quartz/content/Essays/ai-infringement.md

title	tags	date	draft	lastmod
Generative AI: Copyright Infringement's New Trench Coat	essay ai legal copyright	2023-11-04	false	2024-01-28

One ticket to the original, authorized, or in the alternative, properly licensed audiovisual work, please!

A film roll clatters to the ground from underneath a suspiciously camera-shaped bulge in the figure's oversized trench coat.

[!info] I’m looking for input! Critique my points and make your own arguments. That’s what the comments section is for.

Warning

CW: US law and politics; memes

This site contains my own opinion in a personal capacity, and is not legal advice, nor is it representative of anyone else's opinion.

• Also a reminder that I won’t permit inputting my work in whole or part into an LLM.

I've seen a few news articles and opinion pieces recently that support training generative AI and LLMs (such as ChatGPT/GPT-4, LLaMa, and Midjourney) on the broader internet as well as more traditional copyrighted works, without respect to the copyright holders for all of the above. For now, this will be less of a response to any one article and more of a collection of points of consideration that tie together common threads in public perception. I intend for this to become comprehensive over time.

My opinion here boils down to three main points:

• Training a generative AI model on copyrightable subject matter without authorization is copyright infringement (and the proprietors of the model should be responsible);
• Using a generative AI to generate something where the weights used to determine what the AI outputs are based on copyrightable subject matter which was trained on without authorization is copyright infringement (and the proprietors and users of the model should be jointly responsible); and
• Fair use is not a defense to either of the above infringements.

I also discuss policy later in the essay. Certain policy points are instead made in my Essays/plagiarism, and links to that entry will be labeled with 🅿️.

Prologue: why these arguments are popping up

In short, there's a growing sentiment against copyright in general. Copyright can enable centralization of rights when paired with a capitalist economy, which is what we've been historically experiencing with the advent of record labels/publishing companies. It's even statutorily enshrined as the "work-for-hire" doctrine. AI has the potential to be an end-run around these massive copyright repositories' rights, which many people see as beneficial.

However, this argument forgets that intangible rights are not yet so centralized that independent rights-holders have ceased to exist. While AI will indeed affect central rights-holders, it will also harm individual creators and the bargaining power of those that choose to work with the central institutions. For those against copyright as a whole, I see AI as a neutral factor to the disestablishment of copyright. Due to my roots in the indie music and open-source communities, I'd much rather keep their/our/your rights intact.

Reconciling the two views, I'm sympathetic to arguments against specific parts of the US's copyright regime as enforced by the courts, such as the way fair use is statutorily worded. We as a voting population have the power to compel our representatives to enact reforms that take the threat of ultimate centralization into account, and can even work to break down what's already here. But I don't think that AI should be the impetus for arguments against the system as a whole.

The Legal Argument

Fair warning, this section is going to be the most law-heavy, and probably pretty tech-heavy too. Feel free to skip #The First Amendment and the "Right to Read"

The field is notoriously paywalled, but I'll try to link to publicly available versions of my sources whenever possible. Please don't criticize my sources in this section unless a case has been overruled or a statute has been repealed/amended (i.e., I can't rely on it). This is my interpretation of what's here (again, not legal advice or a professional opinion. Seek legal counsel before acting/refraining from action re: AI). Whether a case is binding on you personally doesn't weigh in on whether its holding is the nationally accepted view.

For all of the below analysis, assume that the hypothetical model in question has been trained on some work which has a US copyright registered with the original author.

The core tenet of copyright is that the doctrine protects original expression (of which regulation is authorized by the Constitution as "works of authorship"), meaning you can't copyright facts. There are two ends to the spectrum of arguments made by authors (seeking protection) and defendants (arguing that enforcement is unnecessary in their case). For example, you can't be sued for using the formula you read in a math textbook, but if you scan that math textbook into a PDF, you might be found liable for infringement.

One common legal argument against training as infringement is that the AI extracts facts, not the author's creativity, from a work. But that position assumes that the AI is capable of first differentiating facts and art, and further separating them in a way analogous to the human mind's.

Training

Everything AI starts with a dataset. And most AI models will start with the easiest, most freely available resource: the internet. Hundreds of different scrapers exist with the goal of collecting as much of the internet as possible to train modern AI (or previously, machine learners, neural networks, or even just classifiers/cluster models). I think that acquiring data without authorization to train an AI on is copyright infringement standing by itself.

Acquiring data for training is an unethical mess. In human terms, scrapers like Common Crawl will take what they want, without asking (unless you know the magic word to make it go away, or just Projects/Obsidian/digital-garden#Block the bot traffic!), and without providing immediately useful services in return like a search engine. For more information on the ethics of AI datasets, read my take on Essays/plagiarism#AI shouldn't disregard the need for attribution, and have a look at the work of Dr. Damien Williams (Mastodon).

• Sidebar: and acquiring this data is copyright infringement too, as unlicensed copying. The case is tremendously stupid: MAI Systems v. Peak Computer holds that RAM copying (ie, moving a file from somewhere to a computer's memory) is an unlicensed copy. As of today, it's still good law, for some reason. Note that every single file you open in Word or a PDF reader; or any webpage in your browser, is moved to your memory before it gets displayed on the screen. Bring it up at trivia night: just using your computer is copyright infringement! It's silly and needs to be overruled going forward, but it is good law.

But then a company actually has to train an AI on that data. What copyright issues does that entail? First, let's talk about The Chinese Room.

The Chinese Room is a philosophical exercise authored by John Searle where the (in context, American) subject is locked in a room and receives symbols in Chinese slipped under the door. A computer program tells the subject what Chinese outputs to send back out under the door based on patterns and combinations of the input. The subject does not understand Chinese. Yet, it appears as if whoever is inside it has a firm understanding of the language to an observer of Searle's room.

Searle's exercise was at the time an extension of the Turing test designed to refute the theory of "Strong AI." At the time that theory was well-named, but today the AI it was talking about is not even considered AI by most. The hypothetical Strong AI was a computer program capable of understanding its inputs and outputs, and importantly why it took each action to solve a problem, with the ability to apply that understanding to new problems (much like our modern conception of Artificial General Intelligence). A Weak AI, on the other hand, was just the Chinese Room: taking inputs and producing outputs among defined rules. Searle reasoned that the "understanding" of a Strong AI was inherently biological, thus one could not presently exist.

• Note that some computer science sources like IBM have taken to using Strong AI to denote only AGI, which was a sufficient, not necessary quality of a philosophical "intelligent" intelligence.

Generative AI models from different sources are architected in a variety of different ways, but they all boil down to one abstract process: tuning an absurdly massive number of parameters to the exact values that produce the most desirable output. (note: CGP Grey's video on AI and its follow-up are mainly directed towards neural networks, but do apply to LLMs, and do a great job illustrating this). This process requires a gargantuan stream of data to use to calibrate those parameters and then test the model. How exactly it parses that incoming data suggests that, even if the method of acquisition is disregarded, the AI model still infringes the input.

The Actual Tech

At the risk of bleeding the #Generation section into this one, generative AI is effectively a very sophisticated next-word predictor based on the words it has read and written previously. If some words being associated with others is more popular historically, then that association is more "correct" to generate in a given scenario than other options. As this relates to training, the only data for that correctness determination is corpus training input. This means that training doesn't have some external indicator of semantics that a secondary natural-language processor on the generation side can incorporate: an AI trains only on the words as they are on the page. Training thus can't be analogized to human learning processes, because when an AI "reads" something, it isn't reading for the forest—it's reading for the trees. Idea and expression in training data are indistinguishable to AI.

As such, modern generative AI, like the statistical data models and machine learners before it, is a Weak AI. And weak AIs use weak AI data. Here's how that translates to copyright.

• Sidebar: this point doesn't consider an AI's ability to summarize a work since the section focuses on how the training inputs are used rather than how the output is generated from real input. This is why I didn't want to get into generation in this section. It's confusing, but training and generation are merely linked concepts rather than direct results of each other when talking about machine learning. Especially when you introduce concepts like "temperature", which is a degree of randomness added to a model's (already variant) choices in response to an user in order to simulate creativity.
• ...I'll talk about that in the next section.

"The Law Part"

All of the content of this section has been to establish how an AI receives data so that I can reason about how it stores that data. In copyright, reproduction, derivatives or compilations of works without authorization can constitute infringement. I believe that inputting a work into a generative AI creates a derivative representation of the work. Eventually, the model is effectively a compilation of all works passed in. And finally (on a related topic), there is nothing copyrightable in how it's arranged the works in that compilation even if every work trained on is authorized.

• ==but because training is deterministic, there's not even any expression in how the data is arranged in its model-representation==

And more cynically, I don't think any of this could be workable in a brief. Looking at how much technical setup I needed to make this argument, there's no way I could compress this all into something a judge could read (even ignoring court rule word limits) or that I could orate concisely to a jury. I'm open to suggestions on a more digestible way to go about arguing the principles I'm concerned about based on this technological understanding of AI.

Detour: point for the observant

The idea and expression being indistinguishable by AI may make one immediately think of merger doctrine. That argument looks like: the idea inherent in the work trained on merges with its expression, so it is not copyrightable. That would not be a correct reading of the doctrine. Ets-Hokin v. Skyy Spirits, Inc. makes it clear that the doctrine is more about disregarding the types of works that are low-expressivity by default, and that this "merge" is just a nice name to remember the actual test by. Confusing name, easy doctrine.

Generation

WIP

Detour: actual harm caused by specific uses of AI models

My bet for a strong factor when courts start applying fair use tests to AI output is harm, in that the AI use in the instant case causes or does not cause harm (and I actually wrote this before the Essays/no-ai-fraud-act 's negligible-harm provision was published. -ed.). Here's a quick list of uses that probably do cause harm.

• Election fraud, including even more corporate influence on US elections (not hypothetical in the slightest, and knowingly unethical)
• Other fraud, like telemarketing/robocalls, phishing, etc
• Competition with actual artists and authors (I am VERY excited to see where trademark law evolves around trademarking one's art or literary style. Currently, the arguments are weak and listed in the mini-argument section).
• Obsoletes human online workforces in tech support, translation, etc
• plagiarism##1 Revealing what's behind the curtain

Where do we go from here?

Well, getting to evaluation of the above by courts would be a start. Right now, courts are ducking AI issues left and right on standing and pleading grounds. Once there's more solid (or honestly any) coverage of the legal arguments on the merits, the reasons why the law should be enforced that way as a matter of policy will become more important.

Policy

These arguments will be more or less persuasive to different people. I think there's a lot more room for discussion here because they become relevant to the future direction of the law as well as current enforcement. The most important debate is up first, but the others are not particularly ordered.

[!info] Section Under Construction More topics under this section forthcoming! I work and edit in an alternate document and copy over sections as I finish them.

Fair Use

WIP

The First Amendment and the "Right to Read"

This argument favors allowing GAI to train on the entire corpus of the internet, copyright- and attribution-free, and bootstraps GAI output into being lawful as well. The position most commonly taken is that the First Amendment protects a citizen's right to information, and that there should be an analogous right for generative AI.

The right to read, at least in spirit, is still being enforced today. Even the 5th Circuit believes that this particular flavor of First Amendment claim will be likely to succeed on appeal after prevailing at the trial level. Book People v. Wong, No. 23-50668 (5th Cir. 2024) (not an AI case). It also incorporates principles from intellectual property law. Notably, that you can read the content of a work without diminishing the value of the author's expression (i.e. ideas aren't copyrightable). As such, the output of an AI is not taking anything from an author that a human wouldn't take when writing something based on their knowledge.

I take issue with the argument on two points that stem from the same technological foundation.

First, as a policy point, the argument incorrectly humanizes current generative AI. There are no characteristics of current GAI that would warrant the analogy between a human reading a webpage and an AI training on that webpage.

Second, more technically, #Training above is my case for why an AI does not learn in the same way that a human does in the eyes of copyright law. ==more==

But for both of these points, I can see where the confusion comes from. The previous leap in machine learning was called "neural networks", which definitely evokes a feeling that it has something to do with the human brain. Even more so when the techniques from neural network learners are used extensively in transformer models (that's those absurd numbers of parameters mentioned earlier).

Mini-arguments

A list of smaller points that would cast doubt on the general zeitgeist around the AI boom that I found compelling. These may be someone else's undeveloped opinion, or it might be a point that I don't think I could contribute to in a rigorous way. Many are spread across the fediverse; others are blog posts or articles. Others still would be better placed a Further Reading section, but I don't like to tack on more than one post-script-style heading.

• Cartoonist Dorothy’s emotional story re: midjourney and exploitation against author intent
• Misinformation worries
• Stronger over time
  • One of the lauded features of bleeding-edge AI is its increasingly perfect recall from a dataset. So you're saying that as AI gets more advanced, it'll be easier for it to exactly reproduce what it was trained on? Sounds like an even better case for copyright infringement.
• Inevitable harm
  • Temperature and the very fact that word generation is used mean that there's no way to completely eliminate hallucination, so truth in AI is unobtainable. Xu, et al.
• Unfair competition
  • This doctrine is a catch-all for claims that don't fit neatly into any of the IP categories, but where someone is still being wronged by a competitor. I see two potential arguments here.
    • First, you could make a case for the way data is scraped from the internet being so comprehensive that there's no way to compete with it by using more fair/ethical methods. This could allow a remedy that mandates AI be trained using some judicially devised (or hey, how about we get Congress involved if they don't like the judicial mechanism), ethical procedure. The arguments are weaker, but they could be persuasive to the right judge.
    • Second, AI work product is on balance massively cheaper than hiring humans, but has little other benefit, and causes many adverse effects. A pure cost advantage providing windfall for one company but not others could also be unfair. Again, it's very weak right now in my opinion.

==TODO analyze and applaud https://www.techdirt.com/2023/11/29/lets-not-flip-sides-on-ip-maximalism-because-of-ai/ ==