from the two-wrongs-don't-make-a-copyright dept

There’s a lot wrong with AI. But bad things happen when we try to use copyright law to fix it. Which is what we just told the Southern District of New York, where OpenAI and Microsoft are defending themselves in several cases alleging that their training violated the copyrights of the works the AI models were trained on. We made a few points.

One overarching point is that what the plaintiffs are trying to do is not what copyright law is for. Copyright gives the holder the ability to control how the embodiment of the work is used (ex: you can’t copy a book without permission, or, as the example we used in the brief, perform a play) but not the ability to control the public from benefiting from its contents. Copyright can control such activities as copying, distributing, or performing a work but NOT reading, watching, listening to, or learning from a work. What OpenAI does with its training is the latter set of activities, not the former. But if OpenAI is somehow not able to engage in these activities because of copyright, then no one can. Which would lead to a direct harm to the public, a loss of an important benefit copyright law is actually supposed to give it.

In fact, it is a benefit copyright law must provide the public. The whole point of copyright law is to promote “the progress of sciences and the useful arts,” or, in other words, help ensure we get more innovation and expression. But the plaintiffs in these cases want the opposite: they want copyright to stop these things. They want copyright law to prevent AI from being further developed (indeed, in its complaint the New York Times Company even asked for OpenAI’s models to be destroyed!) by denying it the ability to do the learning that is needed for it to be useful. And they want to prevent any of the further expression that AI helps facilitate, even though fostering new expression is exactly what copyright law is supposed to do.

In fact, it is what copyright law is constitutionally required to do. Moreover, if, rather than fostering innovation and expression, copyright law instead obstructs it, then copyright law is not constitutional. Congress only gets its authority to make copyright law through the Progress Clause (“to promote the progress of science and the useful arts…”) so if it turns out that the law it has passed instead does the opposite of promoting progress, then that law is unconstitutional, as being beyond its authority.

The First Amendment also limits Congress’s authority to pass laws that inhibit free expression. So Congress also would not have had the authority to pass a law that let copyright holders of existing works prevent either the development of AI or any of the downstream, lawful expression AI helps facilitate. If the copyright statute Congress passed were to have that effect, then it would be unconstitutional.

It also would violate the Constitution in another way we’ve discussed before: by violating the First Amendment-protected “right to read,” which is sometimes also referred to as the right to receive information. All those learning activities (reading, watching, listening to, etc.), which the current statute does not actually restrict, are activities that the First Amendment entitles the public to do, and copyright law not only does not prevent, but constitutionally could not prevent.

Meanwhile, we also made an overarching point: even to the extent that AI as we know it may be terrible, and we are looking for a law that can stop it, copyright law is still not the right law to do it. First of all, it’s not tailored to the actual problems we’d want to address (for instance, if using AI in hiring decisions produces discriminatory results because of bias incorporated into the AI model, then the solution to that problem is making sure that AI can train on even more material, not less, because it is the training limits that tend to create that bias). Secondly, the worse we think today’s AI is, the more important it is that something better is developed, but if law can stand in the way of further development, then none will be. None can be, and we’ll be stuck with the technology we have, if not an even worse version of it if we have to degrade the models in response to copyright claims. As well as stuck with an imbalanced, super-powered copyright law that could stand in the way of lots of other non-AI innovation and expression.

And, again, if we have to freeze the technology because of copyright law, then copyright law is definitely running afoul of the Constitution, which allows Congress to pass a copyright law that only does the opposite.

In sum, the worse you think AI is—and we make no quibble with that assessment, although we did point out that it also has some distinct upsides, particularly when it comes to enabling more expression—the more important it is that these copyright claims be dismissed. Because however bad the state of play may be today, a copyright law that would let these claims prevail will only make everything worse.

Filed Under: ai, ai training, copyright, fair use, innovation
Companies: authors guild, microsoft, ny times, openai