Fair use is famously whatever a court says fair use is after you have spent all your money on lawyers, so I fully expect that the Copyright Office will end up allowing the AI companies to, at a minimum, get away with taking material from writers and authors for their training data without compensation that they have already swallowed. This would be a mistake, bad for both artists and the larger culture.
First, we have to understand that these systems do not actually learn. They are imitative systems — they merely copy what they take in and rearrange it. Sometimes, to the point of outright theft. I have written about this before, but they are not creating in the sense that humans do. To be utterly boorish and quote myself:
A member of my writing group told me the other day that my writing reminds them of Terry Pratchett. This is both incredibly flattering and odd — I do not write satire. What I do write are sarcastic fantasy stories that leverage actual, if slightly warped, history to make their thematic points. Pratchett is an obvious touchpoint to work like that to anyone who has read him closely. But an AI would not make the transformation I have (whether my writing is good or not is another question – we are discussing the process of creation). Such systems are not capable of that kind of growth.
Human beings do not, despite what some AI proponents would have you believe, merely remix everything they have ever read or seen. They transform it in ways that go beyond remixing. They process what they learn from based on their own lived experiences, biases, and preferences and, in the best cases, turn that material into something different than what came before. If you study the history of the best impressionists, for example, you can see that they started as masters of realistic paintings. Only gradually did they invent their impressionistic style, taking what they had learned and turning it into something new.
By allowing AI companies to take material for training without compensation, you are violating the spirit of intellectual property laws in this country. The Constitution allows for protection if intellectual property in order to “… promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries.” AI systems that do not have to pay for their training data or get permission to collect that data violate that contract.
Imitative AI systems make their money by, ultimately, replacing human labor. There is no other way for them to be profitable. But they cannot replace human labor without the fruits of human labor to copy, to imitate in the first place. However, if you do not compensate people for the labor you take from them to train upon, if you are allowed to take their labor without permission and/or payment, then eventually there is little economic incentive for people to continue to create art and thus less art. Eventually, we end up with the majority of art being derived from imitative systems, training themselves on themselves. A recursive series of copies that leads to nothing new, nothing but a faded xerox of a decayed culture.
That is not the intention of the intellectual property clause in the constitution. I realize that money in our society drives the law, but allowing AI to treat taking training data as fair use leads to a result that is completely at odds with the spirit of the laws in question. They were meant to protect culture and the people who produce it. Give the AI companies their way on these issues, and you will reduce culture to a cul-de-sac, incapable of moving forward, doing nothing but going round and round forever, imitative AI systems endlessly copying degraded copies of themselves.