This guest post was jointly authored by Mariah Johnson and Marcus Liou, student attorneys in Georgetown’s Intellectual Property and Information Policy (iPIP) Clinic.

Generative AI (GenAI) systems perpetuate biases, and authors can have a potent role in mitigating such biases.

But GenAI is generating controversy among authors. Can authors do anything to ensure that these systems promote progress rather than prevent it? Authors Alliance believes the answer is yes, and we worked with them to launch a new zine, Putting the AI in Fair Use: Authors’ Abilities to Promote Progress, that demonstrates how authors can share their works broadly to shape better AI systems. Drawing together Authors Alliance’s past blog posts and advocacy discussing GenAI, copyright law, and authors, this zine emphasizes how authors can help prevent AI bias and protect “the widest possible access to information of all kinds.” 

As former Copyright Register Barbara Ringer articulated, protecting that access requires striking a balance with “induc[cing] authors and artists to create and disseminate original works, and to reward them for their contributions to society.” The fair use doctrine is often invoked to do that work. Fair use is a multi-factor standard that allows limited use of copyrighted material—even without authors’ credit, consent, or compensation–that asks courts to examine:

(1) the purpose and character of the use, 

(2) the nature of the copyrighted work, 

(3) the amount or substantiality of the portion used, and 

(4) the effect of the use on the potential market for or value of the work. 

While courts have not decided whether using copyrighted works as training data for GenAI is fair use, past fair use decisions involving algorithms, such as Perfect 10, iParadigms, Google Books, and HathiTrust favored the consentless use of other people’s copyrighted works to create novel computational systems. In those cases, judges repeatedly found that algorithmic technologies aligned with the Constitutional justification for copyright law: promoting progress.

But some GenAI outputs prevent progress by projecting biases. GenAI outputs are biased in part because they use biased, low friction data (BLFD) as training data, like content scraped from the public internet. Examples of BLFD include Creative Commons (CC) licensed works, like Wikipedia, and works in the public domain. While Wikipedia is used as training data in most AI systems, its articles are overwhelmingly written by men–and that bias is reflected in shorter and fewer articles about women. And because the public domain cuts off in the mid-1920s, those works often reflect the harmful gender and racial biases of that time. However, if authors allow their copyrighted works to be used as GenAI training data, those authors can help mitigate some of the biases embedded in BLFD. 

Current biases in GenAI are disturbing. As we discuss in our zine, word2vec is a very popular toolkit used to help machine learning (ML) models recognize relationships between words–like women as homemakers and Black men with the word “assaulted.” Similarly, OpenAI’s GenAI chatbox ChatGPT, when asked to generate letters of recommendation, used “expert,” “reputable,” and “authentic” to describe men and  “beauty,” “stunning,” and “emotional” for women, discounting women’s competency and reinforcing harmful stereotypes about working women. An intersectional perspective can help authors see the compounding impact of these harms. What began as a legal framework to describe why discrimination law did not adequately address harms facing Black women, it is now used as a wider lens to consider how marginalization affects all people with multiple identities. Coined by Professor Kimberlé Crenshaw in the late 1980s, intersectionality uses critical theory like Critical Race Theory, feminism, and working-class studies together as “a lens . . . for seeing the way in which various forms of inequality often operate together and exacerbate each other.” Contemporary authors’ copyrighted works often reflect the richness of intersectional perspectives, and using those works as training data can help mitigate GenAI bias against marginalized people by introducing diverse narratives and inclusive language. Not always–even recent works reflect bias–but more often than might be possible currently.

Which brings us back to fair use. Some corporations may rely on the doctrine to include more works by or about marginalized people in an attempt to mitigate GenAI bias. Professor Mark Lemley and Bryan Casey have suggested “[t]he solution [to facial recognition bias] is to build bigger databases overall or to ‘oversample’ members of smaller groups” because “simply restricting access to more data is not a viable solution.” Similarly, Professor Matthew Sag notes that “[r]estricting the training data for LLMs to public domain and open license material would tend to encode the perspectives, interests, and biases of a distinctly unrepresentative set of authors.” However, many marginalized people may wish to be excluded from these databases rather than have their works or stories become grist for the mill. As Dr. Anna Lauren Hoffman warns, “[I]nclusion reinforces the structural sources of violence it supposedly addresses.”

Legally, if not ethically, fair use may moot the point. The doctrine is flexible, fact-dependent, and fraught. It’s also fairly predictable, which is why legal precedent and empirical work have led many legal scholars to believe that using copyrighted works as training data to debias AI will be fair use–even if that has some public harms. Back in 2017, Professor Ben Sobel concluded that “[i]f engineers made unauthorized use of copyrighted data for the sole purpose of debiasing an expressive program, . . . fair use would excuse it.” Professor Amanda Levendowski has explained why and how “[f]air use can, quite literally, promote creation of fairer AI systems.” More recently, Dr. Mehtab Khan and Dr. Alex Hanna  observed that “[a]ccessing copyright work may also be necessary for the purpose of auditing, testing, and mitigating bias in datasets . . . [and] it may be useful to rely on the flexibility of fair use, and support access for researchers and auditors.” 

No matter how you feel about it, fair use is not the end of the story. It is ill-equipped to solve the troubling growth of AI-powered deepfakes. After being targeted by sexualized deepfakes, Rep. Ocasio-Cortez described “[d]eepfakes [as] absolutely a way of digitizing violent humiliation against other people.” Fair use will not solve the intersectional harms of AI-powered face surveillance either. Dr. Joy Buolamwini and Dr. Timnit Gebru evaluated leading gender classifiers used to train face surveillance technologies and discovered that they more accurately classified males over females and lighter-skinned over darker-skinned people. The researchers also discovered that the “classifiers performed worst on darker female subjects.” While legal scholars like Professors Shyamkrishna Balganesh, Margaret Chon, and Cathay Smith argue that copyright law can protect privacy interests, like the ones threatened by deepfakes or face surveillance, federal privacy laws are a more permanent, comprehensive way to address these problems.

But who has time to wait on courts and Congress? Right now, authors can take proactive steps to ensure that their works promote progress rather than prevent it. Check out the Authors Alliance’s guides to Contract Negotiations, Open Access, Rights Reversion, and Termination of Transfer to learn how–or explore our new zine, Putting the AI in Fair Use: Authors’ Abilities to Promote Progress.

You can find a PDF of the Zine here, as well as printer-ready copies here and here.

By Dave Hansen

Have you ever used a photograph to illustrate a historical event in your writing? Or quoted, say from a letter, to point out some fact that the author conveyed in their writing? According to the 10th Circuit, these aren’t the kinds of uses that fair use supports. 

On Thursday, Authors Alliance joined with EFF, the Association of Research Libraries, the American Library Association, and Public Knowlege in filing an amicus brief asking the 10th Circuit Court of Appeals to reconsider its recent fair use decision in Whyte Monkee v. Netflix. 

The case is about Netflix’s use of a funeral video recording in its documentary series Tiger King, a true crime documentary about Joseph Maldanado, aka Joe Exotic, an excentric zookeeper, media personality, exotic animal owner, and convicted felon. The recording at issue was created by Timothy Sepi/Whyte Monkee, as a memorial for Travis Maldonado, Joe Exotic’s late husband. Netflix used about 60 seconds of the funeral video in its show. Its purpose was, among other things, to “illustrate Mr. Exotic’s purported megalomania, even in the face of tragedy.” 

A three-judge panel of the 10th Circuit issued its opinion in late March, concluding that Netflix’s use was not “transformative” under the first fair use factor and therefore disfavored as a fair use. The panel relied heavily on the Supreme Court’s recent decision in Andy Warhol v. Goldsmith, taking that case to mean that uses that do not comment or criticize the artistic and creative aspects of the underlying work are generally disfavored. So, the court concluded: 

Defendants’ use of the Funeral Video is not transformative under the first fair use factor. Here, Defendants did not comment on or “target” Mr. Sepi’s work at all; instead, Defendants used the Funeral Video to comment on Joe Exotic. More specifically, Defendants used the Funeral Video to illustrate Mr. Exotic’s purported megalomania, even in the face of tragedy. By doing so, Defendants were providing a historical reference point in Mr. Exotic’s life and commenting on Mr. Exotic’s showmanship. However, Defendants’ use did not comment on Mr. Sepi’s video—i.e., its creative decisions or its intended meaning.

You can probably see the problem. Fair use has, for a very long time, supported a wide variety of other uses that incorporate existing works as historical reference points and illustrations. Although the Supreme Court talked a lot about criticism and comment in its Warhol opinion (which made sense, given that the use before it was a purported artistic commentary), I think very few people interpreted that decision to mean that only commentary and criticism are permissible transformative fair uses. But as our brief points out, the panel’s decision essentially converts the Supreme Court’s decision in Warhol from a nuanced reaffirmation of fair use precedent into a radical rewrite of the law that only supports those kinds of uses. 

Our brief argues that the 10th Circuit misread the Supreme Court’s opinion in Warhol, and that it ignored decades of fair use case law. We point to a few good examples – e.g., Time v. Bernard Geis (a 1968 case finding fair use of a recreation of the famous Zapruder film in a book titled “Six Seconds in Dallas,” analyzing President Kennedy’s assassination), New Era Publications v. Carol Publishing (a 1990 case supporting reuse of lengthy quotations of L Ron Hubbard in a book about him, to make a point about Hubbard’s “hypocrisy and pomposity”) and Bill Graham Archives v. Dorling Kindersley (a 2006 case finding fair use of Grateful Dead concert posters in a book using them as historical reference points). 

Our brief also highlights how communities of practice such as documentary filmmakers, journalists, and nonfiction writers have come to rely on fair use to support these types of uses–so much so that these practices are codified in best practices here, here, and even here Authors Alliance’s own Fair Use for Nonfiction Authors guide. 

Although it is rare for appellate courts to grant rehearing of already issued opinions, this opinion has drawn quite a lot of negative attention. In addition to our amicus brief, there were amicus briefs filed in support of rehearing from: 

• The International Documentary Association, Film Independent, Kartemquin Educational Films, Women In Film, The University Film And Video Association 
• 31 copyright and media law professors
• The Motion Picture Association 

Given the broad and negative reach of this decision, I hope the 10th Circuit will pay attention and grant the request. 

Today we are very pleased to share an open letter regarding copyright reform on behalf of South African authors. The letter is available here and is also available as a PDF (with names as of today) here.

The letter comes at a critical decision making moment for South Africa’s Copyright Amendment Bill which has been debated for years (read more here and here on our views). We believe it is important for lawmakers to hear from authors who support this bill, and in particular hear from us about why we view its fair use provisions and author remuneration provisions so positively.

We welcome other South African authors to add their names to the letter to express their support. You can do so by completing this form.

Last week you may have read about a website called prosecraft.io, a site with an index of some 25,000 books that provided a variety of data about the texts (how long, how many adverbs, how much passive voice) along with a chart showing sentiment analysis of the works in its collection and displayed short snippets from the texts themselves, two paragraphs representing the most and least vivid from the text. Overall, it was a somewhat interesting tool, promoted to authors to better understand how their work compares to those of other published works. 

The news cycle about prosecraft.io was about the campaign to get its creator Benji Smith to take the site down (he now has) based on allegations of copyright infringement. A Gizmodo story about it generated lots of attention, and it’s been written up extensively, for example here, here, here, and here.  

It’s written about enough that I won’t repeat the whole saga here. However, I think a few observations are worth sharing:  

1) Don’t get your legal advice from Twitter (or whatever its called)

“Fair Use does not, by any stretch of the imagination, allow you to use an author’s entire copyrighted work without permission as a part of a data training program that feeds into your own ‘AI algorithm.’”  – Linda Codega, Gizmodo (a sentiment that was retweeted extensively)

Fair use actually allows quite a few situations where you can copy an entire work, including situations when you can use it as part of a data training program (and calling an algorithm “AI” doesn’t magically transform it into something unlawful). For example, way back in 2002 in Kelly v. Ariba Soft, the 9th Circuit concluded that it was fair use to make full text copies of images found on the internet for the purpose of enabling web image search. Similarly, in AV ex rel Vanderhye v. iParadigms, the 4th Circuit in 2009 concluded that it was fair use to make full text copies of academic papers for use in a plagiarism detection tool.  

Most relevant to prosecraft, in Authors Guild v. HathiTrust (2014)  and Authors Guild v. Google (2015) the Second Circuit held that Google’s copying of millions of books for purposes of creating a massive search engine of their contents was fair use . Google produced full-text searchable databases of the works, and displayed short snippets containing whatever term the user had searched for (quite similar to prosecraft’s outputs). That functionality also enabled a wide range of computer-aided textual analysis, as the court explained: 

The search engine also makes possible new forms of research, known as “text mining” and “data mining.” Google’s “ngrams” research tool draws on the Google Library Project corpus to furnish statistical information to Internet users about the frequency of word and phrase usage over centuries.  This tool permits users to discern fluctuations of interest in a particular subject over time and space by showing increases and decreases in the frequency of reference and usage in different periods and different linguistic regions. It also allows researchers to comb over the tens of millions of books Google has scanned in order to examine “word frequencies, syntactic patterns, and thematic markers” and to derive information on how nomenclature, linguistic usage, and literary style have changed over time. Authors Guild, Inc., 954 F.Supp.2d at 287. The district court gave as an example “track[ing] the frequency of references to the United States as a single entity (‘the United States is’) versus references to the United States in the plural (‘the United States are’) and how that usage has changed over time.”

While there are a number of generative AI cases pending (a nice summary of them is here) that I agree raise some additional legal questions beyond those directly answered in Google Books, the kind of textual analysis that prosecraft.io offered seems remarkably similar to the kinds of things that the courts have already said are permissible fair uses. 

2) Text and data mining analysis has broad benefits

Not only is text mining fair use, it also yields some amazing insights that truly “promote the progress of Science,” which is what copyright law is all about.  Prosecraft offered some pretty basic insights into published books – how long, how many adverbs, and the like. I can understand opinions being split on whether that kind of information is actually helpful for current or aspiring authors. But, text mining can reveal so much more. 

In the submission Authors Alliance made to the US Copyright Office three years ago in support of a Section 1201 Exemption permitting text data mining, we explained:

TDM makes it possible to sift through substantial amounts of information to draw groundbreaking conclusions. This is true across disciplines. In medical science, TDM has been used to perform an overview of a mass of coronavirus literature.Researchers have also begun to explore the technique’s promise for extracting clinically actionable information from biomedical publications and clinical notes. Others have assessed its promise for drawing insights from the masses of medical images and associated reports that hospitals accumulate. 

In social science, studies have used TDM to analyze job advertisements to identify direct discrimination during the hiring process.⁷ It has also been used to study police officer body-worn camera footage, uncovering that police officers speak less respectfully to Black than to white community members even under similar circumstances.

TDM also shows great promise for drawing insights from literary works and motion pictures. Regarding literature, some 221,597 fiction books were printed in English in 2015 alone, more than a single scholar could read in a lifetime. TDM allows researchers to “‘scale up’ more familiar humanistic approaches and investigate questions of how literary genres evolve, how literary style circulates within and across linguistic contexts, and how patterns of racial discourse in society at large filter down into literary expression.” TDM has been used to “observe trends such as the marked decline in fiction written from a first-person point of view that took place from the mid-late 1700s to the early-mid 1800s, the weakening of gender stereotypes, and the staying power of literary standards over time.” Those who apply TDM to motion pictures view the technique as every bit as promising for their field. Researchers believe the technique will provide insight into the politics of representation in the Network era of American television, into what elements make a movie a Hollywood blockbuster, and into whether it is possible to identify the components that make up a director’s unique visual style [citing numerous letters in support of the TDM exemption from researchers].

3) Text and data mining is not new and it’s not a threat to authors

Text mining of the sort it seemed prosecraft employed isn’t some kind of new phenomenon. Marti Hearst, a professor at UC Berkeley’s iSchool explained the basics in this classic 2003 piece. Scores of computer science students experiment with projects to do almost exactly what prosecraft was producing in their courses each year. Textbooks like Matt Jockers’s Text Analysis with R for Students of Literature have been widely used and adopted all across the U.S. to teach these techniques. Our submissions during our petition for the DMCA exemption for text and data mining back in 2020 included 14 separate letters of support from authors and researchers engaged in text data mining research, and even more researchers are currently working on TDM projects. While fears over generative AI may be justified for some creators (and we are certainly not oblivious to the threat of various forms of economic displacement), it’s important to remember that text data mining on textual works is not the same as generative AI. On the contrary, it is a fair use that enriches and deepens our understanding of literature rather than harming the authors who create it.