Tag: input-infringement

The Top 3 Generative-AI Copyright Issues

What are your favorite generative-AI copyright issues? In this capsule summary, Cokato attorney Tom James shares what his three favorites are.

Black hole consuming a star. Photo credit: NASA
Black hole consuming a star. Photo credit: NASA.

Having outlined the top AI legal issues, the time has come now to pick favorites. In this capsule summary, Cokato attorney Tom James shares what his three favorite generative-AI copyright issues are.

Generative artificial intelligence refers collectively to technology that is capable of generating new text, images, audio/visual and possibly other content in response to a user’s prompts. They are trained by feeding them mass quantities of ABC (already-been-created) works. Some of America’s biggest mega-corporations have invested billions of dollars into this technology. They are now facing a barrage of lawsuits, most of them asserting claims of copyright infringement.

Issue #1: Does AI Output Infringe Copyrights?

Copyrights give their owners an exclusive right to reproduce their copyright-protected works and to create derivative works based on them. If a generative-AI user prompts the service to reproduce the text of a pre-existing work, and it proceeds to do so, this could implicate the exclusive right of reproduction. If a generative-AI user prompts it to create a work in the style of another work and/or author, this could implicate the exclusive right to create derivative works.

To establish infringement, it will be necessary to prove copying. Two identical but independently created works may each be protected by copyright. Put another way, a person is not guilty of infringement merely by creating a work that is identical or similar to another if he/she/it came up with it completely on his/her/its own.

Despite “training” their proteges on existing works, generative-AI outfits deny that their tools actually copy any of them. They say that any similarity to any existing works, living or dead, is purely coincidental. Thus, OpenAI has stated that copyright infringement is an “unlikely accidental outcome.”

The “accidental outcome” defense seems to me like a hard one to swallow in those cases where a prompt involves creating a story involving a specified fictional character that enjoys copyright protection. If the character is distinctive enough — and a piece of work in and of itself, so to speak — to enjoy copyright protection (such as, say, Mr. Spock from the Star Trek series), then any generated output would seem to be an unauthorized derivative work, at least if the AI tool is any good.

If AI output infringes a copyright in an existing work, who would be liable for it? Potentially, the person who entered the prompt might be held liable for direct infringement. The AI tool provider might arguably be liable for contributory infringement.

Issue #2: Does AI Training Infringe Copyrights?

AI systems are “trained” to create works by exposing a computer program system to large numbers of existing works downloaded from the Internet.

When content is downloaded from the Internet, a copy of it is made. This process will “involve the reproduction of entire works or substantial portions thereof.” In Comment of OpenAI to the USPTO for example, OpenAI acknowledges that its programs are trained on “large, publicly available datasets that include copyrighted works” and that this process “involves first making copies of the data to be analyzed….” Making these copies without permission may infringe the copyright holders’ exclusive right to make reproductions of their works.

Generative-AI outfits tend to argue that the training process is fair use.

Fair use claims require consideration of four statutory fair-use factors:

  • the purpose and character of the use;
  • the nature of the work copied;
  • the amount and substantiality of the portion copied; and
  • the effect on the market for the work.

OpenAI relies on the precedent set in Authors Guild v. Google for its invocation of “fair use.” That case involved Google’s copying of the entire text of books to construct its popular searchable database.

A number of lawsuits currently pending in the courts are raising the question whether and how, the AI training process is “fair use.” For more information, have a look at Does AI Infringe Copyright? 

Issue #3: Are AI-Generated Works Protected by Copyright?

The Copyright Act affords copyright protection to “original works of authorship.” The U.S. Copyright Office recognizes copyright only in works “created by a human being.” Courts, too, have declined to extend copyright protection to nonhuman authors. (Remember the monkey selfie case?) A recent copyright registration applicant has filed a lawsuit (Thaler v. Perlmutter) against the U.S. Copyright Office alleging that the Office wrongfully denied registration of an AI-generated work. A federal court has now rejected his argument that human authorship is not required for copyright ownership. Thaler v. Perlmutter, 1:22-cv-01564, (D.D.C. Aug 18, 2023) ECF No. 24.

In March 2023, Copyright Office Guidance was published. It states that when AI “determines the expressive elements of its output, the generated material is not the product of human authorship.” Moreover, an argument might be made that a general prompt, such as “create a story about a dog in the style of Jack London,” is an idea, not expression. It is well settled that only expression gets copyright protection; ideas do not.

In September 2023, the Copyright Office Review Board affirmed the Office’s refusal to register a copyright in a work that was generated by Midjourney and then modified by the applicant, on the basis that the applicant did not disclaim the AI-generated material.

The Office also has the power to cancel improvidently granted registrations. (Words to the wise: Disclose and disclaim.)

These are my favorite generative-AI legal issues. What are yours?

Visit my extensive Copyright FAQs page.

Why Machine Training AI with Protected Works is Not Fair Use

… if the underlying goal of copyright’s exclusive rights and the fair use exception is to promote new “authorship,” this is doctrinally fatal to the proposal that training AIs on volumes of protected works favors a finding of fair use.

Guest blogger David Newhoff lays out the argument against the claim that training AI systems with copyright-protected works is fair use. David is the author of Who Invented Oscar Wilde? The Photograph at the Center of Modern American Copyright (Potomac Books 2020) and is a copyright advocate/writer at The Illusion of More.

As most copyright watchers already know, two lawsuits were filed at the start of the new year against AI visual works companies. In the U.S., a class-action was filed by visual artists against DeviantArt, Midjourney, and Stability AI; and in the UK, Getty Images is suing Stability AI. Both cases allege infringing use of large volumes of protected works fed into the systems to “train” the algorithms. Regardless of how these two lawsuits might unfold, I want to address the broad defense, already being argued in the blogosphere, that training generative AIs with volumes of protected works is fair use. I don’t think so.

Copyright advocates, skeptics, and even outright antagonists generally agree that the fair use exception, correctly applied, supports the broad aim of copyright law to promote more creative work. In the language of the Constitution, copyright “promotes the progress of science,” but a more accurate, modern description would be that copyright promotes new “authorship” because we do not tend to describe literature, visual arts, music, etc. as “science.”

The fair use doctrine, codified in the federal statute in 1976, originated as judge-made law, and from the seminal Folsom v. Marsh to the contemporary Andy Warhol Foundation v. Goldsmith, the courts have restated, in one way or another, their responsibility to balance the first author’s exclusive rights with a follow-on author’s interest in creating new expression. And as a matter of general principle, it is held that the public benefits from this balancing act because the result is a more diverse market of creative and cultural works.

Fair use defenses are case-by-case considerations and while there may be specific instances in which an AI purpose may be fair use, there are no blanket exceptions. More broadly, though, if the underlying goal of copyright’s exclusive rights and the fair use exception is to promote new “authorship,” this is doctrinally fatal to the proposal that training AIs on volumes of protected works favors a finding of fair use. Even if a court holds that other limiting doctrines render this activity by certain defendants to be non-infringing, a fair use defense should be rejected at summary judgment—at least for the current state of the technology, in which the schematic encompassing AI machine, AI developer, and AI user does nothing to promote new “authorship” as a matter of law.

The definition of “author” in U.S. copyright law means “human author,” and there are no exceptions to this anywhere in our history. The mere existence of a work we might describe as “creative” is not evidence of an author/owner of that work unless there is a valid nexus between a human’s vision and the resulting work fixed in a tangible medium. If you find an anonymous work of art on the street, absent further research, it has no legal author who can assert a claim of copyright in the work that would hold up in any court. And this hypothetical emphasizes the point that the legal meaning of “author” is more rigorous than the philosophical view that art without humans is oxymoronic. (Although it is plausible to find authorship in a work that combines human creativity with AI, I address that subject below.)

As a matter of law, the AI machine itself is disqualified as an “author” full stop. And although the AI owner/developer and AI user/customer are presumably both human, neither is defensibly an “author” of the expressions output by the AI. At least with the current state of technologies making headlines, nowhere in the process—from training the AI, to developing the algorithm, to entering prompts into the system—is there an essential link between those contributions and the individual expressions output by the machine. Consequently, nothing about the process of ingesting protected works to develop these systems in the first place can plausibly claim to serve the purpose of promoting new “authorship.”

But What About the Google Books Case?

Indeed. In the fair use defenses AI developers will present, we should expect to see them lean substantially on the holding in Authors Guild v. Google Books—a decision which arguably exceeds the purpose of fair use to promote new authorship. The Second Circuit, while acknowledging that it was pushing the boundaries of fair use, found the Google Books tool to be “transformative” for its novel utility in presenting snippets of books; and because that utility necessitates scanning whole books into its database, a defendant AI developer will presumably want to make the comparison. But a fair use defense applied to training AIs with volumes of protected works should fail, even under the highly utilitarian holding in Google Books.

While people of good intent can debate the legal merits of that decision, the utility of the Google Books search engine does broadly serve the interest of new authorship with a useful research tool—one I have used many times myself. Google Books provides a new means by which one author may research the works of another author, and this is immediately distinguishable from the generative AI which may be trained to “write books” without authors. Thus, not only does the generative AI fail to promote authorship of the individual works output by the system, but it fails to promote authorship in general.

Although the technology is primitive for the moment, these AIs are expected to “learn” exponentially and grow in complexity such that AIs will presumably compete with or replace at least some human creators in various fields and disciplines. Thus, an enterprise which proposes to diminish the number of working authors, whether intentionally or unintentionally, should only be viewed as devastating to the purpose of copyright law, including the fair use exception.

AI proponents may argue that “democratizing” creativity (i.e., putting these tools in every hand) promotes authorship by making everyone an author. But aside from the cultural vacuum this illusion of more would create, the user prompting the AI has a high burden to prove authorship, and it would really depend on what he is contributing relative to the AI. As mentioned above, some AIs may evolve as tools such that the human in some way “collaborates” with the machine to produce a work of authorship. But this hypothetical points to the reason why fair use is a fact-specific, case-by-case consideration. AI Alpha, which autonomously creates, or creates mostly without human direction, should not benefit from the potential fair use defense of AI Beta, which produces a tool designed to aid, but not replace, human creativity.

Broadly Transformative? Don’t Even Go There

Returning to the constitutional purpose of copyright law to “promote science,” the argument has already been floated as a talking point that training AI systems with protected works promotes computer science in general and is, therefore, “transformative” under fair use factor one for this reason. But this argument should find no purchase in court. To the extent that one of these neural networks might eventually spawn revolutionary utility in medicine or finance etc., it would be unsuitable to ask a court to hold that such voyages of general discovery fit the purpose of copyright, to say nothing of the likelihood that the adventure strays inevitably into patent law. Even the most elastic fair use findings to date reject such a broad defense.

It may be shown that no work(s) output by a particular AI infringes (copies) any of the works that went into its training. It may also be determined that the corpus of works fed into an AI is so rapidly atomized into data that even fleeting “reproduction” is found not to exist, and, thus, the 106(1) right is not infringed. Those questions are going to be raised in court before long, and we shall see where they lead. But to presume fair use as a broad defense for AI “training” is existentially offensive to the purpose of copyright, and perhaps to law in general, because it asks the courts to vest rights in non-humans, which is itself anathema to caselaw in other areas.[1]

It is my oft-stated opinion that creative expression without humans is meaningless as a cultural enterprise, but it is a matter of law to say that copyright is meaningless without “authors” and that there is no such thing as non-human “authors.” For this reason, the argument that training AIs on protected works is inherently fair use should be denied with prejudice.

***** 
n.b.: The Copyright Office has issued New AI Copyright Guidance. 

[1] Cetaceans v. Bus, holding that animals do not have standing in court, was the basis for rejecting PETA’S complaint against photographer Slater for infringing the copyright rights of the monkey in the “Monkey Selfie” fiasco.

 

AI Legal Issues

AI has been trained not just to perform customer service tasks, but also to perform analytics and diagnostic tests; to repair products; to update software; to drive cars; and even to write articles and create images and videos. These developments may be helping to streamline tasks and improve productivity, but they have also generated a range of new legal issues.

AI is not new. Its implementation also is not new. In fact, consumers regularly interact with AI-powered systems every day. Online help systems often use AI to provide quick answers to questions that customers routinely ask. Sometimes these are designed to give a user the impression that s/he is communicating with a person.

AI systems also perform discrete functions such as analyzing a credit report and rendering a decision on a loan or credit card application, or screening employment applications.

Many other uses have been found for AI and new ones are being developed all the time. AI has been trained not just to perform customer service tasks, but also to perform analytics and diagnostic tests; to repair products; to update software; to drive cars; and even to write articles and create images and videos. These developments may be helping to streamline tasks and improve productivity, but they have also generated a range of new legal issues.

"The March of Intgelligence,"19th century illustration of a fantastical giant robot trampling judges and wreaking havoc
(Public domain)

Tort liability

While there are many different kinds of tort claims, the elements of tort claims are basically the same: (1) The person sought to be held liable for damages or ordered to comply with a court order must have owed a duty to the person who is seeking the legal remedy; (2) the person breached that duty; (3) the person seeking the legal remedy experienced harm, i.e., real or threatened injury; and (4) the breach was the actual and proximate cause of the harm.

The kind of harm that must be demonstrated varies depending on the kind of tort claim. For example, a claim of negligent driving might involve bodily injury, while a claim of defamation might involve injury to reputation. For some kinds of tort claims, the harm might involve financial or economic injury. 

The duty may be specified in a statute or contract, or it might be judge-made (“common law.”) It may take the form of an affirmative obligation (such as a doctor’s obligation to provide a requisite level of care to a patient), or it may take a negative form, such as the common law duty to refrain from assaulting another person.

The advent of AI does not really require any change in these basic principles, but they can be more difficult to apply to scenarios that involve the use of an AI system.

Example. Acme Co. manufactures and markets Auto-Doc, a machine that diagnoses and repairs car problems. Mike’s Repair Shop lays off its automotive technician employees and replaces them with one of these machines. Suzie Consumer brings her VW Jetta to Mikes Repair Shop for service because she has been hearing a sound that she describes as being a grinding noise that she thinks is coming from either the engine or the glove compartment. The Auto-Doc machine adds engine oil, replaces belts, and removes the contents of the glove compartment. Later that day, Suzie’s brakes fail and her vehicle hits and kills a pedestrian in a crosswalk. A forensic investigation reveals that her brakes failed because they were badly worn. Who should be held liable for the pedestrian’s death – Suzie, Mike’s, Acme Co., some combination of two of them, all of them, or none of them?

The allocation of responsibility will depend, in part, on the degree of autonomy the AI machine possesses. Of course, if it can be shown that Suzie knew or should have known that her brakes were bad, then she most likely could be held responsible for causing the pedestrian’s death. But what about the others? Their liability, or share of liability, is affected by the degree of autonomy the AI machine possesses. If it is completely autonomous, then Acme might be held responsible for failing to program the machine in such a way that it would test for and detect worn brake pads even if a customer expresses an erroneous belief that the sound is coming from the engine or the glove compartment. On the other hand, if the machine is designed only to offer suggestions of possible problems and solutions,  leaving it up to a mechanic to accept or reject them, then Mike’s might be held responsible for negligently accepting the machine’s recommendations. 

Product liability

 Example.  Acme Co. manufactures and sells Auto-Article, a software program that is designed to create content of a type and kind the user specifies. The purpose of the product is to enable a website owner to generate and publish a large volume of content frequently, thereby improving the website’s search engine ranking. It operates   by scouring the Internet and analyzing instances of the content the user specifies to produce new content that “looks like” them. XYZ Co. uses the software to generate articles on medical topics. One of these articles explains that chest pain can be caused by esophageal spasms but that these typically do not require treatment unless they occur frequently enough to interfere with a person’s ability to eat or drink. Joe is experiencing chest pain. He does not seek medical help, however, because he read the article and therefore believes he is experiencing esophageal spasms. He later collapses and dies from a heart attack. A medical doctor is prepared to testify that his death could have been prevented if he had sought medical attention when he began experiencing the pain.

It is conceivable that an AI system might engage in activity that is prohibited by an applicable jurisdiction’s criminal laws. E-mail address harvesting is an example. In the United States, for example, the CAN-SPAM Act makes it a crime to send a commercial email message to an email address that was  obtained  by automated scraping of Internet websites for email addresses. Of course, if a person intentionally uses an AI system for scraping, then liability should be clear. But what if an AI system “learns” to engage in scraping?

AI-generated criminal output may also be a problem. Some countries have made it a crime to display a Nazi symbol, such as a swastika, on a website. Will criminal liability attach if a website or blog owner uses AI to generate illustrated articles about World War II and the system generates and displays articles that are illustrated with World War II era German flags and military uniforms? In the United States, creating or possessing child pornography is illegal. Will criminal liability attach if an AI system generates it?

Some of these kinds of issues can be resolved through traditional legal analysis of the intent and scienter elements of the definitions of crimes. A jurisdiction might wish to consider, however, whether AI systems should be regulated to require system creators to implement measures that would prevent illegal uses of the technology. This raises policy and feasibility questions, such as whether and what kinds of restraints on machine learning should be required, and how to enforce them. Further, would prior restraints on the design and/or use of AI-powered expressive-content-generating systems infringe on First Amendment rights?  

Forensic and evidentiary issues

In situations involving the use of semi-autonomous AI, allocating responsibility for harm resulting from the operation of the AI  system  may be difficult. The most basic question in this respect is whether an AI system was in use or not. For example, if a motor vehicle that can be operated in either manual or autonomous mode is involved in an accident, and fault or the extent of liability depends on that (See the discussion of tort liability, above), then a way of determining the mode in which the car was being driven at the time will be needed.

If, in the case of a semi-autonomous AI system, tort liability must be allocated between the creator of the system and a user of it, the question of fault may depend on who actually caused a particular tortious operation to be executed – the system creator or the user. In that event, some method of retracing the steps the AI system used may be essential. This may also be necessary in situations where some factor other than AI contributed, or might have contributed, to the injury. Regulation may be needed to ensure that the steps in an AI system’s operations are, in fact, capable of being ascertained.

Transparency problems also fall into this category. As explained in the Journal of Responsible Technology, people might be put on no-fly lists, denied jobs or benefits, or refused credit without knowing anything more than that the decision was made through some sort of automated process. Even if transparency is achieved and/or mandated, contestability will also be an issue.

Data Privacy

To the extent an AI system collects and stores personal or private information, there is a risk that someone may gain unauthorized access to it.. Depending on how the system is designed to function, there is also a risk that it might autonomously disclose legally protected personal or private information. Security breaches can cause catastrophic problems for data subjects.

Publicity rights

Many jurisdictions recognize a cause of action for violation of a person’s publicity rights (sometimes called “misappropriation of personality.”) In these jurisdictions, a person has an exclusive legal right to commercially exploit his or her own name, likeness or voice. To what extent, and under what circumstances, should liability attach if a commercialized AI system analyzes the name, likeness or voice of a person that it discovers on the Internet? Will the answer depend on how much information about a particular individual’s voice, name or likeness the system uses, on one hand, or how closely the generated output resembles that individual’s voice, name or likeness, on the other?

Contracts

The primary AI-related contract concern is about drafting agreements that adequately and effectively allocate liability for losses resulting from the use of AI technology. Insurance can be expected to play a larger role as the use of AI spreads into more areas.

Bias, Discrimination, Diversity & Inclusion

Some legislators have expressed concern that AI systems will reflect and perpetuate biases and perhaps discriminatory patterns of culture. To what extent should AI system developers be required to ensure that the data their systems use are collected from a diverse mixture of races, ethnicities, genders, gender identities, sexual orientations, abilities and disabilities, socioeconomic classes, and so on? Should developers be required to apply some sort of principle of “equity” with respect to these classifications, and if so, whose vision of equity should they be required to enforce? To what extent should government be involved in making these decisions for system developers and users?

Intellectual Property

AI-generated works like articles, drawings, animations, music and so on, raise three kinds of copyright issues:

  1. Input infringement issues, i.e., questions like whether AI systems that are designed to create new works based on existing copyright-protected works infringe the copyrights in those works. The highest-stake legal battleground today centers on this. I provide a layperson-friendly explanation of this issue and the initial lawsuits raising it in Does AI Infringe Copyright?  
  2. Output infringement issues, i.e., whether output generated by AI tools infringes copyrights in the works on which the AI was trained.
  3. Copyright in output. To what extent, if at all, may copyright be claimed in AI-generated output?
Large computer with a woman reading a notebook and flipping a switch, B&W, from 1950's

Patents, Trademarks, and other IP

Computer programs can be patented. AI systems can be devised to write computer programs. Can an AI-generated computer program that meets the usual criteria for patentability (novelty, utility, etc.) be patented?

Contact Minnesota attorney Thomas James for help with copyright and trademark registration and other copyright and trademark-related matters.