In the Matter of, )

TRANSCRIPT OF PROCEEDINGS

HERITAGE REPORTING CORPORATION

Official Reporters

1220 L Street, N.W., Suite 206

Washington, D.C. 20005-4018

(202) 628-4888

[email protected]

In the Matter of: )

)

INTELLIGENCE AND VISUAL ARTS )

LISTENING SESSION )

)

Pages: 1 through 140

Place: Washington, D.C.

Date: May 2, 2023

Heritage Reporting Corporation

(202) 628-4888

LIBRARY OF CONGRESS

UNITED STATES COPYRIGHT OFFICE

In the Matter of: )

)

INTELLIGENCE AND VISUAL ARTS )

LISTENING SESSION )

)

Suite 206

Heritage Reporting

Corporation

1220 L Street, NW

Washington, D.C.

Tuesday,

May 2, 2023

The parties met remotely, pursuant to the notice,

at 1:00 p.m.

ATTENDEES:

MARIA STRONG, Associate Register of Copyrights

and Director of Policy and International

Affairs

MARK GRAY, Assistant General Counsel

EMILY LANZA, Counsel

NICHOLAS BARTELT, Attorney-Advisor

DAVID WELKOWITZ, Attorney-Advisor

JORDANA RUBEL, Assistant General Counsel

JALYCE MANGUM, Attorney-Advisor

J. SCOTT EVANS, Adobe

BEN BROOKS, Stability AI

ALICIA CALZADA, National Press Photographers

Association

SARAH CONLEY ODENKIRK, Cowan, DeBaets, Abrahams &

Sheppard

KARLA ORTIZ, freelance concept artist

CURT LEVEY, Committee for Justice

REBECCA BLAKE, Graphic Artists Guild

JAMES GATTO, Sheppard Mullin

ALEX RINDELS, Jasper AI

PAUL REINITZ, Getty Images

LUC BOULET, Professional Photographers of America

HEATHER WHITNEY, Morrison & Foerster

DANIEL TAKASH, The Niskanen Center

Heritage Reporting Corporation

(202) 628-4888

ATTENDEES: (Cont'd.)

ZARA VARIN, Dual Wield Studio

DANIEL GERVAIS, Vanderbilt University Law School

JAMES SILVERBERG, American Society for Collective

Rights Licensing

TOM LOCKLEY, Grey Owl Audio

MATTHEW CUNNINGHAM, Cunningham Concept Design

BRIAN FRYE, University of Kentucky College of Law

NETTRIC GASKINS, freelance artist

PHUC PHAM, Freelance Solidarity Project

ANKIT SAHN, Ajay Sahni Associates

JEFFREY SEDLIK, PLUS Coalition

PATRICIA SIGMON, artist/art director

DELANIE WEST, Be Super Creative

Heritage Reporting Corporation

(202) 628-4888

P R O C E E D I N G S 1

(1:00 p.m.) 2

MR. GRAY: Hello, everyone. Thank you very 3

much for joining us today. We are going to start our 4

session now. 5

Welcome to the United States Copyright 6

Office's Listening Session on Artificial Intelligence 7

and the Visual Arts. Today, we are going to be 8

discussing a variety of issues in the visual arts 9

space. 10

My name is Mark Gray, first off. I'm an 11

Assistant General Counsel here in the Office of the 12

General Counsel. 13

Before we start our first panel, I would 14

like to introduce Maria Strong for opening remarks. 15

Maria is an Associate Registrar of Copyrights, as well 16

as the Director of Policy and International Affairs 17

here at the U.S. Copyright Office. 18

Maria? 19

MS. STRONG: Thanks, Mark, and welcome, 20

everybody, to the Copyright Office's Public Listening 21

Session on Artificial Intelligence and Visual Arts. 22

In copyright law, works of visual arts are broadly 23

defined as pictorial, graphic, and sculptural works. 24

Some examples include two-dimensional and three-25

Heritage Reporting Corporation

(202) 628-4888

dimensional works of fine graphic and applied art, 1

photographs, prints and art reproductions, maps, 2

globes, charts, diagrams, models, and technical 3

drawings, including architectural plans. 4

Because the visual arts include a wide 5

variety of works, today, we will ask broad questions 6

to facilitate discussion across each participant's 7

area of expertise. 8

It's likely that almost everyone on this 9

webinar has seen various images that deep learning 10

text-to-image models can produce based on text 11

prompts. We've heard concern from artists and 12

photographers about what the training and deployment 13

of these models might mean for their livelihoods and 14

their industries both in terms of the input of their 15

own images into these models, as well as the 16

excitement and concerns related to the outputs. 17

And the purpose of our session today is to 18

discuss these issues. We want to hear how the public 19

is thinking about policy issues raised by these 20

technologies. 21

To begin to address the copyrightability and 22

registration issues raised by works generated using AI 23

tools, the Office recently issued new registration 24

guidance in mid-March. That guidance makes clear that 25

Heritage Reporting Corporation

(202) 628-4888

applicants have a duty to disclose the inclusion of 1

AI-generated content in works submitted for 2

registration. It outlines how to do so, how to update 3

pending applications, and how to correct the public 4

record on copyright claims that have already been 5

registered without the required disclosure. 6

There was a lot of interest in today's 7

event. Unfortunately, we were not able to accommodate 8

all requests to speak. But this is not the last 9

chance to share your views on AI with the Copyright 10

Office. As we've said before and we'll say again, 11

there are two more listening sessions happening later 12

this month. And down the road, we will be requesting 13

written input through a public notice of inquiry. 14

Please visit our website, copyright.gov/AI, for more 15

information and resources on our AI initiative. 16

Finally, we thank our panelists in advance 17

for contributing to today's discussion and 18

conversation. This is a complex topic and a deeply 19

personal one for all our panelists, whether they are 20

users or developers of AI technology, artists whose 21

works help train that technology, or creators 22

contemplating how AI will affect their careers. We 23

are all looking forward to a thoughtful and respectful 24

dialogue. 25

Heritage Reporting Corporation

(202) 628-4888

Let me turn the mic back to Mark Gray to 1

outline the various logistics for today's session. 2

Thank you. 3

MR. GRAY: Thank you very much, Maria. 4

So, as a quick reminder, before we get into 5

specifics, today's listening session is the second of 6

a series of listening sessions that we are doing here 7

at the Copyright Office going through the end of May. 8

Each of our sessions is going to look at different 9

topics, different types of works, and, as a result, is 10

going to have different panelists and may even use 11

different formats. 12

So, after today, we have two more sessions 13

scheduled. There is a session on May 17, Wednesday, 14

which will be focusing on audiovisual works, which 15

would include movies and video games. And our final 16

session will be on May 31, which will focus on musical 17

works and sound recordings. 18

The purpose behind these listening sessions 19

is to inform the Office's overall AI initiative, so 20

some of the questions our panelists raise may be ones 21

that we seek to explore further in written comments 22

later this year. So please keep in mind that while 23

there are a handful of my colleagues here from the 24

Heritage Reporting Corporation

(202) 628-4888

Office is in the audience and is listening, and all of 1

this is going to help inform our work. 2

The schedule for today, the session format 3

is going to be two panels of different sets of 4

speakers, followed by a third segment where a set of 5

additional speakers will get the chance to share brief 6

remarks. 7

We are making a video recording of this, 8

both of this session as well as the other three. We 9

are trying to get those online within three weeks of 10

each session taking place, so please keep your eyes 11

peeled for that if you have any friends or colleagues 12

who don't have the opportunity to watch the session 13

today. 14

Before we get started, a few Zoom 15

housekeeping notes. If you are a panelist who is not 16

speaking at the current session, please keep your 17

camera and microphone off and on mute. And then, 18

likewise, if you are a panelist, please keep your 19

camera on and be ready to go off of mute when you're 20

speaking. 21

We will be recording the session today. As 22

I mentioned, the recording will try to go up in about 23

three weeks from today. And we have enabled Zoom's 24

transcription functionality for those of you who are 25

Heritage Reporting Corporation

(202) 628-4888

interested in following along with captions. 1

The way we're going to do the first panel is 2

we're going to start with a brief introduction and 3

short statement by each of the panel participants if 4

they so desire. We'd like you to try to keep those to 5

two minutes. We're going to be keeping an eye and the 6

moderators may need to cut you off if it goes a little 7

long just so we can keep everything on schedule. 8

After those introductions and brief remarks, 9

we're going to do a moderated listening session. The 10

panelists have received a set of broad questions in 11

advance. Those are meant to prompt and guide a 12

discussion, but panelists and participants are welcome 13

to share any other relevant perspectives or 14

experiences they think are important for the Copyright 15

Office to hear. 16

If you are a panelist, please try to use 17

Zoom's Raise Hand functionality, and we will try to 18

call on you in the order that you raise your hands 19

just to keep the conversation organized. 20

Please keep in mind this is a listening 21

session and not a debate, so there will be other 22

opportunities in the future for people to engage more 23

directly with competing views. But the purpose today 24

is really to help the Office air out a variety of 25

Heritage Reporting Corporation

(202) 628-4888

ideas and issues and perspectives for us to guide our 1

own thinking. 2

As a final note, I see we have some 3

questions in the Q&A from the audience. 4

Unfortunately, this is a listening session for the 5

participants. We are unable to accommodate audience 6

questions. So thank you so much for your interest. 7

Please keep your eyes out on our website for future 8

public participation and comment opportunities, but we 9

cannot take any comments today, unfortunately, from 10

you. 11

With that, I'm going to hand it over to our 12

moderators for the first session, Emily Lanza and Nick 13

Bartelt. Emily is a Counsel in our Office of Policy & 14

International Affairs. And Nick is an Attorney-15

Advisor in the Office of the General Counsel. 16

Emily, the mic is yours. 17

MS. LANZA: Thank you, Mark, and welcome, 18

everyone. We will begin with introductions in the 19

order as stated on the agenda. 20

So, first up, Scott with Adobe, would you 21

like to go ahead? 22

MR. EVANS: Sure. Thank you for having me 23

today. My name is J. Scott Evans, and I'm Senior 24

Director and Associate General Counsel at Adobe. 25

Heritage Reporting Corporation

(202) 628-4888

For over four decades, Adobe's mission has 1

been to empower our creative community with the tools 2

that they need to express their imagination and earn 3

their livelihoods in areas like photography, art, 4

music, filmmaking, and design. AI and generative AI 5

specifically have profound impact in these areas, so 6

we really wanted to make sure that we, as we harness 7

the power of this new technology, we're doing so in a 8

way that empowers creators. 9

Last month, Adobe launched its generative AI 10

technology, Adobe Firefly. Firefly's initial text-to-11

image model was designed to be commercially safe; that 12

is, it was trained on images licensed from Adobe 13

stock, openly licensed content, and content in the 14

public domain. We want our tools to be good for 15

enterprises and the creative community. 16

When it comes to copyright, we know that the 17

issue of training is one where the creative community 18

has concerns. For this reason, through a technology 19

Adobe developed called Content Credentials, we're 20

enabling artists to attach a do-not-train tag that 21

will travel with their content wherever it goes. With 22

industry adoption, it is our hope that this tag would 23

prevent the training on content that has the do-not-24

train tag. We are working with generative AI 25

Heritage Reporting Corporation

(202) 628-4888

technology companies to respect these tags. 1

From an output standpoint, for much of our 2

professional creative community, generative AI serves 3

as the front door to the creative process. They're 4

changing the image. They're adding colors, editing, 5

adding elements. They're adding their own human 6

expression to the work. So we need a way, a 7

transparent way, to track this expression. 8

Here, Content Credentials can function much 9

like an ingredients label. They'll show you where the 10

image came from and what edits have been made to it. 11

So, for generative AI, it gives the creator a way to 12

show that they started with an AI generated image but, 13

most importantly, to demonstrate the human creativity 14

they brought to the work. 15

Finally, Content Credentials will bring a 16

level of transparency that is much needed with the age 17

of generative technology. Adobe is automatically 18

attaching a Content Credential to images created with 19

Firefly to indicate the image was generated by AI. 20

We're working to drive transparency standards so that 21

together we can deploy this technology responsibly in 22

a way that respects creators and our communities at 23

large. 24

I thank you for having me today, and I look 25

Heritage Reporting Corporation

(202) 628-4888

forward to engaging further on these issues. 1

MS. LANZA: Thank you, Scott. 2

Next up is Ben with Stability AI. 3

MR. BROOKS: Well, thank you to the 4

I lead public policy for Stability AI, a 6

leading developer of open source AI models designed to 7

unlock humanity's potential. These include, as many 8

of you know, the latest versions of Stable Diffusion, 9

which is a model that takes a text prompt from users 10

and translates that prompt into a new image. Users 11

can interact with these models either through a hosted 12

service, like an app or an API, or they can freely 13

use, integrate, and adapt the open source code subject 14

to our ethical use license. Stability has also 15

launched a number of other image models as well as a 16

suite of language models. 17

Stable Diffusion is a type of latent 18

diffusion model. So these models use content to learn 19

the relationship between words and visual features, 20

not unlike a student at a public gallery. Based on 21

this acquired understanding and with creative 22

direction from the user, these models can help to 23

generate new works. In this way, AI should be 24

understood as a tool to help artists express 25

Heritage Reporting Corporation

(202) 628-4888

themselves. It's not a substitute for the artist. 1

Instead, AI can help to simplify the 2

creative process. It can help existing creators boost 3

their productivity as part of a wider workflow. And 4

it can also help to lower barriers to entry for people 5

who simply don't have the resources or training to 6

realize their creative potential today, including 7

those with life-altering injuries or disabilities. 8

As with other assistive technologies, from 9

paintbrushes to cameras to editing software, the user 10

ultimately determines the content and use of any 11

generated images. 12

I do want to acknowledge today the depth of 13

feeling on these issues among creators and developers. 14

AI is changing rapidly, and we understand that it can 15

feel highly disruptive. We welcome a dialogue with 16

all members of the creative community about the fair 17

deployment of these technologies. And through the 18

session today, I can share some details about how 19

we're working towards that goal in practice, whether 20

that be through new training techniques, authenticity 21

standards, and best practices for things like opt-22

outs. So thank you very much. 23

MS. LANZA: Thank you, Ben. 24

Next up is Alicia. 25

Heritage Reporting Corporation

(202) 628-4888

MS. CALZADA: Hi. I'm Alicia Calzada. I'm 1

the Deputy General Counsel for the National Press 2

Photographers Association. 3

First, I really appreciate the invitation 4

for NPPA to be a part of this event. This is very 5

important to us and our members. 6

Before I was an attorney, I was a 7

photojournalist for 20 years. And through NPPA, we 8

serve the -- we are the nation's premier organization 9

for visual journalists. We serve still photographers 10

and videographers, and, frankly, most of our members 11

do both. 12

Some of the things we do include working to 13

support the First Amendment rights of visual 14

journalists. We also advocate for their copyrights 15

and for greater copyright protection and for a strong 16

is the industry standard among visual journalists, and 18

that is, of course, a very important piece of what I 19

hope we'll get into today. 20

NPPA has a few concerns related to AI. The 21

first, of course, is copyright protection for 22

photographers against unauthorized use of their images 23

and unauthorized copying. So we do support 24

legislation that accomplishes that. 25

Heritage Reporting Corporation

(202) 628-4888

For us, it's not just about money. As I 1

mentioned, we care about ethics. And for visual 2

journalists, their reputation is one of their most 3

valuable assets. And so the right to control the use 4

of their image and protect against misuse is very 5

important. When their photos are used in an unethical 6

manner, it impacts them. It impacts the entire 7

industry, frankly. 8

And we also think that news consumers have a 9

right to know the source and the authenticity of the 10

content that they're consuming, the news that they're 11

reading and watching. 12

Finally, a concern that we are monitoring is 13

that journalists, like many photographers, do use 14

technology in some ways that are, in fact, quite 15

ethical, and so we're watching what the Copyright 16

Office is doing as they frame the question of what is 17

copyrightable. We understand that something entirely 18

AI-created might not be copyrightable, but we want to 19

make sure that in making policy we don't risk the 20

copyrightability of photographs that for generations, 21

frankly, have used special timers and triggers, such 22

as the kind of things a sports photographer or a 23

nature photographer might use. 24

So those are some of the things that are 25

Heritage Reporting Corporation

(202) 628-4888

sort of on our radar related to AI, and we're 1

definitely looking forward to this session and to 2

continued conversation on these issues. 3

MS. LANZA: Thank you, Alicia. 4

Next up is Sarah. 5

MS. ODENKIRK: Hi. Thank you very much for 6

including me in today's conversation. My name is 7

Sarah Conley Odenkirk, and I'm a partner with Cowan, 8

DeBaets, Abrahams & Sheppard. I co-head the Los 9

Angeles office and also the Art Law Practice Group. 10

My deep involvement in the implications of 11

emerging technology and visual arts goes back almost 12

30 years with my dedication to representing artists 13

and also working to establish public policy around 14

visual art in public spaces. The combination of these 15

elements in my practice has positioned me well to do a 16

lot of advising around the impact and implications of 17

blockchain technology and now AI, both from the 18

standpoint of the impact on creators as well as on 19

public policy. 20

It's crucial to maintain the focus on the 21

impact the technologies have on artists and artists' 22

abilities to continue to create and innovate. This 23

becomes complex when we cannot easily determine when, 24

where, and how potential copies and other copyright 25

Heritage Reporting Corporation

(202) 628-4888

infringements may be occurring. 1

In order to explore possible futures, we 2

need to start by breaking down the processes used in 3

AI into their component parts as the analysis will 4

likely suggest different solutions at different 5

points. Figuring out fair, enforceable, and 6

economically sound solutions to questions raised at 7

the point of training AIs will differ from determining 8

how to treat the output artists coax from these 9

platforms. We also must distinguish the generic 10

generative process employed by commercial AI platforms 11

from the more bespoke process of generative art as a 12

medium employed by artists. 13

So I urge the Copyright Office to consider 14

the impacts on artists in light of the new structures 15

that are made possible with these technologies and to 16

adhere or even strengthen principles underpinning the 17

artists' innovation and creativity with the market 19

forces. It may be time to consider more than just 20

guidance, more than just analyzing what is considered 21

copying or protectable. 22

I would love to see the Copyright Office 23

take the lead in championing technical solutions that 24

meaningfully address the way content is scraped, 25

Heritage Reporting Corporation

(202) 628-4888

sourced, and used, and explore realistic ways to track 1

IP rights and compensate creators. 2

MS. LANZA: Thank you, Sarah. 3

Next up is Karla. 4

MS. ORTIZ: So back in April of last year, I 5

saw a website called Weird and Wonderful AI Art. It 6

had the names of many of my peers alongside work that 7

looked like theirs but wasn't. I thought it's just a 8

new experiment. Well, I asked my peers whose name I 9

saw on that website, and no one knew what this was and 10

no one had been asked to be a part of it. 11

So we tried to reach out to the folks who 12

were running the website, folks who are also selling 13

merchandise that looked like the studies they were 14

doing. We asked them to please take down the work of 15

the artists who didn't want to be there. Instead, we 16

got ghosted. I thought this was small enough to 17

ignore, but little did I know this would be my first 18

encounter with generative AI. 19

Fast-forward to September-ish, and larger 20

generative AI models like Midjourney and Stable 21

Diffusion are now mainstream. So I research again, 22

and I am horrified by what I found. Almost the 23

entirety of my work and the work of almost every 24

artist I knew was scraped and utilized to train these 25

Heritage Reporting Corporation

(202) 628-4888

for-profit models. I was mortified that this was done 1

without anyone's consent, credit, or compensation, 2

that once AI models are trained on our work, our work 3

could not be forgotten, and that generative AI 4

companies were even encouraging users to use our full 5

names to generate imagery that can look like our work. 6

For example, Polish artist Greg Rutkowski, 7

who in December, between Midjourney, Stability AI, and 8

the very problematic un-Stable Diffusion, Greg's name 9

had been used as a prompt for image generation about 10

400,000 times. 11

If there is one thing I want everyone to 12

remember is that this hyped technology is entirely 13

fueled by the ill-gotten data it is trained upon. It 14

is unlike any tool that has come before as it is an 15

innovation that uniquely consumes and exploits the 16

innovation of others. No other artistic tool is like 17

this, and I know. I've used most of them. 18

In my opinion, to reward tech that relies on 19

the proceeds of theft by granting it copyright would 20

just add insult to injury. 21

Oh, also, my name is Karla Ortiz. I am an 22

award-winning artist who works in film, game, TV, 23

galleries, you name it. I worked on Magic: The 24

Gathering, Guardians of the Galaxy Volume 3, Loki, and 25

Heritage Reporting Corporation

(202) 628-4888

most notably known for my design of Dr. Strange for 1

the film adaptation. 2

I am also a plaintiff in one of the first 3

class-action lawsuits against generative AI companies, 4

specifically, Midjourney, DeviantArt, and, yes, 5

Stability AI. Hi. 6

MS. LANZA: Thank you, Karla. 7

Next up is Curt. 8

MR. LEVEY: Hi there. I'm Curt Levey, 9

President of the Committee for Justice. We're a 10

nonprofit that focuses on a variety of legal and 11

policy issues, including intellectual property, AI, 12

tech policy. 13

There certainly are a number of very 14

interesting questions about AI and copyright. I'd 15

like to focus on one of them, which is the 16

intersection of AI and copyright infringement, which 17

some of the other panelists have already alluded to. 18

That issue is at the forefront given recent high-19

profile lawsuits claiming that generative AI, such as 20

DALL-E 2 or Stable Diffusion, are infringing by 21

training their AI models on a set of copyrighted 22

images, such as those owned by Getty Images, one of 23

the plaintiffs in these suits. 24

And I must admit there's some tension in 25

Heritage Reporting Corporation

(202) 628-4888

what I think about the issue at the heart of these 1

lawsuits. I and the Committee for Justice favor 2

strong protection for creatives because that's the 3

best way to encourage creativity and innovation. 4

But, at the same time, I was an AI scientist 5

long ago in the 1990s before I was an attorney, and I 6

have a lot of experience in how AI, that is, the 7

neural networks at the heart of AI, learn from very 8

large numbers of examples, and at a deep level, it's 9

analogous to how human creators learn from a lifetime 10

of examples. And we don't call that infringement when 11

a human does it, so it's hard for me to conclude that 12

it's infringement when done by AI. 13

Now some might say, why should we analogize 14

to humans? And I would say, for one, we should be 15

intellectually consistent about how we analyze 16

borrow from precedents we know that assumed human 18

authorship than to invent the wheel over again for AI. 19

And, look, neither human nor machine learning depends 20

on retaining specific examples that they learn from. 21

So the lawsuits that I'm alluding to argue 22

that infringement springs from temporary copies made 23

during learning. And I think my number one takeaway 24

would be, like it or not, a distinction between man 25

Heritage Reporting Corporation

(202) 628-4888

and machine based on temporary storage will ultimately 1

fail maybe not now but in the near future. Not only 2

are there relatively weak legal arguments in terms of 3

temporary copies, the precedent on that, more 4

importantly, temporary storage of training examples is 5

the easiest way to train an AI model, but it's not 6

fundamentally required and it's not fundamentally 7

different from what humans do, and I'll get into that 8

more later if time permits. 9

But I think the good news is that the 10

protection for creators of the works that are used as 11

training examples can and will come from elsewhere, 12

where the generated output is too similar -- 13

MS. LANZA: Thank you, Curt. I'm going to 14

have to -- sorry, I'm going to have to cut you off 15

there. 16

MR. LEVEY: Okay. Sure. 17

MS. LANZA: But we'll have time during the 18

question to continue. 19

MR. LEVEY: Sure. 20

MS. LANZA: Rebecca, would you like to go 21

ahead, please? 22

MS. BLAKE: Yes, I'm happy to go ahead. And 23

I'm apologizing in advance for the construction that 24

has just started up outside my window. 25

Heritage Reporting Corporation

(202) 628-4888

My name is Rebecca Blake. I'm the Advocacy 1

Liaison for the Graphic Artists Guild. The Graphic 2

Artists Guild is a trade association representing the 3

interests of visual artists other than photographers, 4

illustrators, designers of all stripes, production 5

artists, cartoon and comic book artists, animators and 6

others. 7

Our mission is to protect the economic 8

interests of our members, and in that vein, we've long 9

advocated for greater copyright protections for 10

individual artists, fair labor and trade practices, 11

and policy which supports small creative businesses. 12

We welcome this opportunity to weigh in on AI 13

generative technologies. 14

Our members include artists who have 15

embraced generative AI in the creation of their own 16

original works and artists who, for various reasons, 17

have not adopted the use of generative AI or, in fact, 18

see it as a threat to their livelihoods. 19

While we support the ethical, legally 20

compliant development of AI as a tool for visual 21

artists, we have serious concerns about the copyright 22

and ethical questions raised by AI generative 23

technologies. These include the inclusion of 24

copyrighted material in the training data sets without 25

Heritage Reporting Corporation

(202) 628-4888

permission or notification, which we see as a 1

protections for artists' works as inputs into AI 3

generative platforms, the unfair competition in the 4

marketplace resulting from the massive generation of 5

images which may ape existing artists' styles or 6

replicate artists' works, confusion with the 7

registration of works containing AI-generated 8

material, and existing barriers to the affordable 9

registration of works created by visual artists other 10

than photographers. And I hope we can go more into 11

this in the subsequent questions. 12

MS. LANZA: Thank you, Rebecca. 13

And last but not least, James, would you 14

like to conclude the introductions, please? 15

MR. GATTO: Yes. Thank you. Hi. My name 16

is James Gatto. I'm honored to have the opportunity 17

to share some views here today on the important 18

office of Sheppard Mullin, where I lead our AI 20

practice. I've been an IP attorney for 35 years. I'm 21

also a member of the ABA IP Section AI Machine 22

Learning Task Force, but the views expressed today are 23

solely my own. 24

I've been doing work with AI for about two 25

Heritage Reporting Corporation

(202) 628-4888

decades, but, like others, I'm seeing a significant 1

increase in that work due to the meteoric rise of 2

generative AI. Clients have a lot of questions. 3

I applaud the Copyright Office's initiative 4

to issue preliminary guidance on the examination of 5

applications involving AI-generated content. I know 6

there's great debate in the community on these 7

guidelines, on authorship issues with AI, the level of 8

human involvement needed, and issues with joint 9

authorship. I hope these listening sessions will 10

result in the Copyright Office keeping an open mind on 11

whether to tweak their guidance and provide further 12

clarity on some of the procedural aspects of the 13

guidance. 14

Some of the issues for which clarity would 15

be helpful are the following: 16

When does the level of detailed input or 17

prompts by a human provide sufficient basis for the 18

output to be deemed original intellectual conceptions 19

of the author and therefore protectable? 20

What is the relevance of predictability in 21

the authorship analysis? This concept was part of the 22

basis for the Kashtanova decision but does not appear 23

in the guidance. 24

What level of detail is needed to comply 25

Heritage Reporting Corporation

(202) 628-4888

with the duty of disclosure regarding use of AI? 1

What is the copyrightability of a work where 2

a human uses AI-generated content as inspiration art 3

but does not copy it? 4

And what are the criteria for determining if 5

AI-generated content is more than de minimis such that 6

it should be explicitly excluded from the application? 7

AI is a powerful tool, and to promote the 8

constitutional mandate, the Copyright Office should 9

develop policy that promotes rather than deters its 10

use. As a result of the guidance in the Kashtanova 11

decision, at least many companies that rely on 12

companies, artists, and many others, are concerned 14

about using generative AI and in some cases restrict 15

or limit employees' use of it. That's not consistent 16

with the goal of promoting the use of technology. 17

So we hope through these sessions we get to 18

a happy medium where artists' rights can be respected 19

and tools can be used to facilitate the creation of 20

their expressive works. 21

MS. LANZA: Great. Thank you, James, and 22

thank you all for those introductions, and welcome 23

again. 24

So, to begin our discussion, I'll start off 25

Heritage Reporting Corporation

(202) 628-4888

with a question. How is the training of artificial 1

intelligence models affecting your field or industry? 2

What should the Copyright Office know about the 3

technology's use of training materials when 4

considering the copyright issues related to training? 5

And, also, please be specific in your 6

answers in terms of kind of which part of the visual 7

arts ecosystem you're talking about. 8

So, great, I already see hands. So, Karla, 9

you're the first on my screen. Can you please go 10

ahead. 11

MS. ORTIZ: Yeah, absolutely. So, 12

basically, the training of artificial intelligence is 13

already affecting my particular field of 14

entertainment, specifically concept art, illustrators, 15

anything that requires a painter. We're already 16

seeing the effects of these tools, you know, in our 17

industries. 18

Something to consider is the training of 19

these tools is very important. When considering these 20

tools, you can't just focus on the output. You have 21

to see the entire process as a whole. And as a whole, 22

these tools, you know, particularly, you know, some of 23

the tools around here, like Stability AI and 24

specifically LAION, under the pretext of research, 25

Heritage Reporting Corporation

(202) 628-4888

gathered 5.8 billion text and image data from across 1

the Internet to train various AI/ML models for 2

commercial purposes. Again, it was trained upon for 3

research and then switched immediately for commercial 4

purposes. Technologists like Andy Baio call this, you 5

know, loophole data laundering. 6

But another thing that's important to note 7

is that this was done without consent, credit, or 8

compensation. The work of myself and almost all of my 9

peers are in those data sets, again, and also our 10

names are, you know, encouraged to be utilized as 11

prompts so that users can get something that mimics or 12

feels similar to our work. 13

I personally am of the belief that the work 14

generated by these models is impressive only because 15

it is based upon the works of artists. And, again, 16

this was done without consent. And we're not even 17

talking about all the issues when it comes to 18

propaganda, identity theft, and so on. 19

One of the things that I will say as well 20

that the copyright should consider, and I won't take 21

much longer of anybody's time so people can have their 22

say, as one of the few artists in this panel, you 23

know, there's various others as well, but I'm a 24

teacher, and I can tell you that anthropomorphizing 25

Heritage Reporting Corporation

(202) 628-4888

these tools to equate it as human-like is a fool's 1

errand. I've spoken to countless machine learning 2

experts, such as Dr. Timnit Gebru, such as Professor 3

Ben Zhao, and they all agree that it's not what's 4

happening. This is a machine. This is mathematic 5

algorithms. You cannot equate it to a human. 6

And to further add and to give the 7

perspective of an artist, an artist doesn't look at a 8

bunch, like 100,000, images and is able to generate, 9

like, hundreds of images within seconds. An artist 10

cannot do that. Yes, I have my influences, but it's 11

not the only thing that goes into my work. My life, 12

my experiences, my perspective, my technique, all of 13

that goes into the work. 14

Furthermore, something that I feel like a 15

lot of people miss in these discussions is technical 16

artistry, and one of the hardest things you can do 17

ever in the arts is be able to successfully mimic 18

another artist's style or another person's work. It's 19

the hardest thing. I consider myself masterful. I 20

can't even do it. In fact, it's so rare that they 21

even have documentaries on Netflix showcasing the few 22

artists that can successfully mimic, let's say, a 23

Leonardo da Vinci. 24

And depending on what that artist does with 25

Heritage Reporting Corporation

(202) 628-4888

that successful mimicry, if they sell it or if they do 1

anything commercial with it, you know, that could 2

potentially be called forgery. So I don't know why -- 3

MS. LANZA: Thanks, Karla. Sorry to 4

interrupt. 5

MS. ORTIZ: Oh, yeah. No, no, no, it's 6

totally cool. 7

MS. LANZA: But, yeah, we just have a 8

couple -- 9

MS. ORTIZ: No, no, no. Totally great. 10

Just wanted to drop that in. 11

MS. LANZA: All right. Thank you, Karla. 12

James, you're next on my screen. Please go 13

ahead. 14

MR. GATTO: Great. Thank you. I'll try to 15

be brief. 16

So, I mean, obviously, one of the core 17

issues with training AI model on copyright-protected 18

content is whether it's infringement and/or if fair 19

use applies, and, largely, that's going to be a fact-20

specific question depending on the details. I think 21

that to the extent there, you know, are any policy 22

considerations or guidance the Copyright Office, you 23

know, can provide in that, that might be helpful, but 24

there is a pretty significant existing body of law on 25

Heritage Reporting Corporation

(202) 628-4888

that, kind of the broad legal test. 1

I think some of the areas that should be 2

considered, consistent with what Scott said from 3

Adobe, there's a lot of tools out there that can be 4

used that help mitigate the problem, and whether those 5

tools should be mandated or, you know, some other role 6

the Copyright Office can play with respect to them 7

would be helpful. 8

Should AI tool providers be required to be 9

more transparent on the content they use to train 10

their models? I think that's an important issue. 11

Should there be greater use of tools that 12

prevent AI from using copyrighted works to train AI? 13

Similar to how robots.txt works to prevent search 14

engines from indexing certain web content. The 15

technology is there and some of the concerns can be 16

abated if these tools become mandated or just widely 17

used. 18

And the last point I'm going to make is 19

maybe not directly relevant to visual arts, but just, 20

you know, there's other content that using it is not a 21

problem because it's licensed, whether it's open 22

source software that's being used to train AI code 23

generators or like images that are under a permissive 24

license, like Creative Commons, as long as there's no 25

Heritage Reporting Corporation

(202) 628-4888

prohibition on commercial use, the use of it may be 1

permissible, but the question is then are there 2

licensed compliance obligations that need to be met 3

and, you know, whether and how those should be dealt 4

with in these contexts. Those are just a few of the 5

issues I think would be helpful to consider. 6

MS. LANZA: Thanks, James. 7

Alicia, you’re next on my screen. Please go 8

ahead. 9

MS. CALZADA: Thanks. That's a really 10

interesting point about things like Creative Commons 11

that actually do have conditions to, you know, what 12

seems like on the surface an unlimited license, but, 13

actually, there are things you have to do in order to 14

earn that license. 15

Back to the question about how it affects 16

our industry, the primary concern, as I mentioned 17

earlier, in our industry really is an ethical one, and 18

journalists rely on copyright as a means of 19

controlling how their work is used. 20

And it's one thing to say, isn't it neat 21

what this computer can do while you're, you know, just 22

goofing off with friends or doing research or that 23

kind of thing, but, you know, when these works start 24

being used to create deepfakes or images that are used 25

Heritage Reporting Corporation

(202) 628-4888

to promote civil unrest, there are a lot of ways that 1

news images can be abused through this kind of a 2

process in very, very negative ways. 3

And the journalism industry really is 4

concerned about where that's going to go and how it 5

impacts the industry as a whole. You know, we already 6

have editors who have for decades, you know, paid very 7

close attention, you know, to work that comes in to 8

ensure the quality of the sourcing and that kind of 9

thing. But, on some level, there's things out there 10

in the world that we worry about people seeing and 11

thinking is journalism when it really isn't. 12

MS. LANZA: Thanks, Alicia. 13

Next, Curt, you're next on my screen. 14

Please go ahead. 15

MR. LEVEY: Sure. Let me first briefly 16

finish what I was saying about the good news for 17

protection for creators, despite the fact that I do 18

think it's getting harder and harder to distinguish 19

between what humans do and machines do. But, 20

regardless of how they're trained, where the generated 21

output is similar to one of the examples in the 22

training data or really any preexisting work, it's a 23

derivative work or an outright copy, and the licensing 24

requirements for derivative works need to be as 25

Heritage Reporting Corporation

(202) 628-4888

strictly enforced as for non-AI works. 1

And then, second, and some of the others 2

have alluded to this, since the source of the training 3

data is typically unlicensed data, I should say 4

publicly available data or web scraping, we need 5

strict enforcement of the website or database terms of 6

service. And Mr. Evans mentioned a do-not-train tag. 7

That's a good example. 8

Also, when you said what should the 9

bit more about temporary storage and why that's not 11

fundamentally required. 12

Generative AI learns from a very large 13

number of examples, and so does a human artist or 14

author. The artist or author is not born with that 15

ability here. He or she learns from countless 16

examples of art, photography, music, written works, et 17

cetera, and, you know, more and more the human views 18

those examples on a website. The human may purposely 19

make copies of the examples he used. And even if he 20

doesn't purposely do it, his computer makes a 21

temporary copy as he views the image, reads the 22

written work, et cetera. Yet we all dismiss that 23

copying as fair use, you know, if we even acknowledge 24

it at all. 25

Heritage Reporting Corporation

(202) 628-4888

So what AI training does is not very 1

different. For convenience sake, the examples are put 2

in a database, which a learning algorithm cycles 3

through, and that is temporary copying. But humans, 4

like I said, often copy for convenience sake as well. 5

And once the AI cycles through the examples in 6

training, the examples can be thrown away. 7

The trained model, consisting of millions or 8

billions of weights, analogous to the synaptic 9

connections in the human brain, retains no copies of 10

the training examples. Human memory, on the other 11

hand, does remember at least some specific examples. 12

So, in some sense, there's less of an infringement 13

danger with AI than humans. But, to be fair, neither 14

humans nor AI depend on retaining the specific 15

examples they learn. 16

So, again, the problem with relying on the 17

temporary copy argument is that it's not really 18

necessary. You could train the AI model by having it 19

scroll through the very same images or written works 20

that the human learns from. In fact, the AI model 21

could learn from, you know, data being relayed by a 22

mobile robot that, you know, visits art galleries 23

throughout the nation. Someday, you know, that may be 24

how it's done. Think Google maps. Either way, my 25

Heritage Reporting Corporation

(202) 628-4888

point is that hanging one’s hat on temporary copying 1

is skating on very thin ice. 2

MS. LANZA: Thank you, Curt. 3

Next up is Rebecca. Please go ahead. 4

MS. BLAKE: Yeah, gosh, there's just so much 5

to unpack from that previous answer. 6

Very quickly -- 7

MS. LANZA: Oh, Rebecca, you muted yourself. 8

Can you unmute? 9

MS. BLAKE: I'm so sorry. 10

MS. LANZA: That’s all right. 11

MS. BLAKE: Very quickly, some of our 12

members completely eschew using AI image generators. 13

They're concerned about the ethical concerns with the 14

way the image data sets were built. They're worried 15

about copyrightability. And they're worried about 16

exposing their clients to infringement. 17

Other members of the Graphic Artists Guild, 18

in fact, use AI image generators. For the most part, 19

we're hearing that they use it for ideation but not 20

for the creation of completed works, or they use it to 21

generate elements of a much larger work, for example, 22

to create background graphics. 23

We do have one member who, in fact, has a 24

career in AI generative for an AI image generator as 25

Heritage Reporting Corporation

(202) 628-4888

part of that new generation that has achieved a 1

career. However, we've been trying to gauge job loss, 2

job creation, job loss, and we're in very, very, very 3

early days to be able to do that. It’s something we 4

need to start tracking now that these generators have 5

been out almost a year. However, we do hear a lot of 6

anecdotal evidence of job loss. It's in particular 7

sectors. That is hampered by the fact that many of 8

the artists working in these areas, in fact, sign NDAs 9

or are reluctant to go on the record discussing 10

projects that they’ve lost because they're afraid of 11

retaliation. They work in a very small industry. 12

Of our members who do use generative AI, one 13

member stated that he was able to take on larger 14

projects with a smaller workforce. So that does 15

indicate that generative AI permits a streamlining and 16

less hiring of artists. And another member stated 17

that because she uses AI generative technology, she 18

was able to cease contracting to a certain number of 19

designers. So, again, that indicates a benefit to one 20

member but at the loss of others. So that's speaking 21

to the job market. 22

But I want to address two other things. 23

First of all was this idea, this equivalency of the 24

way machine learning works to the way human learning 25

Heritage Reporting Corporation

(202) 628-4888

works. This is a false equivalency for a very, very 1

major reason. When a human learns to draw, they will 2

ape, they will copy the styles or the works, existing 3

works of illustrators they admire. This is very 4

common in the learning practice. 5

But there are ethical considerations, 6

practices that professional illustrators follow that 8

take them away from the wholesale copying of either a 9

style or, in fact, of an image itself. This does not 10

occur with machine learning. The machine is not 11

driven by a creative process, a desire to develop 12

one's own style, one's own mark, one's own creativity. 13

It's simply reiterates a style that it has been 14

learned on. So there is no equivalency in the 15

outputs. 16

The second thing I just wanted to touch on 17

very briefly was this idea that there can be tags or 18

codes or metadata which is embedded in images which, 19

in fact, permits one to track whether or not an image 20

can be used for inclusion in a data set, whether it 21

can be ingested into a platform, et cetera. 22

There's a huge issue with that, which is 23

that Section 1202 of the Copyright Act permits the 24

removal of copyright management information, including 25

Heritage Reporting Corporation

(202) 628-4888

metadata, if that removal is done knowingly or with 1

reasonable grounds to know it will induce, et cetera, 2

et cetera, infringement. 3

We believe that section of the Act needs to 4

be modified so the removal of CMI, including metadata, 5

without permission of the copyright holder is 6

prohibited, regardless of whether or not it's done 7

knowingly to permit infringement. We see metadata and 8

CMI as key to being able to protect artists’ works in 9

an AI environment, but that failure in Section 1202 10

needs to be addressed. 11

MS. LANZA: Thanks, Rebecca. 12

Next up is J. Scott. Would you like to go 13

ahead? 14

MR. EVANS: Sure. You know, at Adobe, we 15

believe that if AI is done right, if this is done 16

right, it benefits both creators and consumers of 17

content because it does nothing but amplify human 18

creativity and intelligence. It doesn't replace it. 19

And so what we see as a major issue here is 20

that creators now have limited resources to attribute 21

their work, especially when generative AI comes into 22

play. One of the important things we need to do as a 23

collaboration with artists and technology is to put 24

creators at the forefront of this technology. 25

Heritage Reporting Corporation

(202) 628-4888

Creators want control over their work that 1

is used in generative AI training, and we need to give 2

them the tools in order to make those decisions. We 3

know many creators that are very excited about this 4

technology and want their creativity to be used in 5

training these models. They are very excited about 6

them. But we do understand that there's a segment of 7

the community that is not excited and wants the 8

ability to prevent the use of their art in training, 9

and they should have an ability to do so. 10

And that's the reason Adobe has developed 11

the Content Credentials. We worked very hard with 12

setting up an open-source industry standard with the 13

Coalition of Content Provenance and Authenticity, the 14

C2PA. It's an open standard that platforms and 15

hardware manufacturers can put into their products 16

that will allow you to put these Content Credentials 17

that will surface them to users and developers of AI 18

technology so that those cues can be followed, and 19

that's something that we think that's very important. 20

We also think there may be technology where 21

artists could harness this technology by training 22

models based solely on their own style or brand and 23

then commercializing that and having that technology 24

and understanding that there are different ways that 25

Heritage Reporting Corporation

(202) 628-4888

this technology can be used is very important. 1

One of the ways the Copyright Office, I 2

think, can help in this is to encourage industry to 3

adopt these open standards that will give artists the 4

ability and tools in order to identify whether they 5

want to participate or don't want to participate and 6

encouraging that kind of proactivity among the 7

companies that are developing this technology to give 8

artists a tool to control their creative work. 9

MS. LANZA: Thank you. 10

And, Sarah, please go ahead. 11

MS. ODENKIRK: Thank you. I think I'm going 12

to be reiterating a number of things that have already 13

been said, but, first, I'd like to say that, you know, 14

I think that there's a lot of reasons to be concerned 15

about AI in general. There are big issues, big global 16

ethical issues that definitely need to be addressed. 17

Unfortunately, I think we do need to somewhat separate 18

those questions from these questions that we're 19

talking about with regard to copyright issues in order 20

to parse through things. Otherwise, we're going to 21

very quickly get sidetracked with, you know, scary 22

potential future possibilities, which I don't think we 23

should ignore, but we need to separate that out of the 24

Heritage Reporting Corporation

(202) 628-4888

There's clearly a lot of potential in 1

addressing some of the training issues through 2

metadata and through some of the tools that Mr. Evans 3

was speaking about, as well as some other tools that 4

have been developed and people are looking to in order 5

to protect the content. 6

And I'd like to underscore what Rebecca said 7

with regard to Section 1202 and, you know, really 8

needing to be concerned about the way in which the 9

metadata can be taken off of content and thereby 10

allowing it to be misused and really keeping creators 11

from being able to track that data. 12

So I think that the final point that I want 13

to make is with regard to paying attention to the 14

purpose of the use that the images are being scraped 15

and collected for. And if what we're talking about is 16

using those images for the purpose of creating a 17

commercial venture, a commercial product that's to be 18

used to earn money, that's a very different use than 19

artists looking at images and using tools in order to 20

generate art, and while they're obviously connected, I 21

think we need to look at them very separately in terms 22

of figuring out what policies and laws and approaches 23

we can take to protect creators in the front end of 24

that process. 25

Heritage Reporting Corporation

(202) 628-4888

MS. LANZA: Thank you, Sarah. 1

So, before we move on to Question 2, I just 2

want to make sure everyone had a chance to speak. 3

Ben, would you like to add anything to 4

Question 1 before we move on? 5

MR. BROOKS: Yeah, thank you, Emily. I 6

think just on this question of impact, I think these 7

go to a broader set of issues around style and 8

authenticity, and I do have remarks on training 9

specifically for later. But I want to reiterate what 10

I said at the beginning, which is that we see AI as a 11

tool to help artists express themselves, but it's not 12

a substitute for the artist. 13

That said, we obviously support efforts to 14

improve creator control over their public content, and 15

we're focusing those efforts in three areas in 16

particular. So one is around access to content. 17

Today, already, data sets like LAION-5B respect 18

protocols like robots.txt that indicate whether a 19

website consents to automated data aggregation. But 20

we're also developing new ways to help creators 21

qualify the use of that public content for AI 22

training. 23

So one of the things we've done is we've 24

committed to honoring opt-out requests from creators 25

Heritage Reporting Corporation

(202) 628-4888

in the next wave of Stable Diffusion releases. And 1

going forward, I think this was a point alluded to by 2

J. Scott, we're also exploring new standards for opt-3

outs so that the opt-out metadata will travel with the 4

content wherever it goes subject to some of the 5

problems that have been flagged just a little while 6

ago. 7

The second area we're focusing on is 8

authenticity of content. So we're working to 9

implement content authenticity standards, like C2PA, 10

with the Content Authenticity Initiative so that users 11

and platforms can better identify AI-assisted content. 12

By distinguishing AI-assisted content, these standards 13

can help to ensure that users apply an appropriate 14

standard of scrutiny in their interactions with that 15

content. It can help to limit the spread of 16

disinformation through social media platforms. And, 17

ultimately, it can also help to protect human artists 18

from unfair mimicry or passing off. 19

And the third and final point I just want to 20

make is the work that we're doing to improve the 21

quality of data sets. So, for example, by improving 22

the diversity and reducing the duplication in training 23

data, we can help to mitigate the risk of things like 24

overfitting, which is where the system erroneously 25

Heritage Reporting Corporation

(202) 628-4888

overrepresents certain elements of a particular image 1

from the data set. So, for example, if you've only 2

ever seen sunsets, you might think that the sky is 3

always orange. 4

In addition, by improving diversity in our 5

data sets, we can be more representative of diverse 6

cultures, language, demographics, and values, all of 7

which can help to mitigate the risk of bias in those 8

outputs. 9

So I think the final point on this question 10

I'll add is, you know, we believe the community will 11

continue to value human-generated content, right? We 12

carry a complex digital camera in our pockets 13

everywhere we go, yet we continue to value painting. 14

Likewise, Photoshop didn't destroy photography. We 15

have machines that can run faster than athletes, but 16

we continue to place a premium on sport. And the same 17

will be true of visual arts in the post-AI creative 18

economy, particularly when we have some of these 19

content authenticity standards in place. 20

MS. LANZA: Thank you, Ben. 21

I'll turn it over to my co-moderator, Nick, 22

for the next question. Thank you. 23

MR. BARTELT: Thanks, Emily, and thanks, 24

everyone. 25

Heritage Reporting Corporation

(202) 628-4888

I think the focus of the first question that 1

Emily had asked was a little bit more on the input and 2

the training. So I think we'll shift the discussion a 3

bit to ask, what should the Copyright Office be aware 4

of regarding how these AI systems, and some of you 5

have already touched on this, how these AI systems 6

generate works of visual art? 7

And then, as sort of a subquestion there, 8

I'll ask, because I know we have limited time, is 9

that, you know, are there any copyright considerations 10

that vary based on the type of visual works that are 11

at issue there? 12

So I see James's hand first. 13

And I know, Karla, we had lost you a minute 14

ago, but I see you're on there too, so we'll go 15

through. 16

Go ahead, James. 17

MR. GATTO: Great. Yeah, so, obviously, the 18

operation of the AI tools vary, and each case is fact-19

specific. We recognize it's a challenge for the 20

but there are a number of fact patterns that are 22

common. And I think what would be helpful, one thing 23

that would be helpful, is kind of like the Patent 24

Office did with patent eligibility guidelines, if the 25

Heritage Reporting Corporation

(202) 628-4888

that are used in generative AI that they would deem to 2

be copyright-protectable, that would be helpful. 3

The other thing is, just to take one use case, so 4

I do a lot of work with NFTs as well, and there's a 5

lot of concern around the use of generative AI for 6

some of these NFT projects. If I create NFTs that 7

represent images and, for example, I specify two 8

images of dogs, each having a different collar that I 9

designed and different colors, and I use AI just to 10

generate the permutations of those artistic elements 11

that I created under my control, saying, produce those 12

permutations, AI should just be deemed a tool, even 13

though it's output from a generative AI. 14

The question is, as you scale that up and 15

maybe some of the parameters are a little bit looser, 16

where does the line get crossed between it being my 17

creative expression and it being too much input from 18

generative AI? That's one practical use case that 19

we've seen in a number of these NFT projects. 20

So I think that there's many other -- I'll 21

be brief so others can talk -- but there's many other, 22

I'd say, common use cases that we're seeing, and I 23

think that any input or guidance or examples that the 24

Heritage Reporting Corporation

(202) 628-4888

to, you know, assist those in trying to figure out 1

where the line is and recognizing that, you know, 2

there are fact-specific differences. Thank you. 3

MR. BARTELT: Okay. Thanks, James. 4

Actually, because, you know, Karla had her 5

hand up before we had switched the question, we'll go 6

Karla and then Ben and then Curt. 7

So go ahead, Karla. 8

MS. ORTIZ: Wholeheartedly appreciate it. 9

So something that I think the Copyright 10

Office should be aware of regarding how AI systems 11

generate work of visual art, there's been some talk at 12

the idea of, like, whether these models copy, 13

remember, memorize, whatever the word, overfit, 14

whatever the word really is. 15

Something I'd like the Office to know is 16

that studies are being done concerning these issues. 17

For example, there's research from the University of 18

Maryland and the University of New York. They did a 19

study that found diffusion models generated high 20

fidelity reproductions, which is basically plagiarism, 21

at an estimated 1.88 percent, and it is estimated by 22

these researchers to be higher. 23

Cursory numbers, but let's take a look at 24

like Lensa AI, which uses Stable Diffusion, has about, 25

Heritage Reporting Corporation

(202) 628-4888

you know, this was around December, has about 25 1

million downloads, and gives users about 50 trials 2

each. At 1.88 percent, that's potentially 23,500,000 3

generated images that could be similar to training 4

data. And we see this consistently with, like, for 5

example, Steve McCurry's famous Afghan girl can be 6

perfectly plagiarized by these tools, as it happened 7

in Midjourney. 8

Another thing that I'd like, you know, folks 9

to consider as well is, like, the music, the, you 10

know, Stable Diffusion -- no offense, Ben -- but 11

Stable Diffusion already has made the case for us. I 12

mean AI companies have already made the case for us. 13

For example, Dance Diffusion was a music 14

program developed by Harmonai, which has links to 15

Stable Diffusion, and as they trained their model, 16

they stayed clear from copyrighted data and only did 17

public domain. And one of the things that they quoted 18

on is, because diffusion models are prone to 19

memorization and overfitting, releasing a model 20

trained on copyrighted data could potentially result 21

in legal issues. Why was this done for the music 22

industry but not the visual industries? 23

And this also goes to the point to why opt-24

out is inefficient, regardless of what it should -- 25

Heritage Reporting Corporation

(202) 628-4888

the standard should be opt-in, because opt-out places 1

an undue burden on people who may not know the 2

language, who may not be online, who may not even, 3

like, know what's going on. Companies cannot just 4

arbitrarily grab our copyrighted works, our data, and 5

just say, this is ours, and then later on we have to 6

remove ourselves, which is why opt-out is important. 7

The other thing that's really important, 8

again, is transparency. And I know that Adobe, you 9

know, is mentioning this, but, for example, we need to 10

really know what, for example, open license works 11

mean. We really need to know and have an open data 12

set to see exactly what it means so that licensors, 13

you know, can actually, like, fulfill their licenses. 14

And -- 15

MR. BARTELT: All right. Thanks. Thanks, 16

Karla. So I want to make -- 17

MS. ORTIZ: Oh, I forgot the guidance on 18

you, but you guys go on ahead. If we have time later, 19

let's do it. I'm sorry. 20

MR. BARTELT: Okay. Yeah. I just want to 21

make sure we get through everybody in our remaining 10 22

or so minutes. 23

MS. ORTIZ: Yeah. 24

MR. BARTELT: So I'll go back to you now, 25

Heritage Reporting Corporation

(202) 628-4888

Ben. I think you mentioned, what should the Office be 1

aware of of how these AI systems generate works? 2

MR. BROOKS: Yeah, look, I think it's 3

important that we properly characterize the training 4

process, right? These models are not, as is sometimes 5

being described, you know, a collage machine or a 6

search index for images. These models review pairs of 7

text, captions, and images to learn the relationships, 8

again, between words and visual features, right? So 9

that could be fur on a dog or ripples on water or 10

moods like bleak or styles like cyberpunk. And with 11

that acquired understanding and with creative 12

direction from the user, those models can then help 13

the user to generate new works. So, in this sense, 14

training is, we believe, an acceptable and 15

transformative use of that content. 16

But there are some good instinctive examples 17

as well. Stable Diffusion notoriously struggled to 18

generate hands, right? So it produced three-finger 19

hands or 12-finger hands because it doesn't know that 20

a hand typically has five fingers. And it isn't 21

searching a database of the many images with hands, 22

right? Instead, it has learned that a hand is a kind 23

of flesh-colored artifact typically accompanied by a 24

number of sausage-like appendages, right? 25

Heritage Reporting Corporation

(202) 628-4888

And that all has real implications for how 1

we should think about AI training and generation. In 2

other words, these models are using knowledge learned 3

from reviewing those text-image pairs to help the user 4

generate a new work. They're not using the original 5

images themselves. And those images are nowhere in 6

the AI model. 7

MR. BARTELT: Great. Thanks, Ben. 8

And, Curt, we'll go to you next. 9

And just so everyone knows, we have about 10

two minutes for the remaining five people with their 11

hands up before this panel ends. So, if you have any 12

concluding remarks, you know, kind of work them into 13

whatever you have to say here. Thank you. 14

MR. LEVEY: Well, let me answer the 15

question, but also, in a sense, these are concluding 16

remarks. 17

A couple of the panelists feel strongly that 18

machine learning is not like what humans do, so let me 19

say more about why I believe it is very similar. 20

The human brain consists of neurons connected by 21

synapses of various strength. So, when a human sees 22

an example, those synaptic strengths are slightly 23

modified. Modification takes place slowly. But, you 24

know, given a lot of examples, there's a lot of 25

Heritage Reporting Corporation

(202) 628-4888

modification and learning. That is how we learn. 1

Neural networks consist of artificial neural 2

networks connected by artificial synapses. When the 3

AI is shown an example, the synaptic strengths or 4

weights are slightly modified, and, again, over time, 5

the modifications add up, and we call that learning. 6

And I realize this similarity is difficult to see 7

because, for one thing, humans are very invested in 8

seeing themselves as being very different than 9

machines, you know, myself included. 10

So my advice to the Copyright Office is to 11

look past that difficulty and use the similarity to 12

your advantage to guide your policy development rather 13

than reinventing the wheel. 14

And I would also suggest that your position 15

that only product of human creativity can be 16

copyrighted is something that, you know, you should 17

think about. I don't take a position either way, 18

whether copyright should be granted to AI-generated 19

works. But the Copyright Office should keep in mind 20

that it likely will have to recognize AI authorship as 21

AI becomes more sophisticated and the philosophical 22

distinction between human and machine creativity 23

becomes harder to sustain. 24

You know, already generative AI can pass a 25

Heritage Reporting Corporation

(202) 628-4888

Turing test of sorts in that it's hard to tell the 1

difference between human creation and machine 2

creation, and that’ll be all the more so once 3

artificial general intelligence becomes a reality. 4

And I'll stop there. 5

MR. BARTELT: Okay. Thank you, Curt. 6

Sarah? 7

MS. ODENKIRK: Thank you. And apologies for 8

noise. I'm in the corner of a conference room in the 9

midst of a big conference trying to find a quiet space 10

here. So just a couple of things. 11

I think that, you know, even if we assume 12

that an AI platform has preemptively licensed all of 13

the content that it's using to train the AI and that 14

everything underlying is fine and licensed, we really 15

need to look at the way in which the artist is using 16

that content and how they're interacting with the AI 17

platform in order to come up with their eventual 18

artwork. And that's something that is going to have 19

to be explored in a little bit more detail rather than 20

just having really broad blanket rules about what can 21

come out of that AI interaction and collaboration 22

where artists are using it as a tool. 23

To go back to something that James brought 24

up with regard to generative artists, I think that 25

Heritage Reporting Corporation

(202) 628-4888

point really warrants a little bit more exploration as 1

well because we do have a whole group of artists and 2

they've, of course, received quite a lot more 3

attention lately, especially with the NFT marketplaces 4

opening up. And that has to do with artists who are 5

generating their own algorithms using their own data 6

sets to put the artwork in and generating work. It 7

may be that it's quite a bit generated by the AI 8

platform and not by the artists themselves except that 9

the underlying content is, in fact, created by the 10

artist, including the algorithm. So these are really 11

different ways of looking at tools that kind of 12

emanate from the same place but end up being used in 13

different ways and creating different results. 14

One just last point is that I think that 15

oftentimes we're going to have to still look at and 16

rely on traditional means for figuring out whether an 17

output is substantially similar to either somebody 18

else's work or that, you know, there are copyright 19

infringements in that output. This is where we're 20

going to have to consider things like the fact that 21

many artists don't even register their copyrights. 22

And so whether they have access to the court system 23

because of that, it's really going to impair their 24

ability to seek some sort of resolution of those 25

Heritage Reporting Corporation

(202) 628-4888

issues. 1

And this is where maybe we can take a look 2

at the Copyright Office's Small Claims Board as a 3

place to perhaps think about resolving some of these 4

issues because we are going to have a tremendous 5

number of artist-related issues that maybe don't rise 6

to the level of needing to go to federal court but 7

definitely are going to need to be resolved in a 8

meaningful way for those artists. 9

MR. BARTELT: Thank you, Sarah. 10

J. Scott? 11

MR. EVANS: Thank you. I mean, at Adobe, we 12

look at the laws that exist today. We look at the 13

guidance that you provided. And after many hours of 14

thinking through this, we view text prompts as 15

conceptually an idea, an idea that is put into a 16

machine that will give you many different expressions 17

of that idea. So the expression seems to be being 18

completely generated by a computer, and under current 19

law, that is not copyrightable. 20

But, as we said in our opening statement, 21

most of the creatives that we have talked to don't use 22

generative AI as the end product. What they do is 23

they use it as inspiration or a jumping-off point, and 24

then they take that product and they do other things 25

Heritage Reporting Corporation

(202) 628-4888

to it, just like they would if they took a piece of 1

art that was in the public domain and they created a 2

new work that had elements of creativity and human 3

creativity to that. So that's what we think is 4

important. 5

I think what would be helpful is some 6

additional guidance that you heard from our first 7

speaker from the Copyright Office that gave examples 8

of what you're looking for. And I hate to say this, 9

but it probably would be more helpful if we had a new 10

form that was specifically designed for AI-created 11

works that drilled down on these particular issues so 12

that some of the concerns that we hear from filmmakers 13

and from photographers who use some sort of AI in the 14

generation of the work, but not to the extent or in 15

the way that we're talking about AI that generates the 16

work itself, is used. 17

And so I just think that, you know, we need 18

specific guidance. We need malleability from the 19

just heard, we may need to change that guidance. We 21

need the ability to do that and to follow these issues 22

very closely. 23

But, at Adobe, you know, we think that we're 24

headed in the right direction based on the current 25

Heritage Reporting Corporation

(202) 628-4888

laws that exist today, the guidance that you all have 1

provided. 2

MR. BARTELT: Great. That's good to hear. 3

Thank you, Scott. 4

And, Alicia, you are next. 5

MS. CALZADA: Thanks, and I'll be brief. 6

I just, I hear efforts to minimize the value 7

of the work that are input into these systems, and I 8

just want to make sure that the Copyright Office 9

always keeps in perspective that without the input, 10

which involves copying works in their entirety, 11

there's no output. And so we can't minimize the value 12

of the works that are used to generate AI. That would 13

be inappropriate. Thanks. 14

MR. BARTELT: All right. Thank you, Alicia. 15

And just to wrap up, I think we have about 16

two minutes left, maybe a minute each. We'll go to 17

both Karla and Ben, and then we'll wrap up this 18

session. So go ahead, Karla. 19

MS. ORTIZ: Thank you. So, again, I just 20

want to reiterate that current practicing leading 21

experts in the AI/ML machine-learning industry warn of 22

equating machines to humans. We should heed their 23

warning. 24

Furthermore, no matter how extensive the 25

Heritage Reporting Corporation

(202) 628-4888

prompt will be, it still relies on the training. And 1

machine-learning companies want to automate prompts, 2

as said various times by Stability AI’s CO themselves. 3

So I don't know how much of a standard that will be if 4

it will eventually be automated. 5

And, lastly, honestly, for me, I'm concerned 6

that AI-generated material will gain copyright, that 7

the proceeds of theft will be rewarded. And, 8

potentially, as someone who is likely to be affected 9

in a whole industry that I see from the inside being 10

affected by these tools, I'm very concerned. Thank 11

you. 12

MR. BARTELT: Thank you, Karla. 13

And, Ben, with our remaining minute or so 14

left, go ahead, please, and give us your closing 15

thoughts. 16

MR. BROOKS: Yeah, just a last point on 17

the guidance. 18

MR. BARTELT: Sure. 19

MR. BROOKS: Look, we acknowledge and accept 20

that there is a threshold of authorship below which a 21

work with negligible human input may not qualify for 22

registration. I think our concern is that in the 23

guidance and in the caption of the decision, the 24

Office does not necessarily account for all of the 25

Heritage Reporting Corporation

(202) 628-4888

ways in which human input might rise above that 1

threshold, and some of them have been discussed 2

previously on this session. 3

A creator, to get a desirable input, may 4

fine-tune that model using their own existing content 5

to evoke their own existing style. They might provide 6

detailed prompts that narrowly define the range of 7

possible outcomes. And they also may, as J. Scott 8

mentioned, use the image as a jumping-off point. And 9

they may refine that initial image many times using 10

traditional editing software or further prompting. 11

Any one of these factors may qualify that work for 12

registration. And so a user who has clear expressive 13

intent and takes steps to steer these tools in a 14

particular direction should be able to register their 15

work. 16

I will just finally add that, again, we want 17

to make it clear that training is not about stitching 18

images together, it's about learning hidden 19

relationships. But we do understand that there are 20

other ways to improve creator control of their 21

content, and, hopefully, some of the things we flagged 22

today can be considered by the Office in your future 23

deliberations. 24

MR. BARTELT: All right. Well, thank you, 25

Heritage Reporting Corporation

(202) 628-4888

Ben. And thank you to all of our participants in the 1

first panel. I think we learned a lot, and we 2

appreciate you all participating. 3

So, with that, I am going to hand the mic 4

back over to Mark Gray. 5

MR. GRAY: Thank you, everyone. Thank you 6

to all the panelists. This is the end of the first 7

session. We are now going to take a 10-minute break. 8

(Whereupon, a brief recess was taken.) 9

MR. GRAY: Hello everyone, welcome back. 10

For those of you who've only recently joined us, we're 11

going to do a quick reminder on Zoom housekeeping. 12

For panelists who are speaking but not at 13

this specific session, please keep your cameras turned 14

off, please keep your microphones on mute. 15

We will be recording the session today. 16

That recording will be made available on our website 17

in hopefully three weeks. And we have also activated 18

Zoom's transcription function today for those of you 19

who would like to follow along with captions. 20

We're going to start our panel with brief 21

introductions for each speaker, and, optionally, 22

everyone is allowed to do a short statement if they 23

desire. Please limit any statements to two minutes. 24

We will have to watch the clock and keep things moving 25

Heritage Reporting Corporation

(202) 628-4888

if you go over two minutes. 1

And after those introductions, we're going 2

to have a moderated listening session. The panelists 3

have received the questions in advance, but keep in 4

mind those are just intended as prompts and guidance 5

for discussion. We, of course, welcome you to share 6

any relevant perspectives or experience that you have 7

that you think is important for the Office to hear. 8

For those of you who are on the current 9

panel, please use Zoom's Raise Hand function. We will 10

try to call you in the order that you raise your hand. 11

That will help us keep the conversation nice and 12

orderly. 13

And then, as a final reminder for those in 14

the audience, we are, unfortunately, not accepting 15

audience questions today, so, please, no need to 16

submit questions or raise your hand. We will have 17

opportunities for public participation in the future, 18

including through a written comment period. But, for 19

today, we are trying to focus on the panelists and the 20

information they have to provide. So thank you very 21

much for your understanding. 22

With that, I'm going to introduce our 23

moderators for the second panel, David Welkowitz and 24

Jordana Rubel. David is an Attorney-Advisor in the 25

Heritage Reporting Corporation

(202) 628-4888

Office of the General Counsel with me. And Jordana 1

Rubel is also my colleague, who is an Assistant 2

General Counsel. 3

David, the mic is yours. 4

MR. WELKOWITZ: Thank you, Mark. We'll 5

start with the brief introductions. 6

Jasper AI, Alex, would you like to start, 7

please? 8

MR. RINDELS: Yeah, thank you. My name is 9

Alex Rindels. I'm Corporate Counsel at Jasper. I 10

want to thank the Copyright Office for holding these 11

really useful listening sessions. Jasper is a 12

generative AI tech startup that develops and deploys 13

software tools to assist businesses large and small 14

and individuals in their content creative processes. 15

As it relates to the subject matter of this 16

listening session, Jasper provides a software tool 17

called Jasper Art that receives text inputs from our 18

human users and filters them through generative AI 19

foundation models from various providers and ourselves 20

and spits out output in the form of art, whether that 21

looks like photography or any other computer-generated 22

art. 23

Our users are typically within the 24

professional marketing and professional creative 25

Heritage Reporting Corporation

(202) 628-4888

spaces, and they use these outputted images in 1

furtherance of things like marketing campaigns. And, 2

oftentimes, it's not just the end images that they 3

use, but they use those images for ideation to further 4

their campaign. 5

And because many of our users and customers 6

are creative professionals themselves, we receive two 7

things, a lot of positive feedback that these tools 8

have really freed up their creative processes so they 9

can think more about the things they want to create 10

and direct their efforts to the human side of the 11

creative process rather than the mechanical side of 12

it, and they're really grateful that we're able to aid 13

them in that process. 14

But then, two, we also have an ear to the 15

ground on the concerns that they have in terms of, you 16

know, their work or others' works in the creative 17

space being misappropriated and used for wrong reasons 18

or not having the right rights to use those, and we 19

really take those seriously, and we want to work with 20

our customers and everyone in the community and in the 21

responsible way. Thank you. 23

MR. WELKOWITZ: Thank you. 24

Getty Images, Paul, would you like to begin? 25

Heritage Reporting Corporation

(202) 628-4888

MR. REINITZ: Hi. Yeah. Thank you for the 1

opportunity to speak at today's session. My name is 2

Paul Reinitz, and I am Legal Advocacy Counsel at Getty 3

Images. Getty Images is an established and respected 4

member of the global media. Our growing content 5

library includes over 520 million visual assets 6

representing the work of more than 516,000 creative 7

contributors. 8

Getty Images believes AI and generative 9

models hold the potential to provide significant 10

benefits. However, we see significant risks if the 11

current development and deployment of these 12

technologies are left unchecked. 13

First, bad actors can use generative AI 14

tools to easily create deepfake content and distribute 15

it widely via Internet platforms without check. This 16

undermines public fluency in facts, trust in 17

institutions, and democracy in general. 18

Second, generative models are trained on the 19

existing work of creators. It is important that 20

longstanding IP rights that protect these creators and 21

sustain ongoing creation are not ignored. 22

Transparency and respect for third-party rights are 23

key components of mitigating these risks. Similar to 24

the position expressed by your Office in its recent 25

Heritage Reporting Corporation

(202) 628-4888

registration guidance, we believe that the use of AI 1

tools in the creative process should be disclosed. 2

Further, we believe that, one, AI generative 3

content should be identified as synthetic, and two, to 4

give rights holders visibility into the use of their 5

work, records must be kept detailing how generative 6

models were trained. The latest draft of the EU AI 7

Act codifies transparency regulations of this nature, 8

and we encourage the USCO to collaborate with the EU 9

with the goal of harmonizing standards. We believe 10

that supporting similar policy in the U.S. is crucial 11

to the Office's mission of promoting creativity and 12

free expression for the benefit of all. 13

Thank you for your leadership in this area. 14

I know that the creative and media industries more 15

broadly welcome the opportunity to provide further 16

input. 17

MR. WELKOWITZ: Thank you. 18

From Professional Photographers of America, 19

Luc. 20

MR. BOULET: Thank you. Greetings. My name 21

is Luc Boulet, and I serve as the Government Affairs 22

Manager for Professional Photographers of America. 23

We're the world's largest and oldest photographic 24

association with a membership of 35,000 creative 25

Heritage Reporting Corporation

(202) 628-4888

professionals. Our members include small businesses 1

dotting every city and town across the nation, 2

typically operating one- to two-person shops and 3

earning an average net income of $38,000 per year. 4

PPA acknowledges the profound impact of 5

artificial intelligence on our world, and we firmly 6

believe in the responsible development of this 7

revolutionary technology that will uphold the 8

principles and objectives of the copyright system. 9

We also recognize the potential harm that AI 10

may cause to the intellectual property sector, our 11

leading net export valued at an astonishing $1.6 12

trillion. Our greatest concern with AI is the 13

unauthorized scraping of a photographer's life work. 14

This work is made public and available online by a 15

photographer to promote their business and is then 16

used to create new works by AI engines. 17

AI tools are being designed to directly 18

emulate an artist's style at the click of a drop-down 19

menu. While style is generally not a copyrightable 20

attribute, the act of copying a photographer's image 21

is a violation of copyright. Transparency is possible 22

and achievable. Best practices from corporations, 23

research institutions, governments, and other 24

organizations that encourage transparency around AI 25

Heritage Reporting Corporation

(202) 628-4888

training already exist. 1

With this discussion, we hold two main 2

objectives. They are, number one, it is essential the 3

rights of copyright holders are respected as AI 4

develops and AI laws and policies are formulated. And 5

number two, new determinations on AI policy should be 6

based on the foundation of preserving the rights of 7

carefully considered to achieve this goal. Thank you. 9

MR. WELKOWITZ: Thank you. 10

Next, from Morrison & Foerster, Heather. 11

MS. WHITNEY: Thank you for inviting me to 12

participate. My name is Heather Whitney, and I am an 13

attorney at Morrison & Foerster and a member of the 14

firm's AI Steering Committee. Previously, I was a 15

Bigelow Fellow in Lecture and Law at the University of 16

Chicago Law School and a fellow and faculty affiliate 17

at the Berkman Klein Center for Internet Society. 18

Today, I'm speaking on behalf of my client, 19

Kristina Kashtanova, author of Zarya of the Dawn and 20

Rose Enigma. The Office recently refused to register 21

images Kashtanova created using Midjourney, 22

essentially on the grounds that Kashtanova could not 23

be the author because they could not predict ahead of 24

time what the output would be. In March, we submitted 25

Heritage Reporting Corporation

(202) 628-4888

a copyright registration for Rose Enigma and are 1

waiting to hear back on that application. 2

My point today is a simple one. The Office 3

is not writing on a blank slate when it comes to the 4

copyrightability of outputs created with the 5

assistance of generative AI tools. Images created 6

with these tools are visual works, and the Office 7

should treat all visual works the same. 8

Today, however, the Office's treatment of AI 9

images diverges substantially from its treatment of 10

photographs, with the bar for copyrightability much 11

higher for AI images. This inconsistent treatment 12

threatens to destabilize the registration process for 13

visual works. This inconsistency shows up in several 14

areas, but briefly I'll mention predictability and 15

what it means to be the mastermind of a work. 16

On predictability, photographers receive 17

copyrights in photos without knowing what their photos 18

will look like ahead of time. Garry Winogrand, one of 19

the most influential photographers of the 20th 20

Century, said that he photographs to see what 21

something will look like photographed. Wildlife 22

photographers register photos taken by cameras on 23

motion sensors, photos where the photographer was not 24

present and had no idea what the image was going to be 25

Heritage Reporting Corporation

(202) 628-4888

until they looked later. Photographers do not have to 1

predict how their works will look to be the authors of 2

them, and AI artists should not either. 3

On the mastermind, to be the author of a 4

work, the Office has stated that one must be the 5

mastermind of it. We are all the authors. We could 6

all register the endless, endless photos we take with 7

our phones. If we are the masterminds of those 8

photos, where our creative contributions are so 9

minimal, it is hard to understand why AI artists, like 10

Kashtanova, are not the masterminds of images 11

generated after setting far more parameters and making 12

far more creative choices. 13

In short, whatever the test for authorship, 14

the test should be applied consistently across the 15

visual arts. Thank you for having me. 16

MR. WELKOWITZ: Thank you. 17

From The Niskanen Center, Daniel. 18

MR. TAKASH: Hi. Thank you. My name is 19

Daniel Takash. I'm a Regulatory Policy Fellow at The 20

Niskanen Center. We're a 501(c)(3) public policy 21

think tank. We work on a wide range of public policy 22

issues, from employment and poverty welfare to climate 23

and intellectual property. 24

Thank you so much to the Copyright Office 25

Heritage Reporting Corporation

(202) 628-4888

for putting this together. Thank you so much to 1

everyone for attending. 2

And so I’d like to make three global points, 3

I guess, to frame my comments. 4

First, it makes absolute perfect sense that 5

the Copyright Office would be one of the first 6

agencies to host formal discussions around the nature 7

of artificial intelligence. Just the way that this 8

field is developing and things are shaking out, they 9

have implications for copyright law, perhaps before 10

more than any other field. 11

And I would also like to commend the way the 12

implications of artificial intelligence, especially as 14

it relates to the registration of AI-generated works, 15

both as a matter of policy and a matter of law as it 16

exists today. I think they've gotten it right. 17

But the second point I'd like to make is 18

that as important as copyright is, I don't think it 19

should be the final word or even necessarily the most 20

consequential word on developments in AI. This is a 21

technology that perhaps will stop developing today, 22

and it will just remain a novelty, in which case IP 23

will remain the most important framework under which 24

to regulate it. 25

Heritage Reporting Corporation

(202) 628-4888

But the sky is the limit, and the potential 1

of artificial intelligence across virtually all 2

domains of life are extraordinarily consequential, and 3

for that reason, I think it's important to keep it in 4

perspective. If we're talking about general safety, 5

job dislocation, or other issues that are separate 6

from, even though they may be related to, intellectual 7

property, I think it's important that they take 8

priority in any discussion. 9

And then, finally, I would encourage 10

everyone to remain somewhat forward-looking. The role 11

of copyright is to promote the progress of science, 12

and it would be problematic for us to use that policy 13

regime in order to limit it. 14

We can all look back and laugh at John 15

Philip Sousa, you know, bemoaning the threat of 16

phonograph-dooming music. And I want to be abundantly 17

clear that the threats posed, that fear of creative 18

destruction, is real, sincere, and precedented, and we 19

must be prepared to say what other policy domains 20

separate from copyright law should be best used to 21

address the problems to the extent they exist. 22

Thank you. 23

MR. WELKOWITZ: Thank you very much. 24

From Dual Wield Studios, Zara. 25

Heritage Reporting Corporation

(202) 628-4888

MS. VARIN: Hi. Thank you so much for the 1

opportunity to speak today. My name is Zara Varin. I 2

am the Art Director and Senior Product Designer at 3

Dual Wield Studios. It's a company based and founded 4

on making things we love, inspired by the things we 5

love. 6

My personal background is a little all over 7

the place. After getting out of the Marine Corps, I 8

established a career as a game developer and got a 9

foot into the licensing industry as well. I've worked 10

on everything from video games and comics to action 11

figures, costumes, TV, movies. The point is I've 12

gotten to work on a lot of cool stuff. 13

Before all that, though, I was a fan artist, 14

and I still am. Uniquely, in my day job, we strive to 15

champion and partner with fan artists and indie 16

creators to elevate them in official collaborations 17

with IP holders in ways that center those fan works 18

while compensating and crediting them fairly, all 19

while safeguarding the IP's representation. We bridge 20

the gap between licensors and fandom to ensure 21

creativity isn't stymied but celebrated. 22

Within the licensing industry, we're an 23

outlier in our prioritization of fair wages, ethical 24

production, and credit to the artists that we work 25

Heritage Reporting Corporation

(202) 628-4888

with. For many artists, that credit underscores their 1

portfolio of work in acting as a resume. 2

So, in order to be effective at what I do, I 3

have to keep up with constantly changing technology 4

and tools, and I worry about the current state of 5

generative AI. It grossly undermines credit to the 6

vast swath of works informing their data sets. 7

For artists whose name has been used to 8

populate guidance for these imitative generators, 9

their actual work is becoming mired in a sea of 10

manufactured imposters. It purports to be built for 11

us, but the methodology informing its data sets and 12

lack of clear ethical foundation indicates otherwise. 13

When data set training is reliant upon things like 14

underpaid workers combing through horrifying content 15

to identify and tag it, it demonstrates a worrisome 16

set of priorities. 17

Lastly, the most disheartening aspect of 18

this is seeing how it has scared and dissuaded so many 19

folks, from those just starting to explore creative 20

expression to those who have spent countless hours 21

honing their skills and often sharing to encourage 22

others. Many people whose work was used 23

nonconsensually to train these tools are being 24

jeopardized and devalued in their own industry by 25

Heritage Reporting Corporation

(202) 628-4888

them. 1

I think generative AI genuinely has the 2

potential to become a robust tool for creators, but it 3

requires far greater ethical application and scope 4

before it's something I'm comfortable considering 5

incorporating into my workflows. 6

I'm grateful to discuss things further with 7

all of you. Thank you so much. 8

MR. WELKOWITZ: Thank you very much. 9

From Vanderbilt University Law School, 10

Daniel. 11

MR. GERVAIS: Yes. Thanks, David. Thanks 12

for having me as a panelist today. 13

My name is Daniel Gervais. I'm a Professor 14

at Vanderbilt Law School where I teach AI and IP law. 15

I've written extensively on AI and IP and on the 16

regulation of AI, and also on the legal and functional 17

differences between human and machine learning, 18

intelligence, and thinking. It's all available on 19

SSRN, so if anyone wants to have a look. And, in 20

fact, I just posted a summary checklist of issues on 21

AI and IP to both Twitter and LinkedIn that you may 22

find useful. 23

As to today's topic, there are five legal 24

questions I'm interested in, and I don't think we'll 25

Heritage Reporting Corporation

(202) 628-4888

have time to get into all five, but here they are. 1

The first is, obviously, is the scraping or 2

text and data mining legal? And, here, I'm interested 3

not just in U.S. law. I'm also looking at foreign 4

international law. 5

Second question, does the machine infringe 6

when it produces a new work? And I think the analysis 7

here needs to be separate for different rights in the 8

The third question we've been already 10

discussing, I mean, in the previous panel, and I'm 11

sure we'll get back to it in this panel, is, can the 12

machine be an author? Here, basically, I strongly 13

support the guidance from the Copyright Office. I'd 14

be happy to explain why. 15

The last two questions I don't know if we'll 16

get to would be, is the AI algorithm itself protected 17

by IP law? 18

And the fifth is, is the data set, if you 19

want to call it that to simplify, copyrightable or 20

protected? 21

I'm also interested, and I'll end with this, 22

at a deeper level on the alignment, by which I mean, 23

is it possible to align the future development of AI 24

in this space with the needs for humans to grow and 25

Heritage Reporting Corporation

(202) 628-4888

develop and to use art to communicate, both to send 1

and receive new ideas and messages? 2

So I'll end here and look forward to our 3

conversation. Thank you. 4

MR. WELKOWITZ: Thank you. 5

And, finally, from ASCRL, James. 6

MR. SILVERBERG: Hi. I'm James Silverberg. 7

I'm the CEO of the American Society for Collective 8

Rights Licensing. I'm also a former law professor, 9

and for more than 40 years, I worked litigating 10

ASCRL is a not-for-profit corporation that 12

represents tens of thousands of illustrators and 13

photographers. We collect and distribute collective 14

licensing fees. These come from collecting societies 15

in foreign countries which have implemented proven and 16

successful collective licensing systems to compensate 17

artists for non-author, non-title-specific content 18

use, that is, compensations distributed to authors in 19

spaces where the use of content is certain, but the 20

identification of the specific content or authorship 21

is not always possible or is impractical. These 22

spaces are like the AI space. 23

ASCRL is interested in exploring legislation 24

to provide for collective licensing solutions for the 25

Heritage Reporting Corporation

(202) 628-4888

use of illustration and photography in AI platforms 1

and believes that subscription and advertising or 2

other fees should be collected to compensate authors 3

on a nonspecific basis for the ingestion of their 4

material into AI platforms. 5

We need to be aware that the current 6

of art and authorship in the context of how AI 8

generates visual artwork. The constitutional 9

authorization for legislation to reserve to authors 10

the rights to their creations first found form in the 11

prohibitions against copying expression and 13

permissible exceptions, like the Fair Use Doctrine, 14

with the objective of preserving the economic benefit 15

of artwork for authors with the intention of expanding 16

American culture. 17

The constitutional purpose of reserving 18

authors' rights remains outstanding, but the current 19

AI-generated works where the technical lines of 21

expression copying are not always crossed or where 22

existing fair use factors become a part of the 23

equation. 24

For example, one of the main challenges with 25

Heritage Reporting Corporation

(202) 628-4888

AI-generated works is that the existing copyright laws 1

are ill-equipped to preserve the economic benefit of 2

authors when their material is learned or ingested or 3

when uncopyrightable styles are appropriated. Even 4

when infringement does occur, it can be difficult to 5

detect and identify and prohibitive costs can be 6

associated with pursuing legal action. 7

Much of our discussion today is already 8

focused on the failures, challenges, and uncertainty 9

of applying the existing law and debating its 10

application in the context of injection, and this in 11

and of itself may be proof that copyright law is at 12

best problematic and uncertain as a solution to the 13

problem of author protection. 14

For these reasons, ASCRL believes that we 15

need a new way of thinking about how we should 16

implement the constitutional premise that we reserve 17

to authors their rights and their ability to receive 18

compensation because the current copyright system is 19

not achieving and cannot really achieve the 20

constitutional goal in the AI context. 21

So, to address this challenge, ASCRL 22

recommends that we do not entirely focus on the 23

niceties of infringement, issues of interim copying, 24

fair use factors, and that we move towards 25

Heritage Reporting Corporation

(202) 628-4888

legislatively implementing collective licensing 1

systems like those that are currently used very 2

successfully in many foreign countries. These systems 3

serve our constitutional objectives and facilitate 4

licensing and the use of AI and create a more balanced 5

system that recognizes the needs of the AI community 6

as well as the authors whose works or work attributes 7

are ingested into these systems. 8

We are hoping to level the playing field by 9

requiring non-title-specific, non-author-specific 10

compensation where works cannot be specifically 11

identified in order to compensate for uses where 12

specifics of use are not available. 13

MR. WELKOWITZ: Thank you, James. 14

MR. SILVERBERG: I appreciate your inviting 15

me on the panel, and I look forward to our discussion. 16

MR. WELKOWITZ: Thank you very much, James. 17

And thank you all for introducing yourselves, and 18

welcome again. 19

And to begin the discussion, let's begin 20

with this first question. How is the training of 21

artificial intelligence models affecting your field or 22

industry? And what should the Copyright Office know 23

about the technology's use of training materials when 24

it is considering copyright issues relating to 25

Heritage Reporting Corporation

(202) 628-4888

training? Please be specific in terms of which part 1

of the visual arts ecosystem you're talking about. 2

Let's see. And please use the Raise Hand 3

feature. Ah, good. Zara, you're first on my screen. 4

MS. VARIN: Hi. Thank you. It's Zara. 5

MR. WELKOWITZ: Oh, I apologize, Zara. 6

MS. VARIN: It's okay. 7

MR. WELKOWITZ: I apologize. Sorry. 8

MS. VARIN: Thank you. 9

So I occupy several different unique spaces. 10

I have a foot in a lot of different worlds, both as a 11

game developer and within the licensing industry, and 12

something that we've been seeing a lot across the 13

board is that generative AI has kind of become a 14

digital gold rush. And the training for informing 15

that digital gold rush has come from a lot of my 16

peers. I believe Karla touched on this during the 17

first part of this panel. 18

But it's very disheartening when you Google 19

search somebody's name and what comes up now is a 20

proliferation of images that were created utilizing 21

their art style but were not created by them 22

themselves. And that's creating uncertain authorship 23

and a great deal of uncertainty for people who are 24

actually in these fields who are thinking maybe I 25

Heritage Reporting Corporation

(202) 628-4888

don't want my online portfolio available at all. 1

Maybe I should safeguard or gatekeep my work in a way 2

where I have to have password protection so that I can 3

control the access. And it's essentially denigrating 4

a lot of the work that people have been sharing online 5

since the Internet's inception. 6

And it's incredibly disheartening as well 7

within a production environment that these training 8

methodologies are being done nonconsensually as well. 9

Even people who have attempted to reach out utilizing 10

the opt-out options that some of these models have 11

kind of after the fact offered to artists, they're not 12

having successes with opting out, or they're still 13

seeing their work featured in those models. 14

And it's, I think, very telling that if you 15

ask one of these models to create work based on an 16

artist who wasn't part of that data set, it cannot. 17

It’ll try. It’ll do something. But you're not going 18

to end up with something that is aping the style and 19

imitating something. 20

They're very sophisticated calculators by 21

taking your set of inputs, running it through the 22

algorithms to arrive at an output, but there's no 23

creativity inherent in the process. And the training 24

data utilized to do that is where the creativity 25

Heritage Reporting Corporation

(202) 628-4888

exists. The prompt might be a couple of cool words 1

that somebody's put together. And I don't want to 2

punch down on folks who are starting to learn how to 3

do prompting. But the training data sets themselves 4

are pulled together from a vast quantity of many 5

works, folks who are here representing some of them, 6

and that's causing many creators to second guess 7

whether they want to be creators at all. 8

MR. WELKOWITZ: Thank you. 9

Luc, you're next on my screen. 10

MR. BOULET: Thank you. 11

This question implies that the visual arts 12

industry understands the extent to which AI companies 13

are using and profiting off of the works of others. 14

And this is quite the opposite. Without a transparent 15

and open AI system, it's impossible to determine which 16

AI-generated works are incorporating the copyrighted 17

material of others. And this leaves both the public 18

and copyright holders completely unaware of which 19

photographic works are being used by AI engines and 20

the extent of their use. This is why it's imperative 21

for AI companies to disclose their sources and methods 22

used for when creating their final product. 23

There's growing anecdotal evidence of AI 24

work displacing works created by human authors. A 25

Heritage Reporting Corporation

(202) 628-4888

publishing company looking to support their article 1

with photographic evidence may turn to AI to produce a 2

bespoke image, and others are inputting their selfies 3

into an AI server to generate their latest LinkedIn 4

headshot. 5

It's no surprise that individuals with the 6

least bargaining power are most vulnerable to the 7

negative effects of AI systems. Photographers who 8

rely on their work to make a living often display 9

their images in digital galleries and are promised 10

with the option to opt out of having their works 11

scraped in the future. However, these promises are 12

empty as the timeline for when they may choose to do 13

so has yet to be announced or is unclear at all. 14

license their work, and that wish should be respected. 16

MR. WELKOWITZ: Thank you. 17

Paul? 18

MR. REINITZ: Thanks. 19

So, yes, I think that the short answer is 20

that these technologies are having a big impact, and 21

we expect that the impact is going to continue. As I 22

said in my introductory statement, we believe that 23

there's a lot of potential for these systems to help 24

creativity, but there is also a lot of potential for 25

Heritage Reporting Corporation

(202) 628-4888

harm. 1

Now, to talk about exactly how it's 2

impacting our industry, you know, I would like to 3

point out that, and this kind of goes into the second 4

question as well, that, you know, high-quality content 5

along with captions that explain that content is 6

really valuable in the machine learning process, and 7

because of that, you know, there's a lot of demand for 8

our content. And we’re doing licenses out there. 9

There's a big demand in our industry for people to 10

come to us knowing that we have collected rights over 11

nearly three decades of our existence and that we can 12

license safely for these usage. 13

We also see that our customers are using 14

this technology already. We recently did a scientific 15

survey, and over half, I think it was 56 percent, of 16

our customers that were polled are already using AI in 17

their current workflow. Now, you know, we think that 18

there are lots of great ways to use it and it can help 19

promote creativity. But, again, you know, at the end 20

of the day, these customers are more excited about 21

licensing content from us for commercial use because 22

they know that there's legal certainty in a license. 23

And I think that hopefully we can get to a point where 24

there's confidence in these tools that somebody can 25

Heritage Reporting Corporation

(202) 628-4888

use the output for their creative work. 1

I'll stop there. 2

MR. WELKOWITZ: Thank you. 3

Daniel, you're next. 4

MR. GERVAIS: Thanks. Three quick points. 5

The first is I think the Office in its 6

capacity as advisor to Congress and the courts should 7

bear in mind the international obligations of the 8

United States, in particular, the TRIPS agreement, 9

three-step tests, for example, in providing advice on 10

what can be done without permission or payment. 11

Second, the question of scraping, of course, 12

can happen in many different countries. Many other 13

countries have already adopted exceptions, the EU, 14

Japan, others, Singapore. So one way or the other, 15

you know, text and data mining will happen, but there 16

will be limits, and whatever needs to happen beyond 17

those limits will need to be licensed. So perhaps the 18

Office can play a role in facilitating licensing. And 19

by licensing, I don't mean just that if you're an 20

artist, you get paid once because your work gets 21

scraped and you get paid because there's a licensing 22

regime. A license can also impose contractual limits 23

on the use and reuse of material. 24

Now anything that's scraped can lead to an 25

Heritage Reporting Corporation

(202) 628-4888

output that will potentially create a commercially 1

competitive product, but at least there is some 2

conversation, some exchange of consideration. 3

So I think those are the three roles I would 4

see the Office potentially playing. Thank you. 5

MR. WELKOWITZ: Thank you. 6

James? 7

MR. SILVERBERG: So I wonder if we'll really 8

succeed in disentangling the many problems of AI. Is 9

the problem really whether there's a copyright 10

infringement? Is the problem really identification of 11

works? Is it opt-out? Is it content access 12

restrictions? I think all of these things are 13

important when we're looking at the ingestion problem. 14

But is that really where we should be focusing our 15

inquiry? Isn't the problem that we need to have a 16

clear system, call it copyright or something else, 17

that makes sure that the authors are compensated? 18

MR. WELKOWITZ: Thank you. 19

Heather? 20

MS. WHITNEY: Thanks. I just want to 21

quickly mention something about the contribution of 22

artists who are using these tools in terms of how it 23

has outputs. So I think there's a misconception that 24

all of these people are just sort of typing in some 25

Heritage Reporting Corporation

(202) 628-4888

generic words and then there's an output. I strongly 1

recommend people just Google control net models, 2

Stable Diffusion, and you can see the different kinds 3

of models that are now used on top of kind of like an 4

extension of Stable Diffusion in order to have a much, 5

much greater control over the images that are being 6

used. It's not just the model that's doing the work. 7

And if you look at what they can do with 8

these things, with these kinds of different kinds of 9

models, setting up the composition, choosing sort of 10

the angles, the lighting, all those kinds of things, 11

and you compare that to the things that are said in 12

the Compendium about what is required for someone to 13

have a copyright in a photograph, this is just 14

absolutely without question equal to or greater than 15

the kinds of things that people are doing in that 16

context. 17

So I just think it's helpful that people 18

understand the technology and how it's evolved, and I 19

think the control net models are really a way for 20

people to start to understand where that's going. 21

MR. WELKOWITZ: Thank you. 22

Alex? 23

MR. RINDELS: Yes. Thanks. I'd just like 24

to briefly respond about how the output is positively 25

Heritage Reporting Corporation

(202) 628-4888

affecting the industry that we're in. 1

So, in two regards, the professionals who 2

use our service as creatives, they're able to unleash 3

their creative ability that might have otherwise been 4

tied up in time-consuming creative processes. So many 5

of them use Jasper Art to create ideas that they then 6

build upon or to create end products that they use in 7

marketing campaigns or in other business uses or 8

commercial uses, and this greatly frees up their time 9

to use their creative potential, their intelligence, 10

their efforts in other productive ways. 11

And secondly and probably more importantly 12

are the output in tools like Jasper and others, 13

Stability, it allows people in the creative space who 14

otherwise could not have created output like this to 15

create output, and I'll explain briefly. 16

So we also have a text-generating tool. And 17

we've had numbers of customers who have dyslexic 18

disabilities or otherwise who would have otherwise 19

been unable to create output, and they regularly 20

respond to our tools and tell us how grateful they are 21

that they're now able to take what was in their head 22

and put it to paper. And, previously, it was just 23

basically a mechanical limitation that prohibited them 24

from doing so. And now, with tools like this, it's 25

Heritage Reporting Corporation

(202) 628-4888

untapped, it's unleashed their potential that 1

otherwise was untapped in the industry. 2

So, in the first part, it's freeing up space 3

for people who already have the potential to create 4

art or other output and it gives them more time. But, 5

secondly and probably more importantly, it also allows 6

those who otherwise could not have created artistic 7

output to do so. 8

MR. WELKOWITZ: Thank you. 9

Daniel? 10

MR. TAKASH: Thank you. So, with respect to 11

the images that are training, Niskanen’s policy with 12

respect to all the works we put out, images included, 13

is Creative Commons provided, there's attribution, 14

because we do our best to contribute what we like to 15

think at least is good-quality work so that images can 16

be trained. We are also fond of using AI-generated 17

images just as a substitution for stock photography 18

that we would otherwise license, or we can have some 19

bespoke images that particularly complement some work 20

that we're generating. So we have an interest in 21

making sure that material is out there so that the 22

models can be the best that they can be. 23

That being said, we recognize that there 24

ought to be tools available and the law should 25

Heritage Reporting Corporation

(202) 628-4888

accommodate the use of those tools to allow the 1

authors or the rights holders of works, in this case, 2

visual artists or visual arts, although this principle 3

can be extended to other contexts, that, you know, 4

that creates some ability to remove or make it so that 5

it's much harder for their works to be learned upon. 6

That being said, we should recognize that 7

veto to the rights holder, and there are plenty of 9

examples, you know, say what you will about the 10

quality or the desirability of this outcome overall, 11

where, you know, use can be even in a way that the 12

original rights holder may not approve of. 13

That being said, in order to ensure quality 14

and respect, incentives to produce arts, I think the 15

best way to square the circle is to focus on a regime 16

that deals with remuneration and financing, which 17

deals with much larger policies, as I alluded to 18

earlier in my comments. 19

MR. WELKOWITZ: Thank you. 20

Paul? 21

MR. REINITZ: Hi. Thanks. Yeah, I just 22

wanted to make a quick comment about what Daniel just 23

said. I mean, you just said that you, you know, as a 24

consumer of stock photography, sometimes use generated 25

Heritage Reporting Corporation

(202) 628-4888

content as a substitution. I think that that's fine 1

and I think that the market is probably going there. 2

But I would like to point out that if you are going to 3

be using a substitute and that model that creates the 4

generative content was trained on unauthorized 5

content, you are substituting a product without 6

compensating the artists that were needed to make that 7

model. 8

And, you know, I just, sorry, I just needed 9

to call that out because it just, it’s so real, as you 10

say that, as a user of stock photography. 11

MR. WELKOWITZ: Thank you. 12

James? 13

MR. SILVERBERG: Just to follow up or 14

elaborate on Paul's point, which I very much 15

appreciate, the problem is also particularly paramount 16

for small businesses and medium-size businesses that 17

involve authors who do not have vast amounts of 18

content to aggregate and to license on a large-scale 19

basis. And so, while there are content aggregators 20

that are able to do that on a much larger scale 21

because they occupy a larger market force, hundreds of 22

thousands of individual creators and authors don't 23

have the capacity to engage in a licensing transaction 24

with an AI platform, nor do they have the capacity to 25

Heritage Reporting Corporation

(202) 628-4888

enforce copyrights in copyright litigation or possibly 1

even to identify the use of their work. 2

And so that is why ASCRL is advocating for a 3

remuneration system that will prevent a complete 4

market displacement of these constituents and which 5

will make sure that they get compensated. 6

MR. WELKOWITZ: Thank you. 7

Daniel? 8

MR. TAKASH: Thank you for circling back. 9

Yes. So just to be clear, with respect to 10

Paul's point, I am fully cognizant of the, I guess you 11

could call it, recursive nature between the body of 12

works available for training and the output of it, so 13

I'm fully aware of that. 14

I think the problems -- or not problems, the 15

challenges artificial intelligence creates with 16

respect to copyright is a difference of both degree 17

and of kind, which is why I'd like to underscore again 18

the support for something that moves away from more 19

traditional models of infringement, alluding 20

to -- I can't say I'm super familiar with the model 21

James brought up originally, but something that 22

focuses more on remuneration moving into a very 23

different technological paradigm in order to, as I 24

mentioned before, square that circle. 25

Heritage Reporting Corporation

(202) 628-4888

MR. WELKOWITZ: Thank you. 1

Okay. I think I'm going to turn the mic 2

over to my colleague, Jordana Rubel, for the next 3

question. 4

MS. RUBEL: Thanks, David. 5

I guess I'll start off by maybe just asking 6

a question that relates to the point we were just 7

talking about, which is more solution-oriented 8

thinking about if we get to the place where we are 9

just talking about remedies here. Maybe we can start 10

with James to give a little bit more detail about what 11

ASCRL's proposal is, and if other folks want to speak 12

to that, you're welcome to raise your hands as well. 13

I'll turn it to you, James. 14

MR. SILVERBERG: Thank you. So I appreciate 15

in the United States there's a general lack of 16

familiarity with collective licensing systems, 17

particularly in the visual space, because we don't 18

have any. In foreign countries, there are collective 19

systems in a number of different areas. I'll just 20

mention two, for example, to illustrate how they work. 21

One would be in the area of library lending. 22

As we all know, in the United States, there's a first 23

sale doctrine. If a library buys a book, they can 24

lend it forever and never pay for it again. But, in 25

Heritage Reporting Corporation

(202) 628-4888

foreign countries, a system's been introduced as a 1

secondary or adjunct system to the copyright law 2

involving what's referred to as secondary rights. 3

For library lending, people don't know what 4

books are being lent, people don't know which authors 5

are involved in the lending, people don't know which 6

photographs are in what books or what illustrations 7

are in what books, but fees are paid into a collecting 8

society in order to compensate the authors for the 9

lending of this material. It's done on a non-author, 10

non-title-specific basis. We don't know exactly what 11

the fees are for other than they’re for the lending 12

itself, and then algorithms are established in order 13

to make a fair allocation of the revenue to the 14

individual authors whose works might be embodied. 15

It's a form of rough justice. It's not 16

specific. It's not a perfect system, but it's a way 17

of creating compensation. It is analogous to a 18

system, sort of like a SoundExchange, where fees are 19

being paid for the use of recorded music that's 20

digitally streamed, and that goes into SoundExchange, 21

one of our directors on our Board of Directors, a 22

former executive director of SoundExchange, and those 23

funds are distributed, but they're distributed on a 24

more specific basis where it's known what material is 25

Heritage Reporting Corporation

(202) 628-4888

used. 1

We would be proposing a system where AI 2

platforms would be distributing compensation to 3

collecting societies in order to distribute those 4

according to a fair algorithm which would compensate 5

illustrators and photographers. And this was also 6

espoused by Authors Guild in the other listening 7

session for works in the writing space. 8

MS. RUBEL: Thanks. 9

Daniel, did you want to respond to that, or 10

do you have any other thoughts on this topic? 11

MR. GERVAIS: Just a footnote just to be 12

clear, James is right about public lending, but the 13

way it works is not random. I just want to make that 14

very clear. So, in countries where they have it, I 15

don't know, Germany, Canada, and others, the authors 16

must register their works, and then the collective 17

actually will only pay -- so they will survey certain 18

public and private libraries, usually mostly public 19

libraries, and will only pay if the book is actually 20

in those libraries. And the reason they don't pay 21

per, you know, the number of times that the book was 22

taken out by a user at a library is partly privacy. 23

And so the second regime, though, that is a 24

little different is this thing called extended 25

Heritage Reporting Corporation

(202) 628-4888

licensing that I know the Copyright Office has 1

published a number of reports about where, basically, 2

a collective is given the -- basically, it becomes an 3

opt-out, essentially, so the collective basically has 4

rights to represent a class of right holders, except 5

those who opt out. This is very, well, I can't say 6

very common, it's common in Europe, and other 7

countries are looking at it. Some countries have 8

functional equivalents of that. 9

That is a different thing because what it 10

does is it gives a collective a right to license, but 11

the basis on which then authors will get paid or 12

publishers or whoever else has rights is not 13

predetermined. It has to be decided case by case in 14

the appropriate way. It's not a black box. I want to 15

make sure people understand this isn't just a black 16

box of money that gets, you know, paid somehow. You 17

can use data to actually apportion the funds 18

correctly. Thank you. 19

MS. RUBEL: Paul? 20

MR. REINITZ: Yeah. Thanks. And thanks, 21

Daniel. I was actually going to bring up a similar 22

issue. 23

I would say that, you know, I think the idea 24

of collective licensing is a good one but only if it's 25

Heritage Reporting Corporation

(202) 628-4888

really necessary. Really, you know, it's a complex 1

system to set up. There's a lot of administration, 2

and, you know, there can be a lot of inefficiency in 3

it. 4

From where I currently sit, as I described 5

in my opening statement, we're seeing a lot of 6

interest, and we're doing a lot of licenses directly 7

with people that want to or organizations that want to 8

license content. Now I understand that that doesn't 9

work for everyone, and it's much harder for an 10

individual, you know, photographer to say do that on 11

their own, and maybe we need a solution like what 12

James is suggesting. 13

That said, you know, along with what Daniel 14

described, I think the idea of an opt-out and 15

following these models of extended collective 16

licensing is essential because, you know, if an 17

organization wants to opt out, they should be able to 18

do so. 19

MS. RUBEL: Thank you. 20

Zara? 21

MS. VARIN: Thank you. 22

So, to your question, it's been really 23

interesting to see this conversation contextualized 24

and framed by different industries. So, for instance, 25

Heritage Reporting Corporation

(202) 628-4888

film and music and the music industry as a whole, they 1

have a far more rigorous enforcement of their 2

like, the games industry or visual effects industry or 4

working as freelancers, don't have a single unified 5

front to advocate for us. 6

So there's efforts to unionize, but we don't 7

really have any sort of large collective that is 8

coming to attend events like this, that's speaking on 9

our behalf and helping to advocate for the rights of 10

artists and creators in those industries. 11

I think folks on these sorts of creative 12

teams doing freelance work and within, I think, more 13

of the cutting-edge tech industries where there's a 14

lot of overlap between utilizing technology in our day 15

to day and rapidly adapting to what that new 16

technology is could definitely use some sort of 17

remunerative residual system or something along the 18

lines of what I think James was getting at and what 19

Paul has also touched on. But that has to work and 20

advocate alongside credit to the artists who have 21

helped build that system. That system wouldn't work 22

without all of the pieces that have built that 23

learning model. 24

So I think step one is figuring out, how do 25

100

Heritage Reporting Corporation

(202) 628-4888

we backtrack and ensure that all of the existing 1

learning models are clear, transparent, ethical, have 2

defined what their scope is, and also define what 3

efforts they are making? Because the onus is on them 4

for building these tools to do so in a manner that is 5

not going to undercut people that are already working 6

in those creative industries, because, again, I do 7

think there's a lot of great potential in these tools, 8

but they are tools. They are not creative. 9

Artists necessarily are called skilled. 10

It's not talent. Talent is kind of a very worrisome 11

eugenicist concept. Skill is a thing that you have to 12

work at. There is no skill inherent in the AI 13

generative process. The only skill present is coming 14

from the works that are scraped to put everything 15

together. So whenever we're contextualizing all of 16

this, I want folks to keep that in mind, where is the 17

human labor and where is the skill coming from? 18

MS. RUBEL: Daniel? 19

MR. TAKASH: Oh, sorry, two Daniels. It 20

gets confusing. 21

MS. RUBEL: Sorry. 22

MR. TAKASH: Yes. So, with respect to the 23

model for remuneration and licensing based on the 24

work, I think that cognizant of the transaction costs, 25

101

Heritage Reporting Corporation

(202) 628-4888

which are not insurmountable but are real and there, 1

and additionally, the fact that artificial 2

intelligence is nascent and may not yet, underline 3

yet, have a clear path towards commercialization and 4

ability to generate a source of revenue that can 5

easily be extracted. 6

And balancing, I think, those shortcomings 7

at present with the potential for artificial 8

intelligence, again, across the economy, you know, in 9

a wider way and getting it off the ground, I think it 10

may be worth considering, and I value everyone else's 11

input, some type of alternative source of revenue in 12

the initial phases of any type of licensing that may 13

exist. A popular idea that comes around every now and 14

then, particularly floated by Paul Romer, would be a 15

tax on online advertising revenue, I think that's an 16

attractive opportunity for a source of revenue, but 17

something that I would like to encourage folks keeping 18

in mind at least in the early phases. 19

MS. RUBEL: Okay. I'm going to go to Luc 20

next because he hasn't spoken on this point, and then 21

I'll come back around to James and Paul. 22

Luc? 23

MR. BOULET: Thank you. 24

Our position is that an AI engine cannot 25

102

Heritage Reporting Corporation

(202) 628-4888

capture the beauty of a photograph without initially 1

copying the image. And ingestion of copyrighted works 2

by AI developers without proper authority constitutes 3

great concern. 5

The reason why developers are seeking out 6

copyrighted works is because professional 7

photographers create high-quality photos that are 8

exceptionally well-suited for AI ingestion, and in 9

this case, they're considered so valuable because 10

quality of input determines their quality of output. 11

And I just want to also add that the 12

priorities of individuals using copyrighted materials 13

for AI ingestion, you know, must not and should not 14

take precedence over the rights and interests of 15

creators. The AI systems should not be built on their 16

backs without their consent. And we must not 17

compromise longstanding laws and policies that protect 18

the rights of copyright holders in the pursuit of 19

developing AI technology. 20

MS. RUBEL: Thanks. 21

James? Oh, I think you're still muted. 22

MR. SILVERBERG: I really appreciate what 23

Luc said, and I want him to be right about everything 24

that he said, but I'm a little bit concerned about the 25

103

Heritage Reporting Corporation

(202) 628-4888

discussion about AI ingestion because there are fair 2

use issues and other issues there which make that 3

final conclusion problematic. 4

Additionally, even if there is a 5

determination in these court cases that there's an 6

infringement, I'm not really sure what the utility of 7

that is going to be for individual authors and small 8

businesses who won't be able to monetize or possibly 9

even identify the use of their material on AI 10

platforms, particularly not through expensive 11

litigation. 12

So I think it leaves us in a situation where 13

we're still looking for answers even if Luc turns out 14

to be right about everything that he said, and, again, 15

I hope he is and he may be, but we're still going to 16

have a problem. 17

And to follow up on Zara, I hope I said your 18

name correctly, to follow up on Zara's comment, 19

unfortunately, the onus may not be on the AI platforms 20

to account for what they're doing because, unless what 21

they're doing becomes clearly illegal, they have 22

loopholes through the Fair Use Doctrine or other 23

means, interim copying doctrines and other legal 24

niceties, to evade liability. And they can sit here 25

104

Heritage Reporting Corporation

(202) 628-4888

today and present the argument that they're compliant 1

with the law and may very well be. 2

But our problem isn't whether they're 3

legally compliant and whether they're violating the 4

to make sure that the Zaras and the Karla Ortizs of 6

the world get paid for the use of their style, 7

content, appropriation of their efforts, when the 8

current copyright system sort of is not really well 9

fitted to doing that the way I would like it to be? 10

MS. RUBEL: Paul? 11

MR. REINITZ: Thanks. 12

Yeah, so I also wanted to respond to Zara 13

and also the problem that James is pointing out. And 14

I think, you know, again, as I said in my opening 15

remarks, I think, you know, one of the best solutions 16

we can do to mitigate these risks is require 17

transparency. We need to have obligations on the 18

developers of these models to keep records of the 19

content that they trained on so that Zara or any other 20

creator can tell if their content has been used. 21

And, you know, we are very early on, but 22

this is something that we cannot go back and redo. We 23

need to have these transparency obligations now so 24

that there isn't just the ability to basically sweep 25

105

Heritage Reporting Corporation

(202) 628-4888

this stuff under the rug, and as the legal issues are 1

decided, we need to preserve, basically, the evidence 2

of what's going on. 3

MS. RUBEL: Alex, I'm going to give you a 4

chance to jump in here. 5

MR. RINDELS: Thanks. Yeah, I'd say two 6

points. 7

If society decides that we should pursue 8

some type of remunerative system for this, I think 9

being able to attach copyright protection to the end 10

works themselves would actually be a pretty efficient 11

way to collect fees for that remuneration. So a lot 12

of our end users are constantly asking us whether the 13

images they generate can be protected by copyright, 14

and if they had to file an application like everyone 15

else and that's part of that, meaning they have a duty 16

to disclose that AI tools assisted in their generation 17

of the content, maybe there's some sort of a fee that 18

attaches to that. 19

My second point is I think it becomes very, 20

very difficult for a remunerative system, although 21

there are people much smarter than I who could 22

probably determine where and into what portion of the 23

fees, who they go to. 24

One, some AI models themselves aren't 25

106

Heritage Reporting Corporation

(202) 628-4888

storing the images so that when somebody puts in a 1

text prompt, it's not going and retrieving an image 2

and outputting it or some variation of it. The models 3

themselves temporarily notice patterns in the images 4

and in those patterns create an algorithm, and that 5

algorithm overall is what the model is. And when you 6

put an input for it to generate something, it simply 7

provides a reasonable approximation of what the output 8

should look like based on your text input. 9

So, for that, in that case, it would be very 10

difficult to determine whose input was used to produce 11

your output, so I think that would be difficult for 12

attributing the fees as well. 13

MS. RUBEL: Yes, and we certainly heard that 14

from Stability AI earlier this afternoon as well. 15

Luc, you're going to be the last voice on 16

this question, and then we're going to move to talk a 17

little more about the Copyright Office's registration 18

guidance and related issues. 19

MR. BOULET: Thank you. 20

And I would just say to James's earlier 21

point, that really leads me to the logical conclusion 22

that there has to be an open AI system, that there has 23

to be a transparency element added for the public 24

market. And I would just add that just because it's 25

107

Heritage Reporting Corporation

(202) 628-4888

difficult to have transparency doesn't make it 1

impossible. And, frankly, just because it is 2

difficult, that is the road you must travel in order 3

to use copyrighted content. 4

MS. RUBEL: Okay. Thank you. 5

Last question I want to pose to everyone, we 6

have heard, the Office has heard in response to the 7

guidance we released recently that, and I think 8

Heather started making some points earlier in her 9

responses along these lines, that there's possibly 10

some things about the technology or how the users are 11

interacting with the AI technology that the Office 12

doesn't fully understand or appreciate. 13

So I'm interested in any thoughts you have 14

about what the Copyright Office should know about how 15

the AI systems generate content, what the 16

participation of users might be in different AI 17

models, and any other feedback you have about the 18

guidance that the Copyright Office released. 19

Daniel, why don't we start with you. Daniel 20

Gervais. 21

MR. GERVAIS: Thank you. 22

So, first, as I said in my introductory 23

comments, I support the guidance. I think what's 24

going on -- so, you know, machine learning is almost a 25

108

Heritage Reporting Corporation

(202) 628-4888

synonym of AI these days, and so, you know, machines 1

learn when they produce output. There's a report 2

online I wrote for the European Commission a couple 3

years ago. I looked at every case where people said, 4

look, the machine created this. And at least as of 5

two years ago, there wasn't a single case where there 6

wasn't substantial human edits. 7

But it's called machine learning for a 8

reason, is that the machine will learn those edits, 9

right? And as time passes, there will be more and 10

more productions that have what I call no human cause 11

really. And I think it's perfectly the right decision 12

to say that doesn't have copyright. It would be weird 13

to say you produce something, which is code that 14

produces something, so you have copyright on the code, 15

no question, right, if it’s human written. But 16

whatever the code produces, you also have copyright on 17

that. We don't have that. We've never had that, 18

right? And this is, I think, a line that should not 19

be crossed. 20

So what I think the comment might be would 21

be there are ways in which humans collaborate. The 22

doctrine’s very clear. When you take a public domain 23

picture that's not protected anymore and you do 24

something to it, we know, we can test for whether -- 25

109

Heritage Reporting Corporation

(202) 628-4888

and you can test for whether there was enough done to 1

it to be registrable as a new work, right? So the 2

test is well known. It's just the technology allows 3

humans and the machine to collaborate, as Alex and 4

others mentioned. 5

So I would absolutely urge you to keep the 6

guidance. Just in terms of its application, yes, 7

there may be some ways to dig deeper into technology, 8

but both the transparency and the exclusion of non-9

human works, I think, are very important. 10

MS. RUBEL: Heather? 11

MS. WHITNEY: Thanks. So just on the 12

guidance, a couple of just points very quickly. 13

So one is, as I mentioned, and I think a few 14

people mentioned on the earlier panel, it's not clear 15

when you're reading the guidance what the relationship 16

is between the analysis there and the analysis that 17

the Office presented in Kashtanova letters because, as 18

I mentioned before, that letter is completely 19

dominated by a discussion of predictability, and the 20

inability to predict what the output would look like 21

was basically fatal. 22

But then, in the guidance, there's literally 23

no mention of predictability at all. And while it's 24

possible that the Office has decided that 25

110

Heritage Reporting Corporation

(202) 628-4888

predictability is no longer part of it, it still links 1

to the Kashtanova letter as guidance to artists on its 2

website, and I have heard from many artists through 3

Kashtanova that they're just confused about what that 4

really is supposed to mean. 5

The second thing is that the guidance itself 6

has -- if you look at what the tests are for 7

authorship within the guidance itself, they lead to 8

different outcomes depending on which one you choose 9

to look at. So, for instance, there's a lot of 10

mention about Burrow-Giles with the idea that you are 11

the author if you are the thing, basically, it owes 12

its origin to you. And that's, like, a pretty easy 13

thing to satisfy, and that's what's used a lot, I 14

think, in photography. You click the button or you 15

set up the thing and click a button and now you get 16

the copyright. It would be clear that you would have 17

a lot of copyrightable works if that were the test. 18

And then later you also say that basically 19

you won't register works that are produced by machines 20

or mechanical processes that operate randomly or 21

automatically without any creative input or 22

intervention by a human author. Again, that's like a 23

pretty low standard, any creative input or 24

intervention. But then, at other points, you talk 25

111

Heritage Reporting Corporation

(202) 628-4888

about this idea that you have to have ultimate 1

creative control, that it's you are the one who is 2

contributing all of the traditional elements of 3

authorship. 4

So I think that that is a difficult thing to 5

understand how these things are supposed to work 6

together. And so, in the future guidance, it would 7

just be helpful to sort of make clear what the 8

relationship between those is. 9

MS. RUBEL: Thank you. 10

James? 11

MR. SILVERBERG: Yeah. I really have to 12

echo Heather's sentiments on this point. You know, I 13

think the Copyright Office has the guidance completely 14

correct on the fact that non-human authorship is not 15

copyrightable and is completely correct on the point 16

that human authorship is copyrightable. 17

I think where the guidance, in my opinion, 18

is a little bit off is something that Heather also 19

pointed to, I think, which is that I think the 20

guidance suggests that a substantial contribution 21

needs to be made in order for the human authorship to 22

be copyrightable. But, under the law, I think the 23

standard for copyrightability of a work has a somewhat 24

lower threshold than substantial contribution. 25

112

Heritage Reporting Corporation

(202) 628-4888

And so I think the devil really is in 1

details. And I think greater clarity could be given. 2

I think maybe Daniel suggested that examples be given 3

in order to help clarify what you say in order to 4

render the subject matter of your application 5

copyrightable or not. 6

And just one quick separate point. I think 7

it would be beneficial for the Copyright Office to 8

have a policy where, in circumstances where there's a 9

bulk registration of multiple works, the author can 10

make a singular statement about what the copyrightable 11

or non-copyrightable components are in all of the 12

works so that a photographer registering 750 works 13

doesn't have to do this 750 times. 14

MS. RUBEL: Thank you. 15

Luc? 16

MR. BOULET: Thank you. 17

I just want to state that the U.S. Copyright 18

Office should not, from our position, and does not 19

have the capacity to be engaged in investigations into 20

the boundaries of what is disclaimed as AI-generated 21

and whether or not there was sufficient human 22

involvement in each case. 23

The current registration process for works 24

with AI contribution creates confusion around what AI 25

113

Heritage Reporting Corporation

(202) 628-4888

material should be disclosed in a registration 1

application, and then the guidance applies obligations 2

to disclose AI-generated works without drawing clear 3

lines around what those are. And we don't want the 4

registration process to become more burdensome because 5

the Office is launching investigations into canceled 6

registrations. 7

And the cost of registration is already high 8

with confusion already taking place for many artists, 9

and that standard application severely limits the 10

capabilities of visual artists to register their works 11

in group registrations, leading to greater costs and 12

higher barriers for individual artists. 13

MS. RUBEL: Thank you. 14

Zara? 15

MS. VARIN: Thank you. 16

So, before all of this, in my stress prep, I 17

had the opportunity to watch some of the videos that 18

were featuring other folks from the U.S. Copyright 19

Office, and there were two issues that were cited by 20

Shira Perlmutter, I believe. One is authorship, and 21

the other was ingestion of copyrighted works from 22

machine learning. I've already touched on the latter, 23

so I want to go back to the authorship point, which a 24

couple of other folks have kind of alluded to and 25

114

Heritage Reporting Corporation

(202) 628-4888

mentioned here. 1

I think an important distinction is, at what 2

point does human involvement cross that threshold into 3

authorship, right? And, right now, we don't really 4

have a clear set of guidelines for what constitutes 5

that, especially as it pertains to AI-generated 6

pieces. 7

I really appreciate the way that Creative 8

Commons phrases that copyright law's fundamental 9

purpose is to foster human creativity. Copyright 10

helps protect folks' creative works while ensuring 11

there are clear distinctions and guidelines for what 12

constitutes a copyrightable work. I think, with the 13

growth in this area, there's a great deal of potential 14

to revisit and redefine aspects of what those 15

distinctions even entail. 16

And as other folks have mentioned, we're not 17

necessarily going to be solving all of this, and 18

that's not entirely on the Copyright Office, but 19

you're setting precedents, and I think it's important 20

to consider that as it's going to ripple through every 21

other aspect of every single creative field and all 22

parts in between. 23

I think that there's also, whenever 24

registering stuff, things that I would want to 25

115

Heritage Reporting Corporation

(202) 628-4888

consider seeing. The use of any and all mediums, 1

including generative AI, must be disclosed, which I 2

think that that's already part and parcel of the 3

process, but it needs to be, I think, more 4

distinctively and clearly communicated. It's digital 5

work. This part kind of sucks for some artists, but 6

you can record and maintain a full record of the 7

process that goes into creating something. 8

I regularly screen share or share a process. 9

I use a program called Procreate that can start to 10

finish share what the process looks like on pieces. 11

That is an option that digital artists have and I 12

think is something that, if not necessarily that but 13

something similar, could and should be considered when 14

evaluating the degree of human authorship for any sort 15

of AI-generated works. 16

I think there's another factor here that 17

we've skirted around, and that's fair use. The fourth 18

factor in particular is pertaining to the effect on 19

the potential market, and that's of extreme concern. 20

If somebody's art style becomes a popular prompt 21

fodder and the Internet is inundated with countless 22

generated images in that artist's style, how does that 23

impact the artist? The short version is it sucks. 24

And I've seen many people that have been directly 25

116

Heritage Reporting Corporation

(202) 628-4888

harmed and have lost work because of the lack of 1

really distinct enforcement and ethical development 2

around these AI tools. 3

So I could keep going on the list, but I 4

know that we're running out of time, and I want to 5

give other folks a chance to speak. 6

MS. RUBEL: Okay. Thank you. 7

I think we'll hear last from Paul. 8

MR. REINITZ: Great, and I'll try to be 9

brief. You know, I just want to say this has been a 10

really good discussion, and I think that really good 11

points brought up, especially on this issue. 12

I also would like to commend the Office on 13

putting this guidance out there. I know it's really 14

hard to get your hands around it. And, you know, one 15

of the things could have just been ignore it. And, 16

you know, I think that you've done a really great job 17

in trying to get it out there. Is it perfect? No, 18

but, you know, it will get there. 19

I think one thing in terms that could be 20

helpful for you is to understand that it's not 21

necessarily all or nothing when using generative AI. 22

It's not like I have created an image and it's 23

completely generative AI. There's also applications 24

of it where you could be considered using it to, like, 25

117

Heritage Reporting Corporation

(202) 628-4888

modify an existing image. And I think that, you know, 1

that needs to be taken into account. 2

Yes, the disclosure still needs to exist. 3

And, you know, to simplify things, I think that there 4

should be disclosure anytime there's generative AI 5

used. But, you know, the line might be different. 6

For instance, if somebody, you know, in Photoshop, 7

some of these tools are already using AI, right? And 8

if someone is using AI to, say, correct sharpness or 9

color, and that's something that has traditionally 10

been okay in part of the creative process, you know, 11

using other tools, you know, that needs to be 12

considered, that that's different than creating a 13

generative image just from a prompt. 14

MS. RUBEL: Thank you. And thanks to 15

everybody. This has been a really productive session. 16

We appreciate all of your comments. And I like that 17

we've styled this as a listening session. Really, the 18

purpose was for all of us to be able to listen to one 19

another, and thanks to everyone for sticking with us 20

and sharing your perspectives with us as well. 21

I'll pass it over to Mark briefly. 22

MR. GRAY: Great. Thank you, everyone. So 23

this concludes our second panel. Thank you very much 24

for everyone on the panel for this session. 25

118

Heritage Reporting Corporation

(202) 628-4888

(Whereupon, a brief recess was taken.) 1

MS. MANGUM: Welcome back, everyone, and 2

good afternoon. My name is Jalyce Mangum. I'm an 3

Attorney-Advisor here in the Office of the General 4

Counsel. 5

We're going to begin the final session in 6

just a few. But, first, for those of you who are just 7

joining us, I've got a few Zoom housekeeping points. 8

If you're joining this session, but you're 9

not a speaker for this particular session, please keep 10

your camera off and your mic on mute. 11

We are recording this session today, and the 12

recording will be available on our website in a few 13

weeks. The transcription function is also activated 14

for anyone who wants to follow along that way. 15

In this session, we're going to ask each of 16

our speakers to give brief remarks on the subject of 17

artificial intelligence and visual art. Each person 18

will be limited to two minutes, and I will be watching 19

the time to keep us moving along. 20

We'll call on the speakers in the order 21

listed on the agenda, and we're going to start first 22

with Tom Lockley at Grey Owl Audio. 23

So, if, Tom, you're on, you can get started. 24

MR. LOCKLEY: Awesome. 25

119

Heritage Reporting Corporation

(202) 628-4888

So, before I start speaking, I would just 1

like to thank USCO as well as the panelists for coming 2

here today and speak on this issue. AI and its role 3

in art is a deeply complex issue, and it's good to 4

hear from all stakeholders involved. 5

My name is Tom. I'm a YouTuber, writer, 6

educator, and investor. Over the past year, I've had 7

the chance to participate in a number of AI art 8

communities, including Midjourney and Stable 9

Diffusion. I've also created an essay called The 10

Defense of AI in the Artistic Fields in which I 11

explore technologies like diffusion models and CLIP, 12

along with the surrounding regulatory contexts 13

involving them. 14

In my time here today, I'd like to share an 15

adjacent though nonetheless important thought. AI is 16

a powerful tool for equity and expression in the arts. 17

Creators who lack access to expensive tools in studios 18

or who are unable to produce art in a traditional 19

manner due to illness or disability can rely on AI to 20

provide alternate and often more manageable forms of 21

expression. 22

I would know. Something I didn't mention in 23

my introduction is that I have ulcerative colitis, a 24

disease in which a person's immune system attacks 25

120

Heritage Reporting Corporation

(202) 628-4888

their digestive tract. This can put me out of action 1

anywhere from weeks to months. It also manifests in 2

other ways, from joint pain to a sometimes 3

insurmountable fatigue. I can get around half the 4

issue with tools like scheduled posts and videos, but 5

AI fills the other half. I use it for thumbnails, 6

book covers, character concept art, and branding. Yet 7

the amount of human direction that goes into my work 8

shouldn't be discounted. The process does not end 9

with one prompt, and it can take anywhere from hours 10

to days for me to make a piece that matches or evolves 11

from the initial vision I had for it. 12

I'm not alone in this. In conducting 13

research for my essay, I had the pleasure of speaking 14

with a number of creators who work through their 15

disabilities via AI, some of them for love of the 16

field, while others use their art to generate income. 17

This regulation of this nascent community can lead to 18

damaging outcomes for those who participate in it. A 19

lack of protection for AI artists unfairly penalizes 20

them by opening the door to infringement and art theft 21

by larger players who have both financial means and 22

business motivations to lock up the industry in their 23

favor. 24

Regardless of the outcome of today's 25

121

Heritage Reporting Corporation

(202) 628-4888

conversation, lessons we take from it and the 1

precedents that we eventually set will come to define 2

the next several decades. Let's get it right and make 3

a system that works for all types of creators. 4

Thank you. 5

MS. MANGUM: Thank you, Tom. We really 6

appreciate your comments and your perspective. That 7

was really interesting. 8

Next, we've got Matthew Cunningham from 9

Cunningham Concept Design. 10

Matthew, you're on. 11

MR. CUNNINGHAM: Hi there. Thanks, 12

everyone, for having me and thanks for the great 13

presentation so far. It's been really enlightening 14

listening to everybody's angle on the whole thing. 15

My background, I'm a concept designer in the 16

feature film and television industry. You may know my 17

work from shows like Star Trek: Picard, Season 3, 18

Amazon’s Citadel that just recently was released, as 19

well as a couple of Godzilla movies. I also work as a 20

car designer for companies like BMW, Toyota, Hyundai, 21

you name it. And I'm a former labor union leader with 22

the Art Directors Guild. And I also teach design at 23

the Art Center College of Design in Pasadena, which is 24

the top design school in the world arguably. 25

122

Heritage Reporting Corporation

(202) 628-4888

My reason for being here is I'm part of a 1

larger coalition based in Los Angeles, and what our 2

concerns are fundamentally have to do with the 3

certain companies who are scraping the data and 5

identity of artists. My concern is multi-pronged but 6

I would say primarily is for the artists who are 7

immediately being impacted economically, but also for 8

successive generations of artists and also for the 9

generations of past artists, where I think quite a lot 10

has been discussed regarding the ability to imitate 11

the style of artists. 12

This has very immediate economic impact, but 13

I think also, through the historical lens, going back 14

and, you know, if you visit a museum, you'll see some 15

works of artists that are quite prominent, but these 16

might become muddy waters in due time given enough of 17

an evolution with the technology. 18

So I'm just simply here to speak for artists 19

in both organized labor, unofficially, but mostly for 20

myself and my colleagues in the film industry. 21

MS. MANGUM: Well, thank you, Matthew. We 22

appreciate you being here and for you sharing your 23

thoughts. 24

Next, we have Brian Frye with the University 25

123

Heritage Reporting Corporation

(202) 628-4888

of Kentucky College of Law. 1

Brian? 2

MR. FRYE: Thank you for inviting me to 3

speak. I'm Brian L. Frye, Spears-Gilbert Professor of 4

Law at the University of Kentucky College of Law. 5

I think we're asking the wrong questions 6

about AI and copyright. Everyone is asking whether a 7

training an AI algorithm infringes copyright. The 9

obvious answer is no and no. Copyright only protects 10

works created by people. AI doesn't even create 11

works. It generates content, which we consumers 12

interpret as works. 13

Many years ago, Roland Barthes predicted the 14

death of the author, and AI has written the author's 15

obituary. Likewise, training and AI algorithm doesn't 16

and shouldn't infringe copyright. AI algorithms don't 17

copy works, they merely catalog rhetorical conventions 18

and then deploy them to create conventional content. 19

We should be asking what AI can tell us 20

about what copyright should protect and why. 21

and the Copyright Office have struggled to define 23

creativity. Maybe AI can help. 24

An AI algorithm is essentially a nonsense 25

124

Heritage Reporting Corporation

(202) 628-4888

generator designed to produce banalities. In other 1

words, AI is uncreative by design. An AI algorithm is 2

a machine for regurgitating conventional wisdom. 3

Indeed, we are amused when an AI hallucinates and 4

fails to satisfy our pedestrian expectations. But we 5

can be just as boring as any AI, and there's no point 6

in copyright protecting banalities. 7

Maybe AI can help us limit copyright to 8

works that are actually creative. It's easy, just ask 9

AI to evaluate the creativity of works created by 10

people to determine whether they deserve copyright. 11

No one knows a fake like a faker, and AI is designed 12

to identify banality. That's what makes it a killer 13

app. 14

We don't know how to identify creativity, 15

but AI can tell us what isn't creative, and maybe 16

that's good enough to tell us what is creative, if 17

anything. 18

Thanks again. I'm on Twitter @brianlfrye. 19

MS. MANGUM: Thank you, Brian. Really 20

interesting comments, and thanks for the Twitter 21

tagline. We appreciate that. 22

Next, we've got Nettrice Gaskins, who is a 23

freelance artist. 24

MS. GASKINS: Thank you, and thanks for 25

125

Heritage Reporting Corporation

(202) 628-4888

having me. 1

In 2017, the Andy Warhol Foundation launched 2

a preemptive lawsuit against photographer Lynn 3

Goldsmith, who captured photos of the late musician 4

Prince in 1981 for Newsweek. Warhol was later 5

commissioned by Vanity Fair in 1984 to produce a pop 6

art recreation of one of the images after licensing it 7

for $400. However, Warhol continued to use the image 8

for his portfolio, taking his own spin on the original 9

photo. 10

In response to being sued for her own 11

copyrighted photograph, Goldsmith filed a countersuit, 12

and she lost. New York State District Judge John G. 13

Koeltl ruled in favor of the Warhol Foundation. 14

Koeltl argued that though Warhol used Goldsmith's 15

photograph as a reference image, he removed nearly all 16

of the photograph's protectable elements. Thus, 17

Warhol did not violate the photograph's copyright. 18

Like Warhol, my clients license images from 19

-- for me to use as references for the creation of AI 20

art. Mimicking these references is not a goal. 21

Reimagining, recasting, remixing, and restyling them 22

are. 23

My first commission was in 2019, and I've 24

been using a variety of AI tools to make art since 25

126

Heritage Reporting Corporation

(202) 628-4888

2016, long before prompt-based tools. I use AI to 1

produce variations on reference images, and I use text 2

prompts with keywords that the AI recognizes in order 3

to generate unique visuals. I use image editing 4

software to revise visual elements as well as layer 5

images I've done, and I've done this since the early 6

1990s. Just one of my AI artworks merges multiple art 7

styles and techniques, posing subjects in ways that go 8

far beyond the original image references. 9

For me, as someone with a traditional fine 10

arts background, AI broadens what is possible for 11

artistic production, and these new tools have enabled 12

me to become an art director, maker, and curator of my 13

work. This puts me in a pipeline that previously 14

excluded me. Thank you. 15

MS. MANGUM: Thank you, Nettrice. We 16

appreciate you sharing your experience. Really 17

interesting remarks. 18

Greg Hopwood was unable to join us, so we're 19

going to skip right on to Phuc Pham from the Freelance 20

Solidarity Project. 21

MR. PHAM: Hi, everyone. My name is Phuc, 22

and I am here today to lay out my perspective as a 23

photo editor and a photographer, as well as to 24

represent conversations I've been having with 25

127

Heritage Reporting Corporation

(202) 628-4888

freelancers I organize alongside with in the Freelance 1

Solidarity Project, which is the Digital Media 2

Division of the National Writers Union. 3

Editors of born-digital works that are 4

primarily distributed online, freelance digital 5

content creators, not just those working in the visual 6

arts space, like photographers, animators, 7

illustrators, and graphic designers, but also in every 8

other corner of the media industry, are among those 9

the most likely to be impacted by generative AI 10

technologies. And what I'd like to focus on during my 11

time right now is just a simple example of, like, the 12

sort of innumerable images that train these AI systems 13

to generate their output. 14

You know, these companies that develop AI 15

engines largely obscure details of how their 16

technologies work. As such, it's nearly impossible 17

for an individual artist to seek recompense for their 18

copyrighted work showing up in these training data. 19

Those who wish to remove their images from 20

these data sets have found tools such as "Have I Been 21

Trained?" to even determine exactly which images were 22

used to train these systems. You know, following that 23

discovery, an artist would have to register their 24

works in order to even defend a copyright claim. You 25

128

Heritage Reporting Corporation

(202) 628-4888

know, for example, a photographer whose digitally 1

published works were scraped to train these systems, 2

those works could number in the hundreds of thousands. 3

And at $55 to register 10 images, that quickly can 4

become an exponential amount of money. 5

This process is not only economically 6

draining, could be, but would also demand time and 7

focus away from an artist who definitely won't have an 8

entire corporate department to sort of support these 9

Images or a lot of these other bigger outfits that 11

are, you know, bringing forth lawsuits for this type 12

of usage of their work. 13

And beyond the impracticality of defending 14

individual copyright claims, as someone who works with 15

photographers to commission original works and who 16

organizes to improve industrywide working conditions, 17

I'm simply disturbed, you know, by the implications of 18

these systems and how quickly they are sort of can 19

quickly replace the human spirit and the novelty that 20

working artists bring to these types of works. It 21

seems to me like an untenable future for visual 22

artists who are born-digital and otherwise. 23

MS. MANGUM: Well, thank you. We appreciate 24

your remarks. Really interesting. 25

129

Heritage Reporting Corporation

(202) 628-4888

Next, we've got Ankit Sahn from Ajay Sahni 1

Associates. 2

Ankit? 3

MR. SAHN: Thank you. Good afternoon, 4

esteemed speakers and members of the United States 5

By way of an introduction, I'm Ankit Sahn. 7

I'm an IP lawyer based in India. And I'm the owner of 8

the RAGHAV AI painting tool. I filed what was perhaps 9

the first application at the USCO and the CIPO in the 10

Indian copyright office where a human and AI were 11

identified as co-authors back in 2021. I'm grateful 12

for the opportunity to present my views today. 13

As we continue to rely on AI tools to 14

produce works of art, music, literature, and other 15

creative outputs, creators must be assured that their 16

works will be protected under copyright law. Denying 17

result in a chilling effect on creativity and 19

innovation. 20

Just as when cameras were invented, humans 21

moved up the value chain by becoming photographers, 22

using AI-based tools to create demands human beings to 23

move up the value chain once again. We must recognize 24

the increasing role that AI is playing in the creative 25

130

Heritage Reporting Corporation

(202) 628-4888

processes across all industries while also 1

acknowledging the value of human creativity and talent 2

and thus maintaining that balance. 3

The definition of human creativity in the 4

context of copyrightability, therefore, has to be 5

reconsidered. Works created with the assistance of an 6

AI-based tool could be considered as a separate 7

category of copyright, perhaps with reduced duration 8

and scope of protection to incentivize human effort 9

and promote innovation. 10

As Mahatma Gandhi said, “The future depends 11

on what we do in the present.” Protecting AI-12

generated outputs under copyright law is a crucial 13

step in this direction, and by providing necessary 14

protection and recognition, we can continue to foster 15

innovation, creativity, and ultimately benefit the 16

creative industries, as well as creators of these AI-17

based tools. 18

If AI-assisted works are not protected, on 19

the contrary, it could lead to creators suppressing 20

the fact that they use the assistance of an AI tool to 21

create a work, which would in turn be unfair to 22

creators who have utilized AI tools to enhance their 23

creative outputs and in any case would not be 24

reflective of the correct factual position on the 25

131

Heritage Reporting Corporation

(202) 628-4888

Training on proprietary data, as many 2

speakers pointed out today, is one of the key issues. 3

Collective or possibly compulsory licensing of data 4

could be explored as a possible solution. 5

In conclusion, copyright law must evolve 6

with the times, recognizing the role that AI is 7

playing in the creative industries. Protecting AI-8

generated outputs under copyright law, I believe, is a 9

crucial step in this direction. 10

I am grateful once again for this 11

opportunity. Thank you for your time and attention. 12

MS. MANGUM: Thank you, Ankit. We 13

appreciate your remarks. 14

Next, we're going to move on to Jeffrey 15

Sedlik from the PLUS Coalition. 16

MR. SEDLIK: Thank you. 17

MS. MANGUM: Jeffrey? 18

MR. SEDLIK: Thanks. Yes, I'm Jeff Sedlik. 19

I'm President of the nonprofit PLUS Coalition at 20

PLUS.org. We're currently collaborating with the IPTC 21

on establishing metadata fields to communicate 22

information about works that may incorporate AI and to 23

provide for the expression of permissions and 24

constraints on the use of AI or use of visual works 25

132

Heritage Reporting Corporation

(202) 628-4888

for AI training and for generative AI. 1

I'm also the former president of the 2

American Photographic Artists and a photographer, 3

filmmaker, and professor at the Art Center College of 4

Design, where I teach copyright law and licensing. 5

Like other visual artists, copyright is at 6

the core of my business. To sustain it and to support 7

my family, I rely on revenue from licensing my works 8

throughout the life of my copyrights. I offer 9

licenses of my work to all manner of clients for all 10

manner of media in all manner of purposes, including, 11

importantly, artist reference use, the use of my works 12

by other artists to adapt my work in new derivative 13

works. The exclusive right to adapt a work is often 14

overlooked but is, in fact, one of the core rights of 15

the exclusive right to reproduce, distribute, display, 17

or perform a work. 18

The practice of offering, seeking, and 19

granting artist reference licenses has been in broad 20

use for more than a century, and the use of visual 21

works as AI image prompts falls squarely within the 22

definition of artist reference use. In fact, many of 23

us have offered up our works for paid licensing for AI 24

training and artist reference use in AI generative 25

133

Heritage Reporting Corporation

(202) 628-4888

works for years. 1

This fact, combined with the fact that many 2

stock photo agencies are now routinely licensing their 3

works for AI training, establishes that a marketplace 4

for licensed use of visual works for AI training 5

exists and is therefore usurped when works are used 6

for AI training without authorization. Meanwhile, 7

thousands of copies of my work are included in the 8

LAION database and other databases of, collectively, 9

billions of images used to train AI systems, many 10

without attribution. The more I license my works, the 11

more copies are included in the training sets, and the 12

closer the appearance of generated AI works to my 13

original creations. But, as a professional, I must 14

license my work in order to sustain my business. It's 15

a Hobson's choice. 16

Some suggest that contrary to the core 17

provisions of copyright law, anything on the web is 18

fair game and that if artists don't want their works 19

used for AI training or image prompts, we should 20

remove our works from the web. But the web is the 21

primary medium for licensed use of our creations, and 22

we necessarily depend on the web to monetize our 23

works. 24

It's also very important to recognize that 25

134

Heritage Reporting Corporation

(202) 628-4888

many, perhaps the majority, of copies of our visual 1

works displayed on the web are infringing copies made 2

without our knowledge or permission. These infringing 3

copies are then blindly scraped for inclusion in 4

databases like LAION, which is arguably a database 5

built on infringements. 6

MS. MANGUM: Jeff -- 7

MR. SEDLIK: In addition, we must not forget 8

that the widespread aggregation of visual works -- 9

MS. MANGUM: I'm sorry. We’re going to have 10

to -- I'm going to have to interrupt. It's been a 11

little over two minutes. 12

MR. SEDLIK: Okay. 13

MS. MANGUM: But feel free definitely to 14

submit your remarks when there are opportunities later 15

to do so. Thank you so much for coming. 16

MR. SEDLIK: No problem. 17

MS. MANGUM: Next, we're going to move on to 18

Patricia Sigmon, who is an artist and art director. 19

MS. SIGMON: Thank you so much. My name is 20

Patricia Sigmon, and I am a professional artist and 21

art director who has primarily worked in 2D and 3D 22

animation. I will be speaking on my personal 23

experiences with AI-generated images and my concerns 24

about how they have compromised the ability of artists 25

135

Heritage Reporting Corporation

(202) 628-4888

to retain work, as well as how those problems may 1

multiply in the future without more regulation. 2

At a previous position as an art director, 3

my team of artists realized that character design 4

references we had been given were AI-generated. 5

Company leaders did not consult me about the use of AI 6

beforehand. The team came to me with several 7

concerns, including fears about their jobs being 8

replaced, ongoing legislation against AI, ethical 9

concerns, and the poor quality of the images they were 10

given. While AI-generated images seem like they would 11

be an easy shortcut for gathering references, they 12

often include nonsensical details that do not 13

translate well to designed outfits in 3D models. 14

A major concern was also that work produced 15

by artists for the company could be used to train AI 16

without their consent. There has been ample evidence 17

from previous panelists that the goal of many people 18

who use this technology is to imitate the styles of 19

existing artists. It is not a stretch to imagine that 20

without legal guarantees against this, a company could 21

hire artists, train AI with their work, and eventually 22

use it as a replacement for them as the output quality 23

improves. 24

I compiled all these concerns and statements 25

136

Heritage Reporting Corporation

(202) 628-4888

from the artists and brought them to the heads of the 1

company. I was fired two days after that meeting. 2

I am part of the first wave of artists 3

affected by major companies adopting AI image 4

generators. My goal is not to totally disavow the use 5

of this technology or prevent individuals from using 6

it. I understand that the march of progress cannot be 7

stopped and that AI image generators have uses outside 8

a professional setting. The problem is that companies 9

will always be more concerned with their bottom line 10

than the people they employ. It is imperative to 11

protect the ability of artists to earn a living from a 12

craft that they've perfected through a lifetime of 13

practice without that work being cannibalized. 14

Thank you for your time and for inviting me 15

to this discussion. I think it's really important to 16

have, and I've appreciated hearing everyone's 17

perspectives. 18

MS. MANGUM: Thank you so much for sharing 19

your experience with us, Patricia. We really 20

appreciate it. 21

Last but not least, we've got Delanie West 22

from Be Super Creative. 23

Delanie? 24

MS. WEST: Thank you, Patricia. Thank you 25

137

Heritage Reporting Corporation

(202) 628-4888

so much for that perspective. I appreciate you too. 1

Thanks for the invitation to share my 2

perspective. I'm Delanie West, Founding Creative 3

Director of Be Super Creative. I have 30 years 4

experience in creative marketing, business, and brand 5

development, and I've led creative and product 6

development for U.S., European, and Japanese brands. 7

I serve in a leadership capacity at Women in Toys, 8

Licensing & Entertainment, as well as the Graphic 9

Artists Guild, and other creative industry 10

organizations, but the views I express today are my 11

own. I'm here today to share a perspective of a small 12

business owner doing the work of developing creative 13

for product development. 14

As a value creator who celebrates this new 15

technology, I'm also concerned about the impact of AI 16

on consent, credit, and compensation. AI can 17

revolutionize the creative process, but we must ensure 18

ethical methods to generate outputs that respect 19

intellectual property and current law. 20

I've long worked aside legal IP counsel to 21

protect and defend copyright and patents, and the same 22

process is necessary for ethical AI implementation. I 23

ask that we embrace AI while being mindful of ethical 24

considerations and adhere to the current and future 25

138

Heritage Reporting Corporation

(202) 628-4888

intellectual property laws. 1

The use of AI in product development has 2

tremendous potential for boosting creativity and 3

productivity, but we must approach it with care and 4

responsibility to ensure practice for consent, credit, 5

and compensation. 6

In summary, AI has been a game changer for 7

product development timelines for me, but we creators 8

need the guardrails that enable users to respect the 9

rights of all parties involved. Thank you. 10

MS. MANGUM: Thank you so much. We really 11

appreciate everyone who shared their perspective and 12

their experience. 13

I'm going to turn it over to Mark to close 14

us out. 15

MR. GRAY: Thank you very much, Jalyce. 16

And to echo Jalyce’s remarks, we truly do 17

appreciate, both I, as well as the rest of my 18

colleagues here in the U.S. Copyright Office, we do 19

appreciate you taking the time to talk to us today and 20

to share all of this information. Of course, we're 21

going to continue to think about all of these issues 22

and all of the things that you've told us as we work 23

on our initiative examining issues of copyright law 24

and policy and the intersection with artificial 25

139

Heritage Reporting Corporation

(202) 628-4888

intelligence technology. 1

So looking forward, we have two more 2

listening sessions on the calendar. Our next session 3

is on Wednesday, May 17, which will be focused on 4

audiovisual works, including movies and video games. 5

Our final session is going to be on May 31, focused on 6

musical works and sound recordings. The audiovisual 7

session, unfortunately, is no longer accepting 8

signups. We are wrapping up the selections for those 9

panels now. The music session remains open until 10

May 10. So, if you are interested in speaking about 11

musical works or sound recordings, you can find more 12

information on our website at copyright.gov/ai. 13

As a reminder, we will be making video 14

recordings of this and of the other sessions available 15

to the public on our website. We're aiming for about 16

a three-week turnaround on those. 17

And, as we've said before and as we will 18

continue to remind people, this is not the final word. 19

This is not the final chance to speak to us. There 20

are many more chances coming down in the future. We 21

look forward to hearing from you and thank you very 22

much for spending your time with us today. Thank you. 23

(Whereupon, at 4:00 p.m., the listening 24

session in the above-entitled matter adjourned.) 25

140

Heritage Reporting Corporation

(202) 628-4888

CERTIFICATE

CASE TITLE: Copyright and Artificial Intelligence

Visual Arts Listening Session

HEARING DATE: May 2, 2023

LOCATION: Washington, D.C.

I hereby certify that the proceedings and

evidence are contained fully and accurately on the

tapes and notes reported by me at the hearing in the

above case before the Library of Congress, U.S.

Date: May 2, 2023

Martha Nelson

Official Reporter

Heritage Reporting Corporation

Suite 206

1220 L Street, N.W.

Washington, D.C. 20005-4018