Ep. 58: The U.S. Copyright Office on Artificial Intelligence and Copyrightability, with Jalyce Mangum

Jalyce Mangum, welcome to the show.

How are you doing?

Thanks.

I’m doing pretty great.

A little cold, but pretty good.

Why don’t you introduce yourself to our listeners?

Sure.

So I am an attorney advisor in the Office of General Counsel at the Copyright Office.

I am a supporter of creatives and sort of a in and out creative myself.

I call myself a hobbyologist, but it’s really an honor to be able to support creatives in this way as a public servant at the Copyright Office.

It’s one of the things that drew me here.

So you’re on the show today to talk about the Copyright Office’s report on artificial intelligence and copyrightability.

And as I understand, you were one of the authors of that report.

So could you walk us through just the process of how this was written and what your role in it was?

Sure.

So, of course, it’s a part of an over-arching AI initiative.

We began at the Copyright Office in 2023, and we initiated the initiative with sort of roundtable discussions, bringing in people from the public, artists, technologists, people who are from AI companies.

And we used the information we garnered and gained there to drop this notice of inquiry, asking a bunch of questions about AI and how it impacts copyright.

And we got lots of comments in response to that notice of inquiry, over 10,000.

And so in drafting this report, we really began with parsing through the comments that we received, and over half of the 10,000 comments we received were about copyrightability.

So this report was really drafted as an interoffice collaboration among a number of offices.

Our senior leadership, along with most of the attorneys in the office, were engaged in the AI work.

That’s approximately about a dozen attorneys across all three of the reports that we’re going to issue.

In this report, I was one of the lead drafters, but our team consisted of members from our other divisions, including the Office of Policy and International Affairs, and the Office of Registration and Policy and Practice.

As part of drafting this report, of course, we did review all of those comments, a human laid eyes on all of the comments we received.

The contributors to this report didn’t just include attorneys and didn’t just include me, of course.

The contributors also included our registration examiners who actually deal with claims that we receive, works including AI-generated material.

And they provided really helpful insight.

We also, of course, have graphic designers and public education staff and economists that contributed in a number of ways.

So it was a huge collaboration and we’re really proud of the result.

Could you explain in more detail just how the office works together in generating these reports because I know there are a lot of authors involved and there’s a lot of back and forth.

And just how does that writing process, what does that look like?

I mean, are you sending e-mails back and forth constantly?

Is there a shared document somewhere?

I mean, how does that sort of large scale collaboration look like in real life?

Sure.

So you’re right.

It’s an absolute collaboration and getting it to the place where it is a report that really represents the office’s views.

It takes a lot and it’s a long process, but there are lots of people engaged in that process.

A number of attorneys, registration examiners who filter their experiences through their leadership to inform aspects of the report.

We, of course, start with our overarching outline of how we want to address the topics, the topic of copyrightability.

We started with the comments.

We started reviewing the comments, and themes emerged of things that we had to address.

That got incorporated into an outline.

A number of different attorneys helped draft parts of the outline, and then it got unified to make sure that it’s one voice and one voice that represented the office’s views and not any particular person’s views.

Lots of drafts swapped, lots of e-mails, lots of shared versions that were redlined to infinity, but ultimately the result is a report that reflects the office’s views and guidance.

Well, and I was really interested in the fact that the public comment was so heavily utilized in the final report.

I mean, to the point where you’re citing comments in footnotes of submissions made by some industry people, but also, I assume, just regular folk.

What was the most eye-opening thing to you in that public comment?

Was there anything that stuck out as being really interesting or surprising in some way?

One of the things that touched me the most was that so many people were engaged in so many of the rule-makings that we put out and the reports sometimes that we put out, we don’t really get all of the comments that we want to receive because we do want to hear from a number of different sorts of people.

We want to hear from the small creators.

We want to hear from the people that are just starting out or the people that have been doing it a long time but that haven’t hit the success that they want to.

We want to hear from a number of different types of people.

I was most touched by the fact that we did hear from all kinds of people, from all walks of life.

I think the trend that maybe surprised me was that most people did believe that copyright was human-centric.

The human authorship requirement was really important for most of the commenters that we received comments from.

That surprised me only because maybe I’ve been reading a lot of tech journals and maybe my algorithm is filled with people who believe that AI is the next wave and we’ll take over creativity.

That surprised me, but because it’s so in alignment with where the office is, we do believe that AI is a great addition to how people are creating works, and there’s nothing stopping people from actually using AI to enhance their creative works and incorporating it into their works.

But copyright is human-centric and human authorship is a bedrock requirement for copyright ability.

So, it was great that so many people did see that, even though they don’t, you know, they haven’t studied the Constitution, Title 17 or anything like that, but they just had this instinctive belief that copyright should protect human authorship.

Well, and that’s a great segue to the findings of the report.

And to me anyway, that was, I think, the central finding that human authorship is sort of a non-negotiable part of copyright.

You mentioned it was a bedrock requirement.

Are there other things that would also be considered bedrock in terms of looking at a submission and deciding whether or not it was copyrightable?

Of course, so there are a few, actually.

Under the law, copyright protection subsists or exists in original works of authorship that have been fixed in a tangible medium of expression.

That’s a lot of words.

Then the law goes into listing the categories for which authorship, what is considered a work of authorship.

In that, there are a few requirements.

Of course, the human authorship requirements, which we derive from the Constitution’s use of author and the Copyright Act’s use of author.

But there’s also the originality requirement.

The Supreme Court has said that originality is also a bedrock principle of copyright.

And it actually called it the very premise of copyright law.

And here originality means that to qualify for copyright protection, a work must be original to the author, which means that it is independently created by the author.

But it also must possess at least some minimum degree of creativity.

So, creativity is a really important aspect of copyright ability.

There’s also the fixation requirement, where, you know, work has to be recorded or written in a way that can be perceived for more than a transitory duration of time.

And then there’s the fact that the work has to constitute copyrightable subject matter.

So, you know, one of the enumerated categories that the work goes on to, that the Copyright Act goes on to explain, it has to be a literary work, a musical work, a sound recording, maybe an architectural work.

It has to be, it has to fall in one of those categories of works.

And things that are included are things like ideas, the format and layout of something, or facts themselves aren’t copyrightable subject matter.

So, we’re kind of skipping ahead to later in the report here, but there was this section on the difference between an idea being copyrightable versus the actual expressions of the ideas.

Can you dig in to that a little more?

Because you talk about originality and creativity being important parts of copywritability.

But then later in the report, it also says, well, it’s not ideas that can be copyrighted, it’s those tangible fixed expressions of the ideas.

And I think maybe that’s just the way musicians talk about their art.

But to us, like the idea is the thing you’re copywriting or the originality, the creative part of the music, that’s the thing that you’re copywriting.

A lot of the, sort of, language of music is, you know, patterns and things that you see across all genres or different artists.

And we’re sort of all using the same chord progressions, we’re using the same rhythmic patterns.

And so to us, that almost seems a little backwards, I think.

Yeah, I understand what you mean.

It’s hard to separate the idea from the expression because it doesn’t seem like it’s separate at all, right?

It seems like it’s, like, completely, like, integrated.

I understand that.

The Copyright Act, though, it explicitly excludes copyright protection for ideas, procedures, processes, and that really is to emphasize the fact that what is protected is the expression of your idea.

So, for example, I’ve been really into Wicked, of course.

And so, you know, the idea of a song about flying isn’t protected by copyright.

That’s not copyrightable subject matter.

And you wouldn’t really want it to be.

You don’t want to prevent others from being able to create a song about flying.

But Stephen Schwartz specific expression via the lyrics and the composition of Defying Gravity, it is what, that’s what is protected by copyright.

So, you mentioned, you know, different chord progressions.

We’re all, artists are all using the same building blocks for creativity.

And those individual building blocks, maybe ideas, maybe, you know, too short to be protected by copyright.

But what is protected is the way you combine those building blocks.

It’s the way you use the chord progression and loop it a certain way, adding lyrics or adding different sorts of melodies.

All of that is the expression that’s protected by copyright.

Because we don’t want people to be able to hoard the building blocks of expression.

I think that’s a really great example, and it makes a lot of sense.

One of the things that I think makes music copyright so difficult to understand is that the copyrightable ideas are able to be expressed in so many different ways.

You have the melody and the lyrics.

That’s what I was always taught and understood to be the thing that’s copyrightable.

Like you said, it’s not those building blocks, it’s not the chord structures necessarily, but it’s those melodic ideas, it’s those lyric ideas.

But with music, for example, you can have the lyrics displayed on their own.

You can have various versions of a piece with different accompanying parts.

You can have a recorded version, you can have sheet music, and so you can have the same work appearing in different forms of expression.

And I think that can make it tricky for a lot of musicians to figure out sort of what’s copyrightable and what’s not.

What is important to remember is that no matter what is segmented out, maybe your lyrics do appear on their own.

If they’re sufficiently creative, you created them, they are protected by copyright.

If you decide to change your song and you add something to it, the underlying song, the pre-existing work that you created, is protected by copyright.

But then you may create a derivative work, modifying aspects of the song, adding to it.

If what you added to it is sufficiently creative as well, that is protected by copyright too.

It’s tricky in the context of AI, but when you’re creating all of it, there’s really no issue with the human authorship.

It’s just about the scope of the new version, like the scope of protection for the new version, and it may be technical in the context of actually bringing a litigation or bringing a copyright infringement claim, understanding what version of the song you want to bring the claim on.

That’s a whole other thing and may be complicated.

But in terms of what is protected by copyright, as long as it is original work of authorship, and it’s fixed, it likely has copyright protection if it’s sufficiently creative.

The reason I bring that up is because one of the findings of the report is that in the view of the office, if you have a work that is partially generated by AI and partially created by a human, that it would be possible to separate those portions of the work and apply copyright to the human generated portion without also extending that coverage to the AI generated part of the work.

And for musicians, that brings up a whole host of questions because of the different ways you can put these ideas into expression.

So if I, for example, created a song, I wrote the lyrics and the melody, but then used AI to create the backing track because I don’t have the capability to do that part myself.

You know, in that situation, like, am I creating a split copyright in my work?

We wouldn’t consider it a split copyright.

And here’s kind of where existing principles are analogous and helpful in how we think about the scope of copyright.

So we wouldn’t necessarily, for example, consider if, you know, you took an AI generated output and add it to it, we wouldn’t consider that a derivative work.

But derivative work authorship is helpful and the scope is analogous.

So for a derivative work, when you create something based on a preexisting work, say you have permission to make a different sort of addition or modification to a song and you, you know, create it, you don’t necessarily have copyright protection in what another person created if you didn’t create that preexisting work.

What you have is protection over what you added to the work or the modifications you made to that work, as long as it’s sufficiently creative.

So an example, maybe you, you know, you hear a song and you want to create an acoustic version that maybe adds a bridge or something like that.

So you can take that preexisting song and what, the scope of your protection is what you’ve added to it.

And that’s kind of similar to taking an AI generated output, purely generated by AI, so there’s no copyright protection in that output, but you are rearranging it in some way.

Say maybe you’re taking the loops and changing them, you’re creating a different melody with that output.

Your selection and arrangement and that is what you’re adding to the work, and that is the scope of your copyright protection.

So your protection doesn’t necessarily extend to the pieces that are purely generated by AI, just the same as your protection wouldn’t extend to pieces of a pre-existing work that you didn’t create.

Your protection extends to what you created.

So it’s sort of the same sort of thing.

It’s not a new concept, but it’s just the fact that it’s AI that makes it a bit different.

And we should clarify here that using AI tools in the creation process is not, it’s not exclusionary or disqualifying to the copyright.

Like if you’re using AI plugins to help mix a song or things like that, but you’re still the one primarily creating the work, that it’s fine to use the AI tools as part of the creation, but not the generation of the ideas.

Is that a good way of putting it?

Right, the way we kind of say it is that when you’re using a tool, you’re using a tool to enhance your creativity.

Using AI in a generative sense is when you’re using it to displace human authorship.

When you’re using the AI to mimic a human created thing, it’s actually displacing expression.

So it’s actually creating the melody or creating the lyrics, creating the aspects that would be copyrightable.

But using it to enhance, maybe to change a sound, there’s an example that we have in the report where Randy Travis actually, he is not able to sing because of his health condition, but he was able to use AI.

Someone else sung the melody that they created.

Someone else sung lyrics that a human created, and they used an AI plugin to change the singer’s voice to sound like Randy Travis.

In that case, it’s not like there’s any displacement of human creativity.

They’re just using AI to change the way a voice sounds.

That would be an example of a tool use as opposed to generative use.

That’s a really cool example too.

Yeah.

It’s really awesome they were able to do that.

AI can provide really interesting capabilities and it can enhance creativity in really amazing ways.

We don’t think that using it by itself should preclude expression.

It’s all about how you’re using it.

For what it’s worth, I agree with the findings of the report.

I think human authorship is important and I think that’s something that ought to be protected.

At the same time, I don’t really see how realistically this is enforceable.

I mean, how do you prove that something is AI or more difficult?

How do you prove that something wasn’t AI generated?

You know, it just enters into all of these tangled scenarios that to me anyway make it seem really difficult to enforce this position.

What’s your take on that?

Well, one thing that I think is relevant, of course, it’s important to note that Copyright Office is an enforcement agency, right?

But we do provide guidance and one of the things that we do is we administer the country’s registration system, their copyright registration system.

So one thing that I think is relevant is that for US works, you need to be able to either register your claim to copyright or be refused registration in order to bring a claim of copyright infringement, either in the federal district court or with the Copyright Claims Board.

And under our current rules, if a work contains both AI generated material and sufficient human authorship to support a claim to copyright, the applicant has to disclose that they have included AI generated content.

That is more than a de minimis amount in the work.

So if the content could stand alone be, you know, considered copyrightable, then they would have to exclude that AI generated material.

Now, the office generally accepts the facts that are stated on the application.

We accept what the applicant says, unless there is something that’s implausible or in conflict with something else in the application.

But knowingly making a false representation to the office isn’t legal.

One, and it also puts your registration at risk.

If later it’s found out that, you know, there is a misrepresentation in the application and the applicant provided that knowingly.

So that’s something that’s important, at least for the protection of your registration, which opens the door to you being able to enforce your copyright.

Is that the, going back to what you said in your answer about the de minimis amount, is that sort of the standard, like if you took away the AI portion of a submission, would it stand on its own?

Is that sort of the benchmark for determining whether or not the amount of human authorship is sufficient?

That’s not necessarily the benchmark for, that’s sort of the opposite.

That’s not necessarily the benchmark for determining human authorship, but in the registration context, is the benchmark for determining whether you have to exclude AI generated content.

So you don’t have to exclude, if you use AI in your work or you use AI as part of your work, you don’t have to exclude AI generated content in your registration application in all instances.

It’s only where, if the AI generated content is more than a de minimis amount, meaning it’s standing alone, the AI generated output could be considered a copyrightable work in and of itself.

Then you would have to exclude that in your registration application.

But for determining copyrightability, again, the amount of AI generated content you’re using isn’t the deciding factor.

It’s really the presence of sufficient human authorship.

That’s what gets copyright protection.

So that’s what folks should be looking for.

So it’s more about the significance of the contribution than necessarily the amount.

Right.

It’s really…

There’s no bright line rule, and it really is case by case.

But in the report, we took a look at three different kinds of contributions that people make.

And that would be the prompts.

Lots of people will prompt the system.

And the question was, is that sufficient human authorship?

And we concluded that it’s not based on the generally available technology.

We don’t think that prompts alone provide the sufficient amount of human control to make the users the authors of the outputs that AI generates.

We also looked at people providing expressive inputs.

In some systems, you’re able to actually feed in something that you created, be it lyrics or a melody or an image.

You can actually feed that into the system.

Oftentimes, that will be perceptible in the output that this system generates.

We looked at that as a human contribution that could be considered sufficient human authorship to the extent that human authorship is perceptible in the output.

Then we also looked at modifying and arranging AI generated content and found that selection and arrangement that can be a sufficient human contribution so long as that selection and arrangement is sufficiently creative to warrant copyright protection.

So, there are a number of ways that humans can contribute to an output, either in the generation of the output or after the output has already been generated, that could be considered sufficient human authorship.

Going back to what you said about prompts and prompts not being copyrightable, I think that probably would be the most surprising finding for a lot of people listening, because I think it feels like a lot of work to go through and refine these prompts over and over and over again, to get the result that you’re looking for.

That feels like, I don’t know, just in terms of man hours, that it feels like that gives you, I guess, more ownership or that you’ve put more work into, the creation.

But I also think if you are putting your own original ideas into the prompts, right?

Like if you’re putting lyrics that you’ve written into the prompt and saying, create something like this, or if you’re putting an audio recording that you’ve created into the prompt and asking for creation based off of that, then it starts to feel like those prompts do become something that would be worthy of protection.

That’s a really good question.

Because it brings up a couple of aspects of copyright law that are really important.

One, that copyright law doesn’t really acknowledge what we call sweat of the brow.

So just because something was an endeavor to generate or difficult or took a lot of steps, those things aren’t really acknowledged in terms of determining sufficient human authorship and then also in terms of determining whether there’s sufficient creativity.

It really is about what is perceptible in the output.

Regarding prompts, given the current technology, we found that prompts really alone don’t provide sufficient human control to make users the authors of the output.

And we concluded that based on our analysis of how the existing technology works, of course, currently prompts really function as instructions that do convey those unprotectable ideas.

And while they may communicate a user’s desired expressive elements, they don’t necessarily control how an AI system processes the request and generates the outputs.

Outputs can include different elements that weren’t really requested.

They can throw in a number of different things, and you’re like, where did that come from?

And we have an example of a prompt that we inserted into a system, and this cat that comes out that has a number of random things that just never was even discussed.

And so that’s an example of how a user doesn’t necessarily control the expressive elements that are contained in the output.

But you make a good point that people use prompts in a number of different ways.

They may include things that they’ve created and ask the AI to replicate it in some way or modify it in some way, which is why we separated out pure prompting, text prompting from maybe providing an expressive input.

When you’re providing an expressive input and you’re intending for that input to be perceptible in some sort of way in the output, there’s a different conversation at that point because you are in some way controlling the expressive elements of the output, especially when the output actually has the expressive elements that you input into the system.

Then in that case, we would say there may be sufficient human authorship to the extent that it’s perceptible and to the extent that it’s sufficiently creative, then you may have copyright protection over the elements that you contributed to the output.

Yeah, it gets complicated real fast, doesn’t it?

It does.

It does.

It will continue to, I think.

I think it’ll continue to get more and more complicated.

I think for a lot of musicians, creatives in general, there’s a bubbling level of anxiety, let’s call it.

Right?

About where things are going to go.

I think that kind of attitude is always felt among creatives anytime there’s sort of a new technology introduced.

I think AI has perhaps generated more fear amongst creatives than other technologies of the past because of the scale on which it’s able to generate content.

And I think, you know, looking at the report, one of the main arguments that it brings in favor of human authorship is the principle of wanting to encourage creativity.

Like that’s one of the goals in the Constitution for copyright is providing incentives for creators.

But at the same time, at least within the music industry, because works don’t necessarily have to be registered in order to be exploited, it seems like the flood of content is happening regardless of what the office says is or isn’t copyrightable because you don’t have to necessarily copyright your AI song to upload it to the Internet, or to sell copies, or to put it on a video, or whatever.

To be able to use that song, it may not have copyright protection, but that doesn’t stop the flood of content competing with creators and making it hard for their voices to be heard.

That’s not on the Copyright Office.

That’s just the reality of the situation, I think.

But I guess what I’m getting around to is this.

Do you think AI is so different from other kinds of technologies that have been introduced in the past?

Or is this just another one and a long example of things that have come and changed society, and then we’ll figure out how to adapt to it and move on to the next thing?

Well, I think in terms of what copyright has had to encounter and respond to, there’s always been, as you’ve said, changes in technology that have impacted creatives and have impacted how works are created and exploited.

And so copyright, I think, has always responded, and there have been evolutions, you know, not necessarily in the bedrock principles, but in how they’re applying.

Of course, our report recommends no real change to existing law.

We do think existing law is sufficient to handle the developments that AI is bringing.

But I do think it’s important to just monitor the changes.

And so that’s something we’re doing.

We’re monitoring how AI is impacting different industries, including the music industry.

And that may bring about different things.

We did make specific recommendations in the first part of our report, talking about deep fakes and digital replicas.

In this part, we acknowledge that, you know, particularly as it relates to our administration of the blanket license available under Section 115 of the Copyright Act, under that compulsory license, just as background, royalties for digital phone record deliveries of non-dramatic musical works are paid into a pool for the mechanical licensing collective to divide and distribute to copyright owners.

And we note that, you know, as part of our monitoring, parties have attempted to obtain royalties for streams of AI-generated content.

But when notified, we were responsive and clarified that musical works that lack human authorship are not eligible for the blanket license under Section 115 and should not be getting royalties.

So things like that are important because oftentimes existing law does have these rules and restrictions, and awareness of them helps you strategize and protects against some of these shifts and you just don’t know how much is responsive to the shifts until you are made aware of them.

And so one of the important aspects or roles of this report, of our series of reports, is to make people aware of what existing law says and how it applies potentially to the shifts that AI is bringing.

And is that the main function of the report, kind of bringing it back to the beginning of our discussion?

Is the purpose of the report to inform the general public?

Is it meant to advise judges on how to rule on cases?

Is it for Congress?

Is it all of the above?

I mean, how do you hope this report gets disseminated and incorporated into the conversation?

Yeah, well, the report, we intend for it to offer guidance to Congress, the courts, other federal agencies, and the public.

And it’s really based on our expertise as an agency that administers the Copyright Act.

And it fulfills our statutorily prescribed duty in the Copyright Act.

It says that we are to conduct studies regarding copyright.

And so how people use it is really up to them, but we really do think that it provides helpful tools, helpful background, helpful context for all sorts of parties.

And as a policy report, we’re really, you know, we’re just really providing our expert analysis, you know.

We are looking forward to working with the Congress.

You know, there are times that we do provide guidance to courts and other agencies, so we’re always available to do that.

But yeah, we’re hoping that the report really helps people understand copyright law and copyright law as it applies to things like AI-generated outputs.

So as a normal person, which I host a podcast, so I don’t know that I could be considered normal, but as somebody who doesn’t have a legal or a technical background, what should I be watching for as all of this plays out, as the technology develops, as the law develops?

What are the things to be keeping an eye on?

What do you think are the most significant questions that still need to be resolved?

I think you should be keeping an eye out on the additional guidance that the Copyright Office will be providing.

Excellent plug.

I have to say, it’s, I think, really important as a creative to be engaged with agencies that impact your, your creativity, your industry, how you monetize, how you engage, you know, in the world as a creative.

And if I were a creative person, I would be tapped in to what the Copyright Office is doing.

And I would go to copyright.gov and sign up to receive our announcements because these, you know, big policy reports aren’t the only things that the Copyright Office is doing.

We are issuing regulations and a big part of issuing new regulations is the notice and comment period.

We have to notify the public and they can actually tell us, this is not going to work for us or we hate this rule.

You should revise it like this.

And we read all of those comments and are responsive to them.

And so I think it’s important as a creative to lean in to what the Copyright Office is doing, look out for our additional reports.

Part three of our report will be coming in the coming months and we’ll address different topics, not necessarily copyrightability, new topics related to liability and things of that nature.

So I would sign up to receive notices from the Copyright Office.

That’s what I would probably do.

Well, and we’ll share links to how to do that in the episode notes.

There are some downloadable things on the USCO website and mailing lists you can sign up for and things like that.

So we’ll link to that in the show notes.

Any final thoughts before we let you go?

Any words of wisdom, anything that you wanted to point out before we go?

Well, I’ll reiterate that our report isn’t meant to warn people against the dangers of AI or anything like that.

It acknowledges that people are using AI in really interesting ways that furthers their creativity and pushes them to different limits or different ways of expression that are really helpful to us as a society.

But that human authorship is a better act requirement and copyright is human-centric.

So it’s really important to acknowledge that and to encourage that expression as well.

And then I would also want to leave with your audience of musicians that we have on the Copyright Office’s website a number of really targeted resources specifically for musicians, how to get paid as a musician, how copyright impacts musicians.

There’s a number of really targeted content specifically for musicians.

So I would encourage your audience to go to our engaged pages and look for the content for musicians, for songwriters.

It will really be helpful and demystify a lot of things that you may be concerned about.

Well, thank you again for coming on the show.

This was such a fascinating conversation and the work that that y’all are doing in the office, I think, is tremendously valuable for those of us in the creative industries, trying to navigate all of this.

So thank you again.

And like I said, we’ll share some of those resources in the show notes so people can check it out.

And we’ll be excited for part three.

Awesome.

Thank you so much for having me.

This is awesome.

Thanks again.

Take care.