exact works. A.I. makes it easier to do this, but it doesn't make it possible - It always was.
You were right to focus on the concern about how we tell something is A.I. generated though. For all the people patting themselves on the back about how they can "detect A.I." because they've seen a few people churn out really crappy writing with it, both A.I. art and writing can be quite good and are getting better, it's already possible for it to be impossible to tell in some cases and this will only become more-so.
The problem though is even worse when looked at from the other side. Consider the problem of enforcing your copyright now. Say you create a piece of art and discover someone has plastered it on t-shirts and mugs and is making a tidy profit selling it.
If you try and stop them and they claim that they are perfectly entitled to because you generated the art and thus can't copyright it, now the burden of proof is on you to show you have the rights you're trying to assert - and that isn't always going to be an easy thing to do. Someone could easily bleed you badly in court costs dragging that fight out, and if it's a corporation ripping you off rather than just some regular person?
Technology moves faster than law, as always, we're going to need some careful thought around this one because the answers aren't going to be obvious.