Generative Artificial Intelligence and the Creative Economy Staff Report: Perspectives and Takeaways

Generative Artificial Intelligence

and the Creative Economy Staff Report:

Perspectives and Takeaways

December 2023

US Federal Trade Commission

Introduction ........................................................................................................................ 3

Generative AI and Recent Technological Developments ................................................... 4

FTC’s Interest and Role in AI.............................................................................................. 5

Event Summary ................................................................................................................... 8

Themes ................................................................................................................................ 9

How did participants say data is being obtained or collected?....................................... 9

What harms did participants say they were currently facing?...................................... 12

How did participants view proposed consent defaults?................................................ 14

What are participants doing to understand and address generative AI? ..................... 17

Potential Areas of Further Inquiry.................................................................................... 20

Conclusion ......................................................................................................................... 21

Appendix............................................................................................................................ 22

Appendix I: Participant Bios.......................................................................................... 22

Appendix II: Roundtable Quote Book ........................................................................... 25

“We’re not anti technology.” ...................................................................................... 26

“Our members are keenly aware that their works are being used by AI systems.”... 27

Platforms using user-generated content for training ................................................ 28

Contracts..................................................................................................................... 28

Body “Scans”............................................................................................................... 29

“Losing work” ............................................................................................................. 29

Impacts on opportunities ........................................................................................... 30

“Time spent” ............................................................................................................... 31

Loss of money and/or compensation......................................................................... 31

Accuracy and Fairness................................................................................................ 31

Diminished value of work........................................................................................... 32

Loss of ability to compete........................................................................................... 32

Reputational damage.................................................................................................. 33

Deepfakes, Impersonations, and/or Fan confusion .................................................. 33

Consent, Permission, and/or Opt-out vs. opt-in ....................................................... 35

Machine unlearning.................................................................................................... 36

Transparency and disclosure ..................................................................................... 36

Changing terms of service .......................................................................................... 37

Policy and legislative efforts....................................................................................... 37

Lawsuits ...................................................................................................................... 37

Self-initiated research and investigations.................................................................. 38

Collective bargaining .................................................................................................. 38

Regulation................................................................................................................... 39

Consent ....................................................................................................................... 40

“A system that is opt-in. Not opt-out.”....................................................................... 40

Credit and Transparency............................................................................................ 40

Compensation............................................................................................................. 41

Licensing..................................................................................................................... 41

Introduction

Generative AI is a revolutionary technology with vast implications for how people live,

work, and communicate around the world. There is little doubt that it will be highly

disruptive to markets, businesses, and the public.

As noted in President Biden’s

Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial

Intelligence, “Harnessing AI for good and realizing its myriad benefits requires

mitigating its substantial risks...,” demanding “a society-wide effort that includes

government, the private sector, academia, and civil society.”

The FTC has a role to play

in that broader effort. The FTC is an independent federal agency charged with

promoting fair competition and protecting consumers, workers, and businesses from

unfair or deceptive trade practices. The development and deployment of AI-based

technologies may be proceeding swiftly, but the FTC has decades of experience

monitoring and adapting to novel technologies in new and rapidly changing markets.

Given the FTC’s interest in the AI space and recent emphasis on soliciting input from

individuals who are typically underrepresented in policymaking discussions, FTC staff

recently held a public event for creative professionals entitled, “Creative Economy and

Generative AI.” FTC staff releases this report to summarize the information provided to

the FTC in that roundtable. Some of the issues surfaced at the event implicate the FTC’s

enforcement and policy authority, though some certainly go beyond or outside the FTC’s

jurisdiction.

This report begins by briefly summarizing the technological developments in AI and

generative AI that make this roundtable and report timely. Next, it explains the FTC’s

jurisdictional interest in AI. Finally, it summarizes and memorializes the information

provided by roundtable participants. This staff report is intended as a useful resource

for the legal, policy, and academic communities who are considering the implications of

generative AI.

See, e.g., Lina M. Khan, We Must Regulate A.I. Here’s How, N.Y. Times (May 6, 2023),

https://www.nytimes.com/2023/05/03/opinion/ai-lina-khan-ftc-technology.html.

https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-

safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/

https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-

safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/

Generative AI and Recent Technological Developments

The past year has seen the emergence of tools powered by artificial intelligence that can

generate outputs like text, images, and audio on command. These tools are commonly

referred to as “generative AI.” To output different kinds of content, these models must

be built using vast amounts of existing work.

For example, large language models such

as PaLM 2

and Llama 2

rely on large datasets of text that have been “tokenized”–

divided into smaller chunks of words or even parts of words – which are then analyzed

for patterns that can be reproduced.

Image generators like Stable Diffusion are reliant

on images, paired with their captions, to fuel their models.

The ways these tools are

built and the content the tools output have garnered attention and concern, particularly

from those whose work is being used and potentially replaced.

See, e.g., https://developers.google.com/machine-learning/data-prep/construct/collect/data-size-

quality

See Google, PaLM 2 Technical Report https://ai.google/static/documents/palm2techreport.pdf

See Hugo Touvron et al., Llama 2: Open Foundation and Fine-Tuned Chat Models, arXiv (July 19,

2023), https://arxiv.org/pdf/2307.09288.pdf.

See Madhumita Murgia et al., Generative AI exists because of the transformer, Financial Times (Sep.

12, 2023), https://ig.ft.com/generative-ai/.

See Kashmir Hill, This Tool Could Protect Artists From A.I.-Generated Art That Steals Their Style, The

New York Times (Feb. 13, 2023), https://www.nytimes.com/2023/02/13/technology/ai-art-generator-

lensa-stable-diffusion.html.

FTC’s Interest and Role in AI

The FTC’s economy-wide mission has, over its century long history, adapted to the

development and deployment of new technologies, many of which pose novel and

important challenges to the consumers, workers, and honest businesses who depend on

markets being free and fair. AI is the latest of such challenges.

The Commission’s enforcement authority derives primarily from Section 5 of the FTC

Act, which prohibits unfair or deceptive acts or practices and unfair methods of

competition. The Executive Order on the Safe, Secure, and Trustworthy Development

and Use of Artificial Intelligence encourages the FTC to consider whether to exercise its

existing authorities, as appropriate, to ensure competition in the AI marketplace and to

protect the public from harms that may be enabled by AI.

From an enforcement perspective, the FTC has been using its existing legal authorities

to take action against illegal practices involving AI. For instance, the FTC alleged that

Amazon and Ring used highly private data—voice recordings collected by Amazon’s

Alexa voice assistant and videos collected by Ring’s internet-connected home security

cameras—to train their algorithms while violating customers’ privacy.

The Alexa

matter, in particular, underscored that the Children’s Online Privacy Protection Act

Rule’s prohibition on the indefinite retention of children’s data and similar legal rules

are not superseded by claims from businesses that data must be indefinitely retained to

improve machine learning algorithms.

In recent months, the FTC secured a temporary

restraining order against a business-opportunity seller that claimed to use AI to make

clients profitable and successful.

The FTC has also made clear that a business that

relies on algorithmic decision-making must ensure that the algorithm is not resulting in

EO at 5.3 (a), https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-

order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/

Press Release, Fed. Trade Comm’n, FTC and DOJ Charge Amazon with Violating Children’s Privacy

Law by Keeping Kids’ Alexa Voice Recordings Forever and Undermining Parents’ Deletion Requests

(May 31, 2023), https://www.ftc.gov/news-events/news/press-releases/2023/05/ftc-doj-charge-amazon-

violating-childrens-privacy-law-keeping-kids-alexa-voice-recordings-forever; Press Release, Fed. Trade

Comm’n, FTC Says Ring Employees Illegally Surveilled Customers, Failed to Stop Hackers from Taking

Control of Users’ Cameras (May 31, 2023), https://www.ftc.gov/news-events/news/press-

releases/2023/05/ftc-says-ring-employees-illegally-surveilled-customers-failed-stop-hackers-taking-

control-users.

See Statement of Commissioner Alvaro M. Bedoya, Joined by Chair Lina M. Khan and Commissioner

Rebecca Kelly Slaughter, In the Matter of Amazon Alexa (United States v. Amazon.com, Inc.), at 1 (May

31, 2023), https://www.ftc.gov/system/files/ftc gov/pdf/Bedoya-Statement-on-Alexa-Joined-by-LK-

and-RKS-Final-1233pm.pdf

Press Release, Fed. Trade Comm’n, FTC Action Stops Business Opportunity Scheme That Promised Its

AI-Boosted Tools Would Power High Earnings Through Online Stores (Aug. 22, 2023),

https://www.ftc.gov/news-events/news/press-releases/2023/08/ftc-action-stops-business-opportunity-

scheme-promised-its-ai-boosted-tools-would-power-high-earnings.

unlawful bias.

Furthermore, the FTC charged WealthPress with using deceptive claims

to sell consumers investment-advising services—often claiming that the services’

recommendations were based on an algorithm created by a purported expert.

The rapid development and deployment of AI also poses potential risks to competition.

The rising importance of AI to the economy may further lock in the market dominance

of large incumbent technology firms.

These powerful, vertically integrated incumbents

control many of the inputs necessary for the effective development and deployment of

AI tools, including cloud-based or local computing power

and access to large stores of

training data. These dominant technology companies may have the incentive to use their

control over these inputs to unlawfully entrench their market positions in AI and related

markets, including digital content markets. In addition, AI tools can be used to facilitate

collusive behavior that unfairly inflates prices, precisely target price discrimination, or

otherwise manipulate outputs.

The FTC is empowered under Section 5 of the FTC Act

to protect the public against unfair methods of competition, including when powerful

firms unfairly use AI technologies in a manner that tends to harm competitive

conditions.

With respect to the creative industries, the development and use of AI technology raises

a host of potential competition and consumer protection issues. Although many people

partake in different forms of creative expression as hobbyists or amateurs, millions of

Americans pursue creative work as a profession and many of these artists are self-

Press Release, Fed. Trade Comm’n, FTC Chair Khan and Officials from DOJ, CFPB and EEOC Release

Joint Statement on AI (Apr. 25, 2023), https://www.ftc.gov/news-events/news/press-

releases/2023/04/ftc-chair-khan-officials-doj-cfpb-eeoc-release-joint-statement-ai.

Press Release, Fed. Trade Comm’n, FTC Suit Requires Investment Advice Company WealthPress to

Pay $1.7 Million for Deceiving Consumers (Jan. 13, 2023), https://www.ftc.gov/news-

events/news/press-releases/2023/01/ftc-suit-requires-investment-advice-company-wealthpress-pay-17-

million-deceiving-consumers.

See Staff in the Bureau of Competition & Office of Technology, Generative AI Raises Competition

Concerns, Fed. Trade Comm’n (June 29, 2023), https://www.ftc.gov/policy/advocacy-research/tech-at-

ftc/2023/06/generative-ai-raises-competition-concerns.

The FTC has taken enforcement action to protect competition and innovation in markets for computer

processors with potential AI applications. See Complaint, In the Matter of Nvidia/Arm, Docket No. 9404,

Comm’n File No. 2110015 (Dec. 2, 2021),

https://www.ftc.gov/system/files/documents/cases/d09404 part 3 complaint public version.pdf.

See, e.g., Khan, supra, note 1 (discussing algorithmic price collusion and price discrimination); see also

Huskey v. State Farm Fire & Cas. Co., No. 22 C 7014, 2023 WL 5848164, at *9 (N.D. Ill. Sept. 11, 2023)

(discussing discriminatory insurance claim processing algorithm).

See Fed. Trade Comm’n, Policy Statement Regarding the Scope of Unfair Methods of Competition

Under Section 5 of the Federal Trade Commission Act, Comm’n File No. P221202 (Nov. 10, 2022),

https://www.ftc.gov/system/files/ftc gov/pdf/P221202Section5PolicyStatement.pdf.

employed.

Research has explored generative AI’s early economic impacts on

professional illustrators,

but artists in other creative fields, such as acting or

screenwriting, have also expressed concerns over the ways that generative AI might

affect their livelihoods.

Uncompensated and unauthorized appropriation of creators’

content may also diminish incentives to invest and produce content, affecting quality

over the long term.

Various competition and consumer protection concerns may arise when AI is deployed

in the creative professions. Conduct–such as training an AI tool on protected expression

without the creator’s consent or selling output generated from such an AI tool, including

by mimicking the creator’s writing style, vocal or instrumental performance, or

likeness—may constitute an unfair method of competition or an unfair or deceptive

practice. That is especially true when the conduct deceives consumers, exploits a

creator’s reputation or diminishes the value of her existing or future works, reveals

private information, or otherwise causes substantial injury to consumers. In addition,

conduct that may be consistent with other bodies of law nevertheless may violate

Section 5.

See Office of Research & Analysis, Artists in the Workforce: Selected Demographic Characteristics

Prior to COVID‐19, National Endowment for the Arts (July 2022), .

See Harry H. Jiang et al, AI Art and its Impact on Artists, AIES '23: Proceedings of the 2023

AAAI/ACM Conference on AI, Ethics, and Society, at 363-374 (Aug. 29, 2023),

https://dl.acm.org/doi/10.1145/3600211.3604681.

See Simone Shah, The Writers Strike is Taking a Stand of AI, TIME (May 4, 2023),

https://time.com/6277158/writers-strike-ai-wga-screenwriting/.

See FTC v. Real Prods. Corp., 90 F.2d 617, 619 (2d Cir. 1937) (“A copyright is not a license to engage in

unfair competition.”).

Event Summary

In October 2023, the FTC held a virtual roundtable discussion to better understand the

concerns about generative AI and specifically its impact on creative fields.

Chair Khan,

Commissioner Slaughter, and Commissioner Bedoya provided remarks. The moderated

discussion, which was public, consisted of twelve participants who represented a wide

variety of creative professions, including visual artists, screenwriters, actors,

programmers, editors, musicians, and models.

Each participant gave brief remarks about the changes their field was experiencing with

the advent of generative AI, and how they were responding to those changes. These

remarks were followed by a brief Q&A. A recording of the event

along with a

transcript

are available on the FTC event web page.

During the event, participants acknowledged the potential benefits of generative AI

tools, and many had a long history of incorporating new technologies in their practices.

Participants also described concerns about the ways generative AI could be an avenue

for their own exploitation.

Though participants came from different fields, a few consistent themes emerged:

• Concerns about how their work was being collected and used to train generative

AI models;

• The impact that generative AI outputs are already having on their industry and

livelihoods;

• Issues associated with solutions being proposed by AI companies to address

creators’ concerns; and

• Alternative approaches that creators are pursuing to protect themselves and their

industry, including by enshrining their right to choose whether they want to use

AI in their work through union contracts.

The next section of this report expands on each of the four themes.

See https://www.ftc.gov/news-events/events/2023/10/creative-economy-generative-ai

See https://www.ftc.gov/media/creative-economy-generative-ai-discussion-october-4-2023

See https://www.ftc.gov/system/files/ftc gov/pdf/creative-economy-and-generative-ai-transcript-

october-4-2023.pdf

See Writers Guild of America, Summary of the 2023 WGA MBA

https://www.wgacontract2023.org/the-campaign/summary-of-the-2023-wga-mba.

Themes

How did participants say data is being obtained or collected?

Participants said that their work was being used to train and finetune generative AI

models without their consent. Throughout the event, participants touched on different

ways their work was being collected, either because it was publicly posted online by

themselves or others, or because expansive interpretations of prior contractual

agreements led others to make their art available to train AI. In addition, artists often

produce work for hire and do not own the copyright on those creative works, further

limiting their ability to control how their work is used. Participants said the nature of

their work often leaves them without legal protection, and that the lack of transparency

around data collection practices made it difficult for them to know when their works

were being taken.

Participants said that human-created work, including their own, was

necessary for generative AI tools to function. As Duncan Crabtree-Ireland, the

National Executive Director and Chief Negotiator for SAG-AFTRA, said during the

event, “No AI algorithm is able to make something out of nothing.”

These claims are in line with research and reporting on generative AI. Many factors

impact an AI model’s performance, but one key factor is the quality of the data used to

train it. Recent research has found that not only is it necessary to carefully curate the

data sets used to train a generative AI model,

but removing low-quality content and

even up-sampling higher-quality sources can result in performance improvements.

The AI research community generally agrees that it is critical that the content used

should be diverse

and that generally means it must be created by an actual person.

Generative AI models, said Douglas Preston, an author and participant of the event,

“would be lame and useless without our books. Just imagine what it would be like if it

was only trained on text scraped from web blogs, opinion screeds, cat stories,

pornography and the like.”

In addition to the quality of the data used to train AI models, research suggests another

key factor is the quantity. Popular large language models, for instance, were built using

billions, even trillions, of tokens,

which in turn necessitates similarly massive amounts

of content. Reporting suggests that this content mostly comes from scraping from the

Internet.

Participants said that their work was being taken from the Internet and is

being used to train or finetune generative AI models without their

awareness or consent. John August, a screenwriter and member of the Writers Guild

of America West, said that large language models “have scraped massive volumes of

data, including our words and our unique perspectives.”

For many creative professionals, publicly posting to the Internet is a necessary part of

the job. Steven Zapata, a concept artist and illustrator speaking on behalf of the Concept

Art Association, said that, “to advertise our work, most of us put our art online, on social

media and our personal websites. This leaves it exposed to unethical scraping practices.”

These “unethical scraping practices” have been questioned within academia,

and AI

researchers have clearly stated that using training data that has been obtained from

public sources does not inherently mean that “authorial consent” has been obtained.

In addition to the scraping of work belonging to creative professionals, Bradley Kuhn, a

policy fellow at the Software Freedom Conservancy, pointed out that depending on the

platforms they use, creative professionals “may have already agreed for their own

creative works to become part of the company's machine learning data sets” because of

what is said in those platforms’ terms of service agreements. Several tech companies

made the news over the summer after they updated their terms of service to include

references to building AI with user data,

eliciting backlash from artists in at least one

instance.

In some cases, participants said they weren’t even the ones to post their works online in

the first place. Tim Friedlander, president and founder of the National Association of

Voice Actors, pointed out that, “it's incredibly easy to use AI to capture the voice of an

See Jordan Hoffman et. al, Training Compute-Optimal Large Language Models, arXiv (Mar. 29,

2022), https://arxiv.org/pdf/2203.15556.pdf

See Touvron et al, supra note 5.

See Ilia Shumailov et al., The Curse of Recursion: Training on Generated Data Makes Models Forget,

arXiv (May 31, 2023), https://arxiv.org/abs/2305.17493.

See, e.g, Wayne Xin Zhao et. al, A Survey of Large Language Models, arXiv (Nov. 24, 2023),

https://arxiv.org/pdf/2303.18223.pdf.

See Kevin Schaul et al., Inside the secret list of websites that make AI like ChatGPT sound smart, The

Washington Post (Apr 19, 2023), https://www.washingtonpost.com/technology/interactive/2023/ai-

chatbot-learning/.

See e.g., Signe Ravn et. al, What Is “Publicly Available Data”? Exploring Blurred Public–Private

Boundaries and Ethical Practices Through a Case Study on Instagram, Journal of Empirical Research on

Human Research Ethics, Volume 15 Issue 1-2, at 40-45 (May 19, 2019)

https://journals.sagepub.com/doi/full/10.1177/1556264619850736; See also Antony K. Cooper et. al,

On the Ethics of Using Publicly-Available Data, Responsible Design, Implementation and Use of

Information and Communication Technology, at 159-171 (Mar 10, 2020)

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7134274/.

See Leo Gao, et. al, The Pile: An 800GB Dataset of Diverse Text for Language Modeling, arXiv, at

Section 6.5 (Dec. 31, 2020), https://arxiv.org/abs/2101.00027.

See Matt G. Southern, Google Updates Privacy Policy To Collect Public Data For AI Training, Search

Engine Journal (Jul. 3, 2023) https://searchenginejournal.com/google-updates-privacy-policy-to-collect-

public-data-for-ai-training/490715/; See also Brian Merchant, Column: These apps and websites use

your data to train AI. You’re probably using one right now., Los Angeles Times (Aug. 16, 2023)

https://www.latimes.com/business/technology/story/2023-08-16/column-its-not-just-zoom-how-

websites-and-apps-harvest-your-data-to-build-ai.

See Michael Kan, Artists Drop Twitter Over Elon Musk's Plan to Train His AI Project on Tweets,

PCMag (Aug. 1, 2023), https://www.pcmag.com/news/artists-drop-twitter-over-elon-musks-plan-to-

train-his-ai-project-on-tweets

actor from content available on the internet, and to use that sample to create whole

works for sale or non-commercial distribution.”

Several participants referenced a

dataset called Books3, which The Atlantic reported was built from a library of pirated

books.

Preston, for instance, said that the dataset had, “all my books, all 40 of them on

it, including many different editions.” Another participant, Clarkesworld magazine

founder and editor Neil Clarke, said that anthologies he had edited and published could

also be found in the dataset.

Participants said the nature of most paid creative work means that artists

often have little control over how their creations are later used. Participants

explained that many creative professionals do work for hire, meaning the rights to their

creative works belong to the client or company that hired them. “WGA writers do not

hold copyright to most of the scripts we write; those are works made for hire, so

studios— our employers— hold the copyright,” August said. This means that creative

professionals don’t always have control over how their works are used in the future.

Steven Zapata, a concept artist and illustrator speaking on behalf of the Concept Art

Association, said that with work-for-hire agreements, a visual artist could spend decades

of their life working for a studio and that studio “can take all your work, train a model

on it, and then have a free and tireless replica of you to use in perpetuity.”

Friedlander said that voice actors were experiencing that now: “Contracts we signed

years ago are now being used to justify the inclusion of our audio in synthetic voice

models.”

Some creative professionals have even less insight into what they’ve been forced to give

up. Sara Ziff of the Model Alliance noted that many fashion models do not actually see

the terms that their management companies have agreed to with clients, so many do not

actually know what happens to the digital body scans that some clients are starting to

require.

These uncertainties are exacerbated by the fact that AI developers do not

publicly disclose what works are included in their training data, according

to participants.

Preston said that, “[AI companies] refuse to answer any questions from the Author's

Guild about what data sets they're using, where they're getting their books, and how

they're being used. There's no transparency at all. It's an absolute black hole.”

Music artists have suffered similar problems with generative AI. See Sharon Adarlo, Nicki Minaj

enraged by deepfake video, Futurism (Jul. 12, 2023) https://futurism.com/the-byte/nick-minaj-

deepfake-video; see also Noah A. McGee, Heard the Fake Clip of Drake Rapping ‘Munch’? Here’s Why

It’s Causing Concern for Major Labels, The Root (Apr. 14, 2023) https://www.theroot.com/heard-the-

clip-of-drake-rapping-munch-here-s-why-it-1850336887.

See Alex Reisner, These 183,000 Books Are Fueling the Biggest Fight in Publishing and Tech, The

Atlantic (Sep. 25, 2023), https://www.theatlantic.com/technology/archive/2023/09/books3-database-

generative-ai-training-copyright-infringement/675363/.

Researchers have found that some AI developers have become less open over time about

what can be found in their models’ training data.

When asked about how participants found out about the inclusion of their work in the

training of these models, Umair Kazi stated that “There is a lack of transparency from AI

developers about training data sets, which makes it very difficult to ascertain which

works were actually used to train the models and how.”

What harms did participants say they were currently facing?

Participants’ concerns were limited not just to how their own work was being used.

Throughout the event, participants discussed a wide range of harms they associated with

the outputs of generative AI tools. These harms included the ways that generative AI

could be used make it more difficult to find human-made work, mimic creative

professionals’ unique styles causing market confusion and reputational damage, and

lead to loss of opportunity and income.

Participants said that generative AI outputs are starting to appear in the

venues where creative professionals compete for work, making it more

difficult for consumers and potential publishers to find human-made work.

Kazi, the Authors Guild director of public policy and outreach, said that his group was

already seeing AI being used to generate low-quality eBooks that displace human-

authored books in major online book retailers. In one instance earlier this year, he said,

AI-generated books began dominating the young adult romance bestseller list of a

popular online bookseller.

Neil Clarke, the editor and founder of the sci-fi short story magazine Clarkesworld,

described being inundated with hundreds of submissions that appeared to be AI-

generated, leading him to temporarily close submissions.

Clarke said the magazine

had a standing “no-AI” policy even prior to the influx of submissions, and said his

workload has “easily doubled” as he sorts through a stream of suspicious submissions.

Participants expressed concerns about generative AI tools being used to

mimic their own unique styles, brands, voice and likenesses, which could

allow strangers and former clients to create knockoffs of their work. Karla

See David Gray Widder et. al, Open (For Business): Big Tech, Concentrated Power, and the Political

Economy of Open AI, SSRN Electronic Journal (Aug. 18 2023),

https://papers.ssrn.com/sol3/papers.cfm?abstract id=4543807.

See Jules Roscoe, AI-Generated Books of Nonsense Are All Over Amazon's Bestseller Lists, VICE (Jun

28, 2023), https://www.vice.com/en/article/v7b774/ai-generated-books-of-nonsense-are-all-over-

amazons-bestseller-lists.

See Mia Sato, AI-generated fiction is flooding literary magazines — but not fooling anyone, The Verge

(Feb. 25, 2023), https://www.theverge.com/2023/2/25/23613752/ai-generated-short-stories-literary-

magazines-clarkesworld-science-fiction.

Ortiz, a concept artist and illustrator, said that text-to-image generators can let anyone

produce work “in the style of” a named artist, and that her own name has been

referenced thousands of times by people prompting generative AI systems to produce

work that looked like her own. Similarly, Kazi of the Authors Guild alleged that

generative AI was being used to create unauthorized derivative works in the style of

popular authors, and described one person’s attempts to write the concluding books of

George R.R. Martin’s popular “A Song of Ice and Fire” series.

AI generated knock-offs have found their way onto online book publishing platforms,

said John August of the Writers’ Guild of America West, where they are being sold to

customers who might confuse them with authors’ actual works.

In addition to creating

and selling their own work, now authors have to fend off AI-fueled fraudsters: “They’re

having to fight to get those fakes taken down, and protect their brands,” said August.

Participants said that the threat of AI fakes can also come from former clients.

Friedlander, the NAVA president, gave the example of a New York voice actor who was

let go after working for a company for several years. According to Friedlander, the

company told the voice actor that it finally had enough of the actor’s audio, and the

company was now planning on creating a synthetic version of their voice.

Participants said that when consumers mistake AI-generated work made in

the likeness or style of a particular artist as the actual work of that artist, it

could jeopardize the actual artist’s reputation and ability to earn income.

Zapata, the illustrator speaking on behalf of the Concept Art Association, said AI-

generated work can end up online with the artist’s name attached, even though they

didn’t make the work and have no control over the content or the quality of the piece.

“The negative market implications of a potential client encountering a freely

downloadable AI copycat of us when searching our names online could be devastating to

individual careers and our industry as a whole,” he said.

Participants said that they have also seen generative AI tools being used to harass

creative professionals and confuse consumers. Friedlander referenced an incident from

February, in which anonymous trolls made synthetic versions of multiple voice actors,

and tweeted false audio clips of them divulging their actual home addresses and saying

homophobic and racist slurs.

Jen Jacobsen, the executive director of the Artist Rights

Books allegedly written using generative AI have been published using authors’ actual names. See Ella

Creamer, Amazon removes books ‘generated by AI’ for sale under author’s name, The Guardian (Aug. 9,

2023), https://www.theguardian.com/books/2023/aug/09/amazon-removes-books-generated-by-ai-for-

sale-under-authors-name.

See Joseph Cox, Video Game Voice Actors Doxed and Harassed in Targeted AI Voice Attack, VICE

(Feb. 13, 2023), https://www.vice.com/en/article/93axnd/voice-actors-doxed-with-ai-voices-on-twitter.

Alliance said that generative AI tools have been used to create false depictions of artists

selling products that the artists never endorsed.

“It's not only confusing to fans, but humiliating to the artists themselves and

undermines their public image,” said Jacobsen.

Participants were concerned that creative professionals are already losing

work because of generative AI. Friedlander described a recent incident where a

voice actor in Washington state lost out on an audiobook job when the company told

them it “decided to take the job in-house.” Friedlander said that around the same time,

the company published a press release stating that they would be using a synthetic voice

startup for all their audiobook productions. He said this move was “effectively replacing

all of those human narrators with synthetic voices.”

Kazi of the Authors Guild said that freelance journalists and professional writers of web

and marketing content were reportedly losing work “at an alarming rate.” He described

the plight of an unnamed Guild member, who works as a marketing and web content

writer and reportedly lost three-quarters of their work because their clients switched to

using AI instead.

Ziff, the Model Alliance founder, said that earlier this year a major clothing company

announced that they were creating AI-generated models to increase the number and

diversity of their models.

“In an industry that has historically been discriminatory,

creating digital representations of models of various ages, ethnicities, and body types

rather than hiring and paying a diversity of real models is concerning,” she said.

Ziff pointed out that the use of AI fashion models would not just impact human models.

She said that fashion workers of all kinds—including photographers, stylists, and hair

and makeup artists—were concerned about their use, because it could impact all their

livelihoods.

How did participants view proposed consent defaults?

Some AI developers have started offering people, including creative professionals, the

choice to “opt-out” of their work being used to train future models, through methods

See e.g. Carson Blackwelder et al., Tom Hanks warns fans about fake ad featuring fabricated image of

his likeness, ABC News (Oct. 2, 2023), https://abcnews.go.com/GMA/Culture/tom-hanks-warns-fans-

fake-ad-featuring-fabricated/story?id=103658615.

See Jess Weatherbed, Levi’s will test AI-generated clothing models to ‘increase diversity’, The Verge

(Mar. 27, 2023), https://www.theverge.com/2023/3/27/23658385/levis-ai-generated-clothing-model-

diversity-denim.

such as direct opt-out forms,

voluntarily complying with third-party lists,

and public

commitments to respect the Robots Exclusion Protocol.

Participants raised multiple

concerns about these kinds of opt-out frameworks, ranging from the practical, like not

knowing whether their data was used and, thus, whether opt-out is even needed, to

more fundamental issues with the approach, like shifting the burden from companies to

creators. Participants also discussed the need for solutions that would not only limit the

harm moving forward but also address the harm that has already occurred.

Participants said that opt-outs put the burden on creators to police a

rapidly changing marketplace, where new companies and AI models are

emerging every day. Jacobsen likened these to “a new form of uncompensated labor”

that AI developers are requiring people to perform if they want to avoid being exploited.

“Such burden shifting is not only unfair, it is morally wrong and antithetical to basic

principles of artistic integrity,” said Jacobsen.

August echoed that sentiment, “There are so many companies out there developing and

training AI models, to be forced to continually track all of them down to opt out is an

enormous administrative burden on individual artists.”

Ortiz pointed out that these frameworks require a certain level of technical and legal

expertise about a fast-moving industry, and that raises questions about whether the

choices these frameworks offer are truly accessible to all creators:

Does that mean we have to opt out on each and every one of them? That's a full-

time job. What about if those models update? What about if they don't publicize

and they use third parties? What if those models in the opt-out forms are not an

artist's native language? What about artists who never spend time online or

don't even know this is happening?

Participants wanted AI developers to take an opt-in approach instead. Clarke said that

opt-out frameworks, in contrast to a framework that seeks explicit consent prior to

usage, “is what you get when the fox designs the chicken coop.”

Participants also emphasized that because of the lack of transparency about

what is being used as training data, it was unclear which work they would

need to request to have removed, or even if their work was included at all.

Kazi said the fact that writers and editors were able to check whether their books could

be found in the Books3 dataset was “an anomaly” and not the norm. Because that

See Kyle Wiggers, OpenAI unveils DALL-E 3, allows artists to opt out of training, TechCrunch (Sep.

20, 2023), https://techcrunch.com/2023/09/20/openai-unveils-dall-e-3-allows-artists-to-opt-out-of-

training/.

See Kyle Wiggers, Spawning lays out plans for letting creators opt out of generative AI training,

TechCrunch (May 3, 2023), https://techcrunch.com/2023/05/03/spawning-lays-out-its-plans-for-

letting-creators-opt-out-of-generative-ai-training/.

See e.g., Ben Welsh, Who blocks OpenAI, Google AI and Common Crawl?

https://palewi.re/docs/news-homepages/openai-gptbot-robotstxt.html

dataset was publicly available, anyone could inspect and audit it, and build tools to make

examining the dataset accessible to the public.

Researchers have found that this is not

the case for all datasets used to train AI models,

and participants noted that as well.

“There is a lack of transparency from AI developers about training data sets, which

makes it very difficult to ascertain which works were actually used to train the models

and how,” said Kazi.

Ortiz said that lack of transparency makes it difficult for individuals even to know which

works they should request be removed, “Existing opt-out procedures often ask users to

list works used to train the model they own, but as we just mentioned, that training data

is secret, so it's an impossible task.”

Participants said that transparency around training data and what works were used to

build them was greatly needed. “Divulging your dataset should be compulsory,” said

Zapata.

Participants took issue with the fact that most opt-out frameworks were

about

future

models, and often did nothing to address past usage.

Throughout the event, many of the participants called for consent and compensation

because they said AI models were trained without them. Participants' consent was not

sought during the development of these AI models, and they did not have an

opportunity to negotiate for fair compensation for the works used.

Zapata said that mainstream AI developers’ opt-outs only apply to the future

developments. Even if someone chooses to follow a developer’s opt-out instructions,

Zapata said, “it’s already too late to get out of the most current model.”

This may be, as Zapata noted, due to technical limitations. Developing methods of

effective machine “un”-learning is currently a research topic of academic and corporate

interest.

Many of participants’ concerns surrounded past actions, which are not fully addressed

by giving a choice about future use. Said August, “The advent of AI doesn't change

fundamental ways that the world is supposed to work, and that world works in

permission first.”

Reisner, supra note 38.

See Rishi Bommasani et. al, The Foundation Model Transparency Index, arXiv (Oct. 19, 2023),

https://arxiv.org/abs/2310.12941.

See, e.g., NeurIPS 2023 Machine Unlearning Challenge https://unlearning-challenge.github.io/

What are participants doing to understand and address generative AI?

Throughout the event, participants described the approaches that they are using to

address and understand the ways that generative AI is reshaping their respective lines of

work. Participants said they have been conducting research to better understand

generative AI. They have engaged in labor strikes of near historic lengths as they

negotiate for protections against AI in their collective bargaining agreements. They have

backed legislative efforts on both the state and federal level. They have joined class-

action lawsuits, and they have attempted to engage with AI developers directly.

Participants described research efforts they were undertaking to

understand the emerging generative AI landscape. Clarke said that after he

experienced an influx of what he suspected were AI-generated submissions, he spoke

with fraud detection experts and AI researchers to better understand how to combat

false submissions to his magazine. He spent time evaluating many of the public and

commercial AI detection tools, because he was uncertain about their accuracy claims

and whether they would be suitable to use.

Kazi described the ways the Authors Guild has probed generative AI chatbots to better

understand what might have been included in their training data, since the developers

do not disclose what they contain.

Ziff said the Model Alliance is partnering with the Worker Institute at Cornell University

to develop a research study that would better help them understand the impact of

generative AI on fashion workers, with a particular focus on workers of color.

Union representatives who participated said they have been bargaining

over AI or are planning to when their current agreements are up. August, the

screenwriter and member of the Writers Guild of America West’s Negotiating

Committee, said the Guild’s recently ratified agreement could offer helpful guidance

when thinking about future public policy on AI, and laid out the new guardrails the

Guild has set in its agreement:

Our agreement defines that AI is not a writer and the material it generates is

not equivalent to human writing for purposes of our contract. That means that

AI cannot rewrite us, nor can it compete with a human writer for credit and the

associated financial benefit of that credit. Further, the studios now have to tell

us if they are providing us with material generated by AI and then it cannot

require us to use AI tools.

August emphasized that the Guild was able to win these protections because they are a

strong union that successfully carried off a nearly five-month strike—one of the longest

in the Guild’s history, according to Entertainment Weekly.

August reminded the

audience that most writers and artists don’t have unions to protect them. That fact was

echoed by participants who represented worker advocacy groups like the National

Association of Voice Actors and the Model Alliance. These kinds of groups can advocate

for and support their members, they but cannot collectively bargain agreements with

companies.

John K Painting of the American Federation of Musicians, another union that was

represented during the event, said that in addition to collective bargaining agreements,

legislative lobbying was an important mechanism for AFM to win protections for its

members. This approach was not just limited to the organized labor groups represented

at the event.

Throughout the event, participants mentioned specific pieces of legislation

they supported and hoped would help creative professionals and add

guardrails for how generative AI is built and used. On the state level, Ziff of the

Model Alliance said her organization supported the Fashion Workers Act in New York,

which would establish basic labor protections for models and content creators in the

state’s fashion industry. Ziff said the bill would help address the lack of transparency

around how models’ body scans are being used.

On the federal level, NAVA’s Friedlander said the organization recently endorsed the AI

Labeling Act of 2023,

which would require generative AI developers to include “a clear

and conspicuous disclosure identifying AI-generated content and AI chatbots,”

according to a press release from U.S. Senator Brian Schatz (D-Hawai’i), who

introduced the bill.

Jacobsen of the Artist Rights Alliance also referenced the Protect Working Musicians

Act of 2023,

which Jacobsen said would give small and independent musicians an

antitrust exemption to negotiate collectively with AI developers and streaming

platforms.

Additionally, at least one participant mentioned the idea of a federal right of publicity.

State-level laws have been passed in places like New York

and California,

but not

See Lester Fabian Braithwaite, Writers Guild officially ratifies new contract after months-long strike,

Entertainment Weekly (Oct. 9, 2023), https://ew.com/tv/writers-guild-ratifies-new-contract-after-

months-long-strike/.

N.Y. Legis. S. 2477A. Reg. Sess. 2023-2024 (2023).

https://www.nysenate.gov/legislation/bills/2023/S2477/amendment/A

AI Labeling Act of 2023, S. 2691, 118th Cong. (2023).

Press Release, U.S. Senator Brian Schatz, Schatz, Kennedy Introduce Bipartisan Legislation To

Provide More Transparency On AI-Generated Content (Oct. 24, 2023),

https://www.schatz.senate.gov/news/press-releases/schatz-kennedy-introduce-bipartisan-legislation-to-

provide-more-transparency-on-ai-generated-content.

Protect Working Musicians Act of 2023, H.R. 5576, 118th Cong. (2023).

NY CLS Civ R § 51 (2021)

Cal Civ Code § 3344

every state has its own version. Right of publicity laws generally protect a person’s

likeness from being misused for commercial purposes without their consent and could

potentially give creative professionals greater control over how things like their voices or

personal styles are being used. Since the event took place, a bipartisan group of senators

released a discussion draft of the No Fakes Act of 2023,

which would create such

federal protections specifically addressing misuse of generative AI.

A few participants said they were engaged in ongoing class-action lawsuits

that they hoped would help address some of the harms they said were

caused by generative AI developers. One such participant was Doug Preston, a

fiction author, who said he and his co-plaintiffs were seeking damages for the

unauthorized use of their copyrighted work. Preston said that moving forward, he and

his co-plaintiffs wanted AI developers get permission from authors, properly license

their books, and compensate them fairly for that use.

At least one participant mentioned proactively working with an AI

developer on a project. Friedlander, the NAVA president, said that he was working

on developing a synthetic version of his voice that would allow him to perform work he

would not otherwise be able to do, like read a newspaper cover-to-cover every morning.

Friedlander emphasized that not only was this being done with his consent, control, and

compensation, but more generally that generative AI tools should enhance and not

replace the work of creative professionals.

Press Release, U.S. Senator Chris Coons, Senators Coons, Blackburn, Klobuchar, Tillis announce draft

of bill to protect voice and likeness of actors, singers, performers, and individuals from AI-generated

replicas (Oct. 12, 2023), https://www.coons.senate.gov/imo/media/doc/no fakes act draft text.pdf.

Potential Areas of Further Inquiry

The event outlined the diverse palette of issues and experiences that creative

professionals face across the industry. Further research is required to help scholars, law

enforcement agencies, and other civil society organizations understand how generative

AI can impact communities and how those harms can be mitigated. Participants

highlighted a multitude of questions and open areas for further research.

Ortiz, for instance, raised several questions during the event about how opt-out

frameworks are being effectively communicated to people whose work has been

included in training sets, highlighting issues such as barriers of language and technical

expertise. Ortiz’s questions prompt other questions, such as: What other barriers do

existing opt-out frameworks present? How are the accessibility and usage rates of

options presented by these frameworks being measured?

Zapata noted another major barrier is a lack of disclosure around what is currently

included in training sets. Kazi and Clarke’s discussion of the Books3 dataset also raises

questions about what is needed beyond simply publishing a dataset. Books3 was always

public but not easily interpretable by the public prior to investigative reporting on the

topic. What efforts are being made to disclose the contents of training sets? What

practices are being developed to ensure that transparency is meaningful and furthers

understanding for a public beyond those with technical expertise in machine learning?

Another open question that Zapata touched on was how AI developers can effectively

comply with opt-out requests for existing models. What is the current state of machine

“unlearning” research? What is needed to effectively remove the contributions from

work no longer desired in a model, and how can that be verified? What measures of

efficacy are being developed as this research evolves?

Participants discussed many stages of model development and deployment where issues

may arise, starting from the collection and inclusion of work in training data without the

affirmative consent of its creators, to downstream misuse of these models and other

unintended effects. Friedlander, for instance, highlighted a new problem that voice

actors are now facing—proving the provenance of their own work to clients. Said

Friedlander, “an extra burden has now been placed on voice actors to prove that the

audio they’re delivering is not AI-generated.” For this example, and others, what is the

scope of the de facto burden placed on artists in the aggregate, to what extent is it

unremunerated, and is the allocation of such costs justifiable and efficient?

Beyond the individual effect that generative AI has on any given creator, how might the

practices of AI developers affect these markets over the long term? For example, would

the uncompensated and unauthorized use of creators’ content dissuade individuals and

firms from investing in high-quality content in the first instance?

Conclusion

The creative professionals at the roundtable discussion raised a number of concerns

regarding the impact that AI is having on them and on consumers. Although many of

the concerns raised at the roundtable lay beyond the scope of the Commission’s

jurisdiction, targeted enforcement under the FTC’s existing authority in AI-related

markets can help to foster fair competition and protect people in creative industries and

beyond from unfair or deceptive practices.

For decades, the FTC has used its authorities to address deceptive or unfair acts or

practices and unfair methods of competition as it relates to new and transformative

technologies. There is no “AI exemption” from the laws on the books. The FTC will

continue to vigorously use the full range of its authorities to protect Americans from

deceptive and unfair conduct and maintain open, fair, and competitive markets.

The FTC continues to listen and learn about the latest trends—and recognizes that the

topics covered above are by no means exhaustive. The FTC will continue to closely

monitor and scan the developments of these products, services, and tools. As the

generative AI industry continues to develop, the FTC will remain vigilant and ready to

use the full panoply of its law enforcement and policy tools to foster fair competition,

protect consumers, and help ensure that the public benefits from this transformative

technology.

Appendix

Appendix I: Participant Bios

John August is a screenwriter and member of the Negotiating Committee for Writers

Guild of America West (WGA), a labor union representing thousands of members who

write content for motion pictures, television, news and online media. He is a Negotiating

Committee Member at WGA West and was a former board member of WGA. John is a

screenwriter whose credits include Big Fish, Aladdin, Corpse Bride and Charlie’s Angels.

He also wrote the Broadway musical Big Fish, and the Arlo Finch novel trilogy. He lives

in Los Angeles.

Neil Clarke is the multi-award-winning editor of Clarkesworld Magazine, a science

fiction and fantasy magazine, and over a dozen anthologies, including the Best Science

Fiction of the Year series. An eleven-time finalist and the 2022 winner of the Hugo

Award for Best Editor Short Form, he is also the three-time winner of the Chesley Award

for Best Art Director. In 2019, Clarke received the SFWA Kate Wilhelm Solstice Award

for distinguished contributions to the science fiction and fantasy community.

Duncan Crabtree-Ireland is the National Executive Director and Chief Negotiator at

SAG-AFTRA, the most distinguished entertainment and media union in the world. In

this capacity, he oversees the world’s largest and most influential entertainment union,

comprised of more than 160,000 members worldwide who work in film, television,

broadcast news, commercials, music, video games and more. Crabtree-Ireland has

played a critical role in many of SAG-AFTRA’s signature achievements over the past two

decades, both in his current role and prior to that as longtime chief operating officer and

general counsel. Crabtree-Ireland is a strategic and creative lead negotiator, and has

personally led or overseen negotiations for SAG-AFTRA’s Netflix Agreement and its

Videogames, Commercials, Music and Network Television contracts, among others.

Crabtree-Ireland leads the union’s technology and innovation team. He is also the co-

host of the SAG-AFTRA Podcast and the SAG-AFTRA Podcast en Español.

Tim Friedlander is President and Founder at National Association of Voice Actors

(NAVA), a non-profit association created to advocate and promote the advancement of

the voice acting industry through action, education, inclusion and benefits. Tim is a Los

Angeles based voice actor, studio owner, advocate, and educator. He is the voice of the

PBR Summer Series (Professional Bull Riders) on CBS and the English voice of Ares in

the Netflix show Record of Ragnarok. As owner of the soundBOX: Group, a voice over

centric studio group with 3 locations, Tim has consistently provided safe space for voice

actors to train, learn, and work. Tim is co-founder and President of The National

Association of Voice Actors and in 2023, helped found United Voice Artists, an 18

nation, 36 member Federation of global voice over associations.

Jen Jacobsen is Executive Director of the Artist Rights Alliance, an artist-led

organization advocating for musicians to receive fair treatment and compensation in the

digital marketplace. She was previously the Executive Director of Kid Pan Alley (KPA),

an arts non-profit that teaches songwriting in schools. A 25-year music industry veteran,

Jen served as Vice President, Industry and Government Relations, for Sony Music

Entertainment and Vice President, Global Public Policy for Time Warner, Inc. Jen

received her B.A. in History from Yale University, and her J.D. from the University of

Virginia School of Law. She is a graduate of the Nashville-based Leadership Music

program and a lifelong musician.

Umair Kazi is the Director of Policy & Advocacy at the Authors Guild, the oldest and

largest professional writers’ organization in the United States, where he develops and

advances initiatives supporting writers' interests. His work encompasses copyright, AI,

labor, antitrust, and free expression issues. Umair previously served as the Guild’s Staff

Attorney, counseling members on professional legal matters. Umair is also a writer and

a translator from Urdu. He holds a J.D. from the University of Iowa College of Law and

an MFA in creative writing from Columbia University.

Bradley M. Kuhn is the Policy Fellow at Software Freedom Conservancy, a nonprofit

organization centered around ethical technology. Kuhn began work in the software

freedom movement as a volunteer in 1992 — as an early adopter of Linux-based

systems, and contributor to various Free and Open Source (FOSS) Software projects,

including Perl. Kuhn has industry experience as a computer systems administrator, a

software developer, and a high school Computer Science teacher. Kuhn has been lauded

with multiple awards for his lifelong work in enforcement and compliance of copyleft

FOSS licenses (such as the GPL). Kuhn holds an M.S. and summa-cum-laude B.S. in

Computer Science.

Karla Ortiz is a Puerto Rican, internationally recognized, award-winning artist. With

her exceptional design sense, realistic renders, and character-driven narratives, Karla

has contributed to many high profile film and TV productions, including Jurassic World,

World of Warcraft, Rogue One: A Star Wars Story, Thor: Ragnarok, Black Panther,

Avengers: Infinity War, The Eternals, Loki, HBO’s The Nevers and most notably, her

design of Doctor Strange for Marvel Studios’ Doctor Strange. Karla’s work is also

recognized in the fine art world, showcasing her figurative and mysterious art in notable

galleries, such as Spoke Art and Hashimoto Contemporary in San Francisco, Nucleus

Gallery, Thinkspace and Maxwell Alexander Gallery in LA, and Galerie Arludik in Paris.

She currently lives in San Francisco with her cat, Bady.

John K. Painting is the Director of the Electronic Media Services Division and

Assistant to the President for the American Federation of Musicians of the United States

and Canada (AFM), the largest union of musicians in the world. Painting has worked in

Electronic Media since he began his career at AFM Local 802 (New York City) in 2011.

In May 2019, Painting was hired by the Federation as EMSD Assistant Director,

working under former EMSD Director Pat Varriale. During that time, Painting was

responsible for the administration of the National Public Television Agreement as well

as a wide variety of special projects, most notably contracts pertaining to the rapidly

shifting world of Internet streaming since the start of the COVID-19 pandemic. Both at

the Local and Federation level, Painting has been regularly involved on the bargaining

team for EMSD contract negotiations over the last decade. He also sits on the AFM’s

Education Committee and has worked to create new and updated educational content to

better explain Electronic Media Services agreements to both musicians and employers.

Douglas Preston is the author of 38 books, of which 32 have been New York Times

bestsellers, some reaching the #1 position. Two of his novels, co-written with Lincoln

Child, were chosen in a National Public Radio poll of 17,000 readers as being among the

one hundred greatest thrillers ever written. His recent nonfiction book, The Monster of

Florence, is being made into a television series. Preston worked as an editor at the

American Museum of Natural History and taught nonfiction writing at Princeton

University. He served as president of the Authors Guild from 2019 to 2023.

Steven Zapata is an American artist with 12 years of experience in commercial design,

illustration, and art education. He has contributed to video games and movies such as

The Elder Scrolls Online and Disney’s Noelle. Steven has also designed theme parks and

experiences both domestically and abroad including the USA Pavilion for World Expo

2020, Warner Bros. World in Abu Dhabi, and the Harry Potter Studio Tour in London.

He has also taught design and storytelling at Art Center College of Design in Los

Angeles, and has a YouTube channel where he shares his love of drawing with students

from across the globe.

Sara Ziff is the founder and executive director of the Model Alliance, a nonprofit

research, policy and advocacy organization for people who work in the fashion industry.

At the Model Alliance, she established the first industry-specific support line for fashion

workers and played a leading role assisting survivors during the #MeToo movement.

Sara has successfully championed legislation to advance workers’ rights and protections,

and she is currently working to pass the Fashion Workers Act in New York State. She

received her B.A. from Columbia University and her M.P.A. from Harvard Kennedy

School of Government.

Appendix II: Roundtable Quote Book

This quote book is a compilation of direct quotes from the participants in the FTC’s

October 2023 Creative Economy & Generative AI roundtable aggregated into some

common themes. This summary aims to be a resource that shares the specific

perspectives and lived experiences of the participants of the event on the impact of

generative AI on their industries and is by no means a complete representation of every

perspective on this topic.

• “AI and its algorithms must be here to serve us, not the other way around.” -

Duncan Crabtree-Ireland

• “We publish stories not only from established writers, but also new voices from

all over the world. These people represent the future of my field, and it is

essential to maintain the avenues through which they can be discovered.” - Neil

Clarke

• “The problem is not quality at the moment, but rather the sheer speed and

volume at which these works can be produced.” - Neil Clarke

• “The best way to describe them is to call them spam, but it's spam at a level we've

never seen before.” - Neil Clarke

• “Pulitzer Prize winning author and Authors Guild councilmember Min Jin Lee

has likened the experience of having her work used by AI to identity theft. ‘AI

company stole my work, time, and creativity,’ she recently wrote on X/Twitter.

"They stole my stories. They stole a part of me." It's not just well-known and

bestselling authors that are suffering.” - Umair Kazi

• “ChatGPT would be lame and useless without our books. Just imagine what it

would be like if it was only trained on text scraped from web blogs, opinion

screeds, cat stories, pornography and the like.” - Douglas Preston

• “But supporting creativity by stealing from creators is like claiming you're

supporting the candy store by shoplifting.” - Douglas Preston

• “This is our life's work. We pour our hearts and our souls into our books. They

aren't just products, they're a part of us.” - Douglas Preston

• “The founders of our country wrote copyright protection into the very first clause

of the Constitution—it was that important to them. They believed that their

scrappy little country one day would become the creative engine of the world, and

that's what we've become, and we can't allow AI developers to ignore copyright

protection and injure the entire literary community of our country in their mad

rush to succeed. They can succeed and they can also partner with America's

authors in a mutually beneficial relationship.” - Douglas Preston

• “Not all voice actors are celebrities or well-known voices. Most are blue collar,

working class voice actors who are working 40 plus hours a week. Over 60% of

the voice actors are located outside of LA and New York. We're not anti-tech or

anti-AI, as many have said before.” - Tim Friedlander

• “Musical expression transcends culture and time. It will always be a part of our

society, but that doesn't guarantee it as a viable career. The end game must be the

protection of the profession.” - John Painting

• “This is what you get when the fox designs the chicken coop. The wrong people

are making the decisions when you have something like this happening. And

there's no standard, there's no consequence for them ignoring these things at the

moment either. So we really need to have them seeking consent, explicit consent,

to include these works.” - Neil Clarke

• “Two months ago or a few months ago, he received a phone call from somebody

claiming to have kidnapped his daughter, was demanding ransom, played a

recording or a voice of his daughter screaming, ‘Papa, help me. Help me. Papa,

save me.’ He said, ‘I knew deep in my bones, that was my daughter's voice.’ His

daughter was safe. It was a scam, it was a fake. But he said ‘The voice that I heard

on that line was my daughter's voice. It was the same voice I heard when I

dropped her off at school. And it was the same voice that I heard when I picked

her up when I found out she was safe.’ The danger to consumers is real. It is

tangible. And it is here now.” - Tim Friedlander

“We’re not anti technology.”

• “But to be clear, we at SAG-AFTRA are not opposed to new technologies and

we're not opposed to the existence or even the use of AI. AI in particular is an

essential tool that is going to be widely used in the decades to come and it can

have a variety of creative and innovative uses in the entertainment industry, and

the implementation of AI we are looking to achieve would result in the

technology augmenting human creativity rather than replacing it.”

- Duncan Crabtree-Ireland

• “The ability to record audio on our computer at home was game changing. I'm a

child of the 80s. I still remember the time that I died of dysentery on the Oregon

Trail for the first time. We are not anti-tech or anti-AI. We are pro voice actor.” -

Tim Friedlander

• “As many others have said, we're not anti-technology. We're anti-exploitation.

And we believe this round table is an important step toward ensuring our creative

community is afforded the protections we need and deserve.” - Sara Ziff

• “Musicians are not against AI. We don't want AI to be banned. In fact, musicians

have been using AI-driven tools for years to auto tune vocals, generate beats,

assist with studio production, and in other parts of the creative process as well.” -

Jen Jacobson

“Our members are keenly aware that their works are being used by AI systems.”

• “Shouldn't the individuals whose intellectual property was used to train the AI

algorithm be at least equally protected? An actor's brand is their voice, as is their

likeness and their unique persona, and no company should be able to appropriate

that and use it however they wish without permission.” - Duncan Crabtree-

Ireland

• “Large language models like the one that underpins ChatGPT have scraped

massive volumes of data, including our words and our unique perspectives. This

is theft, not fair use.” - John August

• “Our works— protected by copyright and our own contractual rights—are being

used entirely without our authorization, without any attribution or

compensation. Right now, you could ask ChatGPT to write something “in the

style of” a particular writer, and they would try to do that—appropriate the

unique voice of a writer without that writer's consent.” - John August

• “Meanwhile, Big Tech has launched a campaign to manufacture consent about

these systems. Big Tech claims that the rules, licensing, and legislation that is

applied to creative works since the 1800s in the United States are suddenly moot

simply because machine learning is, in their view, too important to be bogged

down by the licensing choices of human creators of works.” - Bradley Kuhn

• “These companies describe their stolen training material as publicly available

data or openly licensed content to disguise their extensive reliance of

unauthorized use of copyrighted and private data. These companies even break

their own promises not to sell users information.” - Karla Ortiz

• “To advertise our work, most of us put our art online, on social media and our

personal websites. This leaves it exposed to unethical scraping practices, which

can result in AI models being created that can mimic the signature styles of

particular artists.” - Steven Zapata

• “Our members are keenly aware that their works are being used by AI systems.

We get reports. Our members try out different prompts and AI systems and chat

box reveal details about their works. And of course, several of our members have

also reported finding their books in Books3, which is a data set of containing

200,000 books that's downloaded from a pirate source, which was used to train

Meta's LLaMA, Bloomberg's GPT, and others.” - Umair Kazi

• “Today's reckless, careless rush to launch new generative AI products, we are

seeing what is euphemistically referred to as AI training or learning, but which is

in fact illegal copying of artistic works on a massive scale without consent or

compensation and often without the artist even knowing.” - Jen Jacobsen

Platforms using user-generated content for training

• “We knew these models were trained without permission on the copyrighted

works of others, and it didn't sit right with us.” - Neil Clarke

• “Zoom is among the many Big Tech companies that have sought to cajole users

into consent for use of their user data as training input for machine learning

systems.” - Bradley Kuhn

• “For example, data laundering, where a company outsources its data collection to

a third party under the pretext of research to then immediately use that for

commercial purposes. An example of this is LAION. Specifically, it's more

popular LAOIN-5B, a dataset that contains 5.8 billion text and image pairs, which

again includes the entirety of my work and the work of almost everyone I know.” -

Karla Ortiz

• “The massive datasets that text-to-image models are trained off of contain the

copyrighted works of artists, whether it be work done for ourselves or work done

for companies in a work-for-hire capacity.” - Steven Zapata

• “Unfortunately, in today's reckless, careless rush to launch new generative AI

products, we are seeing what is euphemistically referred to as AI training or

learning, but which is in fact illegal copying of artistic works on a massive scale

without consent or compensation and often without the artist even knowing.” -

Jen Jacobson

Contracts

• “Contracts we signed years ago are now being used to justify the inclusion of our

audio in synthetic voice models. And every time that happens without the

consent, control, and compensation of the voice actor involved, the value of that

voice actor's product, their voice and the sound of their voice, is diluted and

unfairly diminished. We consider this to be a form of theft. We agree to license

our voice to clients. The client is buying the exclusive rights to that recording,

that performance, but also the sound of our voice. We could potentially now be in

conflict with a clone of our own voice, which causes harm to both the voice actor

and the company. Pepsi and Coke can't have the same voice providing their

commercials. Chevy and Ford can't have the same voice. Currently the only

protections that we have from voice actors having their audio uploaded to these

sites are the terms of service that people agree to.” - Tim Friedlander

• “When a model signs a management agreement, she typically hands over a power

of attorney, and thus rarely, if ever, sees her own contracts with the brands. As a

result, models have little transparency into their own business dealings.” - Sara

Ziff

• “Also, the nature of the typical work-for-hire contract has put us in a bad bind.

Almost every commercial artist today works under all-encompassing work-for-

hire contracts that surrender all rights to a company.” - Steven Zapata

• “Models have very little insight into how their work or likeness is being used in

general, let alone in the context of generative AI. Normally they don't see their

contracts with the brands and often don't know how their image will be used,

whether how much they'll be paid. So generative AI introduces the potential for

further exploitation in an already exploitative work environment.” - Sara Ziff

Body “Scans”

• “Increasingly, companies are asking models to undergo scans that generate a 3D

model of their body or face. In our poll, nearly 18% of models who responded had

been asked to undergo a scan by a brand or a management company.” - Sara Ziff

“Losing work”

• “Once they start employing bots to submit these words, what we currently receive

in a month may arrive in hours. Quantity and the lack of reliable detection tools

will eventually break our ability to do our job.” - Neil Clarke

• “In one instance earlier this year, AI-generated books started dominating

Amazon's bestseller list in the young adult romance category. We have seen AI-

generated books pop up for sale on closely related or very similar topics as those

authors have listed for pre-orders, a growth in the number of unauthorized

summaries of books, and fake books using names of renowned authors.” - Umair

Kazi

• “Freelance journalists and professional writers of web and marketing content are

reporting losing work at an alarming rate. An Authors Guild member who writes

marketing and web content reported losing 75% of their work as a result of

clients switching to AI.” - Umair Kazi

• “A content writer featured in a piece about ChatGPT's impact in the Washington

Post stated that he had lost a half of his annual income. There's a widespread and

real concern among writers that AI will decimate the profession. In our recent

survey, 69% of authors said that generative AI threatens their careers, and an

overwhelming 90% said that authors should be compensated if their works are

used in training. These concerns are born out of the experience of enduring

longterm precarity.” - Umair Kazi

• “A voice actor in New York worked for a company for three years, and year four,

they were let go because they were told the company had enough of their audio,

and they were going to now create a synthetic version of their voice.” - Tim

Friedlander

• “And most recently, a voice actor in Washington state lost an audiobook job when

they decided to take the job in-house. Around that same time, the audiobook

company made a press announcement that they were now using ElevenLabs for

all of their audiobook productions, effectively replacing all of those human

narrators with synthetic voices.” - Tim Friedlander

• “Fashion workers are worried about the threat of these AI models replacing jobs

– not only for models, but also photographers, stylists, and hair and makeup

artists among others.” - Sara Ziff, Model Alliance

• “Perhaps most harmfully, these exploitative products compete directly with

artists and are already replacing us. That may be the most disturbing harms of

generative AI. Not vast murdering sci-fi nonsense, but one built on works taken

without credit, consent, compensation and transparency, and marketed and used

as a replacement for the creators of those works at a fraction of the cost. The

whole process is rotten.” - Karla Ortiz

• “Compared to previous technological shifts, the job losses here could be steep:

recording and scoring of all types can be wholesale replaced if composers and

songwriters can replace the output of live performing instrumentalists with an

AI-generated facsimile. Producers may replace the composers and songwriters

themselves. Live touring across all genres would be impacted if bands and artists

could tour with AI- generated backing musicians, holographic orchestras, AI

backing tracks.” - John Painting

Impacts on opportunities

• “Members in our community have expressed particular concern about companies

using AI-generated models as part of their diversity and inclusion initiatives. For

example, Shudu, a digital model who was created through AI in 2017 by the

world's first all-digital modeling agency, has appeared as a face of high-end

brands such as BMW and Louis Vuitton. Critics have called this a form of "digital

blackface” since Shudu is a Black woman, and the creator who profits off her

image is a White man.” - Sara Ziff

• “And earlier this year, Levi's announced that they are creating AI-generated

models to increase the number and diversity of their models. In an industry that

has historically been discriminatory, creating digital representations of models of

various ages, ethnicities, and body types rather than hiring and paying a diversity

of real models is concerning.” - Sara Ziff

• “When used ethically and in a manner that recognizes intellectual property

rights, AI can help people in their careers and can further opportunities. It can

create access to employment for people with disabilities and those who would

otherwise be prevented from pursuing work in the entertainment industry.” -

Duncan Crabtree-Ireland

“Time spent”

• “Unfortunately, we still have to review each of these suspicious submissions

manually, but for now, we've been able to stay open. Between the time spent

reviewing submissions and maintaining the software to control them, my

workload has easily doubled. It's been exhausting, and this is only the early days.”

- Neil Clarke

• “Additionally, an extra burden has now been placed on the voice actors to prove

that the audio they're delivering is not AI-generated. Many voice actors have

reached out to NAVA asking for some way to prove that they delivered human-

generated audio.” - Tim Friedlander

• “Does that mean we have to opt out on each and every one of them? That's a full-

time job. What about if those models update? What about if they don't publicize

and they use third parties? What if those models in the opt-out forms are not an

artist's native language? What about artists who never spend time online or don't

even know this is happening?” - Karla Ortiz

Loss of money and/or compensation

• “My own name, Karla Ortiz, has also been used thousands of times. I never gave

consent. I never got credit. I never got compensation.” - Karla Ortiz

• “A machine-generated creation which utilizes a performer's voice, face, or in our

case instrumental sound without the actual participation of that performer in the

creation is tantamount to replacing the performer entirely. It removes the human

from the creative process and the performing process. When that happens, the

work is gone, the wages are gone, the secondary residuals are gone.” - John

Painting

• “For musicians, like all the other creators represented here today, the unethical

use of AI poses an existential threat to our livelihood.” - Jen Jacobson

Accuracy and Fairness

• “To make matters worse, I observed that foreign authors were far more likely to

be incorrectly flagged as AI and an experienced eye remains the only reliable

method of detection.” - Neil Clarke

• “And earlier this year, Levi's announced that they are creating AI-generated

models to increase the number and diversity of their models. In an industry that

has historically been discriminatory, creating digital representations of models of

various ages, ethnicities, and body types rather than hiring and paying a diversity

of real models is concerning.” - Sara Ziff, Model Alliance

• “With that in mind, there's a real risk that AI may be used to deceive investors

and consumers into believing that a company engages in fair and equitable hiring

practices and is diverse and inclusive, when they are not.” - Sara Ziff

Diminished value of work

• “It is using stolen goods to undercut the price of a seller and create market

confusion, and it's not a hypothetical.” - John August

• “While it is likely that there are many positive uses for this technology, I don't

believe that authors, artists, translators, narrators, and others should be required

to sacrifice their work to improve or build these models. Progress can still be

made responsibly without stepping on creatives.” - Neil Clarke

• “There's a serious risk of market dilution from machine generated books and

other works that can be cheaply mass-produced, and which will inevitably lower

the economic and artistic value of human created works.” - Umair Kazi

• “These pieces have subject matters and qualities that we have no control over.

And the negative market implications of a potential client encountering a freely

downloadable AI copycat of us when searching our names online could be

devastating to individual careers and our industry as a whole.” - Steven Zapata

Loss of ability to compete

• “Without guardrails, AI poses a profound threat to writers and the integrity of

our creative work. AI and its use raises major questions of intellectual property,

transparency of competition, all of which require careful oversight to protect the

creative economy.” - John August

• “It's difficult enough to compete against other humans. Now mom and pop voice

actors across the country have to compete against digital clones of themselves

and against multimillion and billion dollar tech companies. It's incredibly easy to

use AI to capture the voice of an actor from content available on the internet, and

to use that sample to create whole works for sale or non-commercial

distribution.” - Tim Friedlander

• “Musicians’ work is being stolen from them and then used to create AI-generated

tracks that directly compete with them. For example, we might see dominant

streaming platforms packing playlists with AI music that they obtain free of

charge or at a massive discount, which then lowers their own royalty obligations

and diminishes artists wages.” - Jen Jacobsen

• “This conduct may violate “right of publicity” laws in several states, but it needs

to be recognized universally as a misappropriation that causes real harm not only

to the artists, but to the entire market by confusing consumers and creating

unfair competition.” - Jen Jacobsen

• “The increasing scale of machine-generated music dilutes the market and makes

it more difficult for consumers to find the artists they want to hear. It makes it

harder for artists to connect with their fans, and it devalues human creativity.” -

Jen Jacobsen

• “Without transparency, we can't even know the extent of how much of these

companies have taken. They took our work and data to train for-profit

technologies that then directly compete against us in our own markets using

generative media that is meant to mimic us.” - Karla Ortiz

• “Meanwhile, Big Tech has launched a campaign to manufacture consent about

these systems. Big Tech claims that the rules, licensing, and legislation that is

applied to creative works since the 1800s in the United States are suddenly moot

simply because machine learning is, in their view, too important to be bogged

down by the licensing choices of human creators of works.” - Bradley Kuhn

Reputational damage

• “In February, Twitter doxxing: Voice actors such as Michael Schwalbe and Abbey

Veffer had their voices cloned and had Twitter accounts made that gave out their

personal addresses in their own voice and said racist and homophobic things.” -

Tim Friedlander

• “It's hard to imagine anything more personal to an artist or to anyone than being

depicted as doing or saying things that they would never do or say. It's not only

confusing to fans, but humiliating to the artists themselves and undermines their

public image.” - Jen Jacobson

• “Right now, authors are finding AI generated knock-offs of their work published

on Amazon. They're having to fight to get those fakes taken down, and protect

their brands.” - John August

Deepfakes, Impersonations, and/or Fan confusion

• “It is using stolen goods to undercut the price of a seller and create market

confusion, and it's not a hypothetical. Right now, authors are finding AI

generated knock-offs of their work published on Amazon. They're having to fight

to get those fakes taken down, and protect their brands.” - John August

• “From electronics to organic eggs, consumers expect to be told the origin of a

product and its authenticity. Consumers make choices based on that information.

The same will likely be true with AI.” - John August

• “We traced the source of much of this activity to YouTube, TikTok, and blog

content hosted by side hustle con artists that would waive a stack of cash and

make false claims about earning riches from ChatGPT. Their intent was to earn

advertising revenue or sell classes to those that tried and failed. They knew these

tactics would not get people published. We happened to be on a list of a hundred

or so magazines that were used as part of these schemes. Over the months that

followed, copycats would refine their approach.” - Neil Clarke

• “We are already seeing that AI is being used to generate low quality eBooks,

impersonating authors, and displacing human authored books in the

marketplaces like Amazon.” - Umair Kazi

• “Generative AI is being used to create unauthorized derivative works such as a

developer using ChatGPT to write the concluding books in George R. R. Martin's,

“A Song of Ice and Fire” Series and chatbots like the Dan Brown Chatbot.” -

Umair Kazi

• “And at the same time, many authors were discovering that ChatGPT-3 knew

everything about their books as well. And some realized it was even being used to

create works that imitated their own. My friend George R.R. Martin, who was

already mentioned, was very disturbed when AI was used to write the last book in

his Game of Thrones series using his characters, his plot lines, his settings—even

his voice.” - Douglas Preston

• “Thousands of synthetic voices are currently online on websites from video games

all over the internet. Fan dubs are being created and turned into adult content,

and then sharing those fan dubs on YouTube using the voices of those voice

actors.” - Tim Friedlander

• “Those who had been scanned described not being given information about how

their scans would be used, unknowingly handing away rights to their image, and

not being fairly compensated. For people whose livelihoods are their image, this

is particularly troubling in light of the rise in deepfake technology, specifically

deepfake pornography.” - Sara Ziff

• “The companies offering these models often encourage users to request work in

the styles of particular artists by name, and many of these generations end up

online with our names attached to these pieces that we didn't make.” - Steven

Zapata

• “And perhaps even more disturbingly, AI models are now using artists’ faces,

voices, and performances without permission to make digital impersonations

that not only create consumer confusion, but also cause serious harm to both fans

and artists. These deep fakes have depicted a band canceling a concert that wasn't

actually canceled. They've shown artists selling products that the artists never

endorsed. We've seen false depictions of musicians badmouthing their own fans.

This isn't a hypothetical harm. This type of consumer deception and fraud are

happening right now.” - Jen Jacobson

• “The ability to create a synthetic voice from anyone who has recorded audio is

easy and simple and dangerous. Currently now it only takes three seconds of

source audio to create a realistic voice clone. And this synthetic content can be

used to deceive consumers into believing that a trusted voice is communicating

with them. This can lead to relying on false and misleading information and

potentially even implicate the human whose voice has been used to harm people.”

- Tim Friedlander

Consent, Permission, and/or Opt-out vs. opt-in

• “If consumers take anything away from my comments today, I hope they

remember to carefully read the terms and conditions of all software platforms

they use, as they may have already agreed for their own creative works to become

part of the company's machine learning data sets. I admit it may take you a week

to read all of those terms, but it's sadly the only way you'll know what rights

you've inadvertently given away to Big Tech.” - Bradley Kuhn

• “AI developers have copied millions of copyrighted works without permission.

These works are not only copied many times in the course of compiling training

data sets and ingestion, but are embedded in the very fabric of the language

models.” - Umair Kazi

• “And at one point, I asked it to write a poem in heroic couplets about one of my

characters. And I was floored at the level of detail it knew when it generated this

poem, and that's when I realized it must've ingested many of my books.” -

Douglas Preston

• “OpenAI illegally ingested our books to create a product that is currently valued

at tens of billions of dollars, and they did this without our consent or

compensation. And as Umair mentioned, the average full-time author in America

makes only $20,000 a year. This is a classic case of Robin Hood in reverse,

stealing from the poor to give to the already obscenely rich.” - Douglas Preston

• “In their race to be first, AI developers are swallowing everything they can get

their hands on without regard to copyright ownership, intellectual property

rights, or moral rights. And they're doing this without the slightest consideration

given to supporting the livelihood of America's creative class.” - Douglas Preston

• “I personally am working on a synthetic voice that I have consent, compensation,

and control for. There are some things that humans can't physically do, such as

narrate the New York Times cover-to-cover every morning, or provide a realistic

voice for someone who is nonverbal. But this tech should enhance and not

replace voice actors.” - Tim Friedlander

• “So we need regulation, intervention, and oversight. We as creators should have

complete control over how our work is used, but we need help. Some of the

potential actions and remedies that we hope to see include, first and foremost,

ensuring that all commercial AI models utilize only public domain content or

legally licensed datasets acquired in an opt-in capacity. Opt-out is completely

insufficient here. This could mean current companies shifting to the public

domain and possibly destroying their current models in the process so that opt-in

becomes the standard.” - Steven Zapata

• “Mainstream models like DALL-E 3 don't reveal their training data and don't let

you search it, but they do offer an inefficient one by one opt-out system that you

can use if you think maybe your art is in there.” - Steven Zapata

• “What's new today, though, are the expansive AI models that ingest massive

amounts of musical works and mimic artists voices without obtaining creators

consent or compensating them.” - Jen Jacobson

• “The final point I want to make is about the importance of choice. Many AI

companies who have illegally vacuumed up hundreds of thousands of musical

works and recordings now say that artists can simply contact the company and

“opt out.” This is essentially proposing a new form of uncompensated labor that

musicians and composers have to perform if they want to avoid exploitation.” -

Jen Jacobson

• “Opt-out is completely insufficient here. This could mean current companies

shifting to the public domain and possibly destroying their current models in the

process so that opt-in becomes the standard.” - Steven Zapata

• “Basically, tech companies must respect artists ownership rights. Those seeking

to profit from others works should have the burden of obtaining permission.

Explicit opt-in is the only way forward. It's really how we ensure generative AI

models exclude unauthorized works from the beginning.” - Karla Ortiz

Machine unlearning

• “But because these AI systems can't unlearn, this will only remove the images

from future training datasets used by this one company and it's already too late to

get out of the most current model.” - Steven Zapata

• “Yeah, so opt-out is an ineffective and inappropriate standard for commercial use

of copyrighted works including a generative AI. Once a model is trained on data,

it cannot be deleted unless the whole model is retrained from scratch. By the time

a model is made public, it's already too late to opt out.” - Karla Ortiz

Transparency and disclosure

• “We also need transparency on datasets, and divulging your dataset should be

compulsory.” - Steven Zapata

• “Existing opt-out procedures often ask users to list works used to train the model

they own, but as we just mentioned, that training data is secret, so it's an

impossible task. And four, there are hundreds of AI models already in the market

and more.” - Karla Ortiz

• “But as far as how we know our books are being used in AI training, we have

absolutely no idea. It's a black hole. OpenAI is training ChatGPT 5 right now, is

building it. They refuse to answer any questions from the Author's Guild about

what data sets they're using, where they're getting their books, and how they're

being used. There's no transparency at all. It's an absolute black hole.” - Douglas

Preston

Changing terms of service

• “Companies can change their terms of service, and currently it's easy as checking

a box that says, "I have the right to upload this audio." We've seen damages this

year very clearly.” - Tim Friedlander

Policy and legislative efforts

• “The fight for protection over our craft and livelihoods doesn't stop at the

bargaining table. While we have been able to achieve groundbreaking protection

for writers, we need public policy solutions, too.” - John August, WGA West

• “The Guild’s new agreement offers helpful guidance in thinking about future

public policy on AI. Our agreement defines that AI is not a writer and the

material it generates is not equivalent to human writing for purposes of our

contract. That means that AI cannot rewrite us, nor can it compete with a human

writer for credit and the associated financial benefit of that credit.” - John August

• “And as such, NAVA recently endorsed the [US Senators] Schatz and Kennedy’s

AI Labeling Act of 2023, which is Senate Bill 2691 that we are fully endorsing.” -

Tim Friedlander

• “To address these concerns, we first aim to pass the Fashion Workers Act, our

signature bill, which would establish basic labor protections for models and

content creators working in New York's fashion industry. This would help

address the lack of transparency that leaves models in the dark about how their

digital image is being used, and establish a necessary foundation for regulation

around generative AI in the fashion industry.” - Sara Ziff

• “Second is artists need to have enough power to negotiate fair license terms with

these gigantic AI developers. There's actually a bill in Congress sponsored by

representative Deborah Ross that would give small and independent musicians

an antitrust exemption so they can come together and negotiate collectively, both

with AI developers and streaming platforms, which is something they do not have

the leverage to do currently.” - Jen Jacobson

Lawsuits

• “I'm also the plaintiff in a class action lawsuit against OpenAI, along with 15

other authors and the Guild itself. And we're asking for damages for unauthorized

use of our copyrighted work and training and building ChatGPT.” - Douglas

Preston

• “And that's why we joined together, the 17 of us authors, in a class action lawsuit

on behalf of all professional novelists against OpenAI. There's nothing

complicated about this lawsuit.” - Douglas Preston

• “As a side note, due to all of this, I am also a plaintiff in a class action against

generative AI image companies as well.” - Karla Ortiz

Self-initiated research and investigations

• “I evaluated many of the public and commercial detection tools and found their

claims significantly overstated. The number of false positives and false negatives

made them unusable.” - Neil Clarke

• “So we at the Authors Guild investigated, and here are some of the facts we

uncovered. ChatGPT3 used more than 150,000 copyrighted books to feed into its

AI system, which led us to the next question. "Where did OpenAI get our books?"

They're not just sitting out there on the web unprotected. Well, the Authors Guild

found that OpenAI got many, if not most, of these books from pirate websites

such as LibGen run out of Russia.” - Douglas Preston

• “We're also developing a research study in partnership with the Worker Institute

at Cornell University to better understand the impact of generative AI on fashion

workers, particularly workers of color, and develop policy recommendations.” -

Sara Ziff

Collective bargaining

• “We won these protections because we're a strong union that successfully carried

off a nearly five month strike. But we need to remember that most writers and

most artists in this country don't have unions to protect them.” - John August

• “We stand in solidarity with our fellow creative industry artists who are in

unions. But unlike them, 80% of the voiceover industry is non-union, meaning we

lack the protections and contract that organize workers enjoy.” - Tim Friedlander

• “Models are typically hired as independent contractors through management

companies which, unlike talent agencies, are held to very few legal standards. So

when we talk about how generative AI is impacting workers, we need to consider

the context of an industry that is truly like the Wild West– where workers have

few protections at baseline and also cannot collectively bargain here in the US.” -

Sara Ziff

• “At The Model Alliance, we believe now is a critical time for solidarity between

workers across creative fields who contribute heavily to our culture and economy.

Unfortunately, it's not enough to win protections through collective bargaining

agreements. There are many workers, including members of our community, who

cannot engage in collective bargaining, and so we have to ensure that they are

included.” - Sara Ziff

• “And unfortunately, visual artists don't have strong union representation to push

back on this. As it stands, you can work hard for a company like Disney for 25

years and they can take all your work, train a model on it, and then have a free

and tireless replica of you to use in perpetuity.” - Steven Zapata

• “The solutions sought have been traditionally approached in two ways: collective

bargaining with industry and legislative lobbying. Both paths tend to seek

secondary income to those performers whose work has been diminished by

advancing technology.” - John Painting

Regulation

• “We need oversight.” - Steven Zapata

• “Regulation of this industry is needed sooner than later, and each moment they

are allowed to continue their current practices only causes more harm. Their

actions to date demonstrate that they cannot be trusted to do it themselves.” -

Neil Clarke

• “Generative AI poses a serious threat to the writing profession, and we believe

that guardrails around its development and use are urgently needed.” - Umair

Kazi

• “No copyright for AI-generated outputs. We oppose efforts to deem AI-generated

content protectable under copyright law or through creation of even a limited

suite generous right”

- Umair Kazi

• “We're here today because a future with unregulated AI will hurt concept artists

and all other sorts of artists across many fields. We need regulation, intervention,

and oversight.” - Steven Zapata

• “If the FTC is not able to do this through rulemaking, we would request the FTC's

support for federal legislation to establish that right. Transparency of ingested

content on which these foundational models are trained in order to know if our

voice is present. Protections prior to the generation of any AI-created content

that might include voices of professionals that have not provided consent and are

not being compensated. Protection of our voices' biometric data for privacy and

commercial purposes. An independent third party to verify that audio files are

ethically sourced. And finally, clear labeling of any AI-generated content to

ensure the consumers are fairly informed.” - Tim Friedlander

• “Regulatory agencies should act now to protect artists, consumers, and other

Americans from this unconscionable exploitation. Regulatory agencies should

demand full transparency from generative AI companies and opt-in only

practices.” - Karla Ortiz

• “Lastly, regulatory agencies should strongly consider seeking algorithmic

disgorgement on products built on data acquired without consent, credit, or

compensation, regardless whether that company is transparent or not. Urgent

measures like these will be needed to avoid, in my opinion, the diminishing or

outright destruction of most, if not all creative professional livelihoods and the

protections of all of our rights.” - Karla Ortiz

• “In my opinion, there's no reason that big tech shouldn't be regulated to make

these systems transparent, completely end to end.” - Bradley Kuhn

Consent

• “Consent and compensation. Require all generative AI companies to seek

permission for the use of creative works and to fairly compensate creators.” -

Umair Kazi

• “And going forward, we're asking that OpenAI and other AI developers get

permission from authors, properly license our books, and compensate us fairly

for that use.” - Douglas Preston

• “We also think authors and artists should have the right to say that they don't

want their identities, works, voice or style used in outputs.” - Umair Kazi

• “That's why we think it's critical that we require artists have affirmative consent

before the work can be used to train generative AI models and that they have to

be compensated fairly when they do so. The same should be true for all artists,

including artists like us who do work for hire and don't hold the copyright on our

work.” - John August

“A system that is opt-in. Not opt-out.”

• “And this system needs to be opt-in and not opt-out. As Jen just said, there are so

many companies out there developing and training AI models, to be forced to

continually track all of them down to opt out is an enormous administrative

burden on individual artists. It's not practical. It has to be opt-in rather than opt-

out.” - John August

Credit and Transparency

• “Credit and transparency. Create obligations for all AI companies to disclose

what data sets and works they use to train the systems.” - Umair Kazi

• “Labeling AI-generated content.” - Umair Kazi

• “As far as what we want, we want AI companies to be required to fully disclose

the complete lists of copyrighted works, books in particular is the medium that

we deal with most often, that are in the training data sets or provide specific links

to where the data sets were obtained from. And anyone compiling a training

dataset should be similarly obligated to disclose the sources. And in the case of

copyrighted works, a complete list of works that have been included in the

dataset.” - Umair Kazi

• “And we believe that there should be a requirement to conspicuously label fully or

substantially AI generated words in online marketplaces.” - Umair Kazi

• “In my view, the public should have access to the input set, have access to the

source code of the software that does the training and generation, and most

importantly, access to the source code that does these forms of backend

generation exclusion, the latter of which I think would expose the duplicity of big

tech's policies here.” - Bradley Kuhn

• “At a minimum the consumers should know when AI is used to generate voices

and receive a warning that the information they're going to receive may not be

accurate.” - Tim Friedlander

• “We need compulsory transparency and tools to verify compliance.” - Steven

Zapata

Compensation

• “Permission and payment for use in outputs. Require all AI companies to seek

permission and pay compensation when creative works are used in outputs or

when names or identities or titles of works are used in prompts.” - Umair Kazi

• “We should also have AI companies pay a fine for their past practices and pay all

affected artists a fee per generation. This is to compensate artists for utilizing

their works and names without permission, should be retroactive for as long as

the company has been for-profit. We must close research to-commercial

loopholes, interpreted or actual, that allow for-profit companies to monetize the

results of non-commercial research.” - Steven Zapata

Licensing

• “And the third thing I would say is there is not a one-size-fits-all licensing system

that will work for all creators or even for all musicians. Assuming there is a level

playing field for negotiating, we think the best way for musicians to license their

work is in the free market, which may look different for every use, every artist

and every company.” - Jen Jacobson

• “Without a doubt, licensing will be essential in the future, but we must

accomplish that through an opt-in system, otherwise there would be no real

negotiating leverage for creators. And the focus of licensing, I think should go

towards new opt-in foundation models, not the fine tuning of existing unethical

models. As to when companies hold the rights to work done for hire and want to

license or train off of that, we need regulation.” - Steven Zapata

• “I mean, visual artists, for example, lack the union representation to push back

against contracts that claim all ownership. And without regulation, I think

predatory contracts will just run rampant in this sector. And collective licensing

is also troubling. The early experiments we are seeing with "contributor funds"

from companies like Shutterstock are paying out less than pennies. I mean actual

fractions of a penny per used image. That's all they want to pay a creator for their

life's work and to create a tool that will directly compete against them forever.” -

Steven Zapata