How Copyright Should Work in the Age of AI

"Thanks to generative AI systems like ChatGPT, GPT-4 and Dall-E 2, novels can now be generated in days, songs composed faster than you can play them and pictures drawn in seconds -- all based on massive machine-learning models that have sampled and remixed the canon of published works.

That raises some difficult issues around copyright rules. So as part of our series of ethical dilemmas sparked by AI, we wanted to know how readers think copyright law needs to change.

Specifically, we asked: It isn't currently possible to copyright the output of an AI. Should it be? If so, who should own the copyright: the user, the AI provider, the people whose material trained the AI? All of them?

Here are some of the responses.

Training data voids copyright

People using generative AI gave up the right to owning any results the moment they engaged with any tool that was trained on other people's work. Without a viable way to discern whether someone else's art/words/music was used in the making of something AI-generated, there is no room for any claim of copyright.

This could change if there are more data sets that have been legally obtained and curated. It's in the interest of AI-application users to push for transparent data sets, because that's the only shot they have of claiming even partial ownership of output. It's likely, however, that if ownership rights are further delineated, the AI-application developers will want a share, and they are likely to have grounds for a claim, too.

-- Rebecca Kobayashi, Fairfax, Va.

It's too easy to copy now

Copyright is needed in an environment where it is difficult to generate an original, and easy to generate a copy. For example, copyright could protect Monet from somebody making an exact replica of his paintings, but wouldn't protect him from someone painting a different scenery using the impressionist style. And such a protection was neither necessary nor desired because making another impressionist painting was no easy task.

That is, until today, when imitating a style or combining several styles is as easy as running a copy machine. Therefore I expect copyright to be replaced by another law, which will also consider the copying of style. It may merely require an acknowledgment, as is the case today with quoting another author in a paper. It may allow authors or publishers to prohibit their work from being used for training just as they can prevent replication today. Or it may involve compensation for making works of arts available for training. I do not know how flexible that law will be, but I do expect copyright laws to be replaced in this way.

-- Daniel Brand, Kailua, Hawaii .

Humans only

Only humans should be granted copyrights. AI should be held responsible if it violates copyright protections during any of its operations.

AI is generally used in activities or transactions intended to create value. If AI profits from the work of others without credit or permission it should be considered a violation of the law.

-- Peter Goulet, Lee's Summit, Mo.

An art to the science

ChatGPT's responses improve with the quality of the prompts. There's a strategy and perhaps an art to prompting this tool so that it returns a better answer. In that context, then yes, being that the answer is the result of a human's prompting (which is theirs, right?), then the human should be able to copyright it. Again, the owner of the "David" sculpture is not the guy who made the chisel and hammer.

-- Timothy McGuckin, Reston, Va.

Copyright for AI? Of course not.

No, you should never be able to copyright data sorting. That would be like copyrighting a Google search -- ridiculous! All copyrights belong to those who created the original material that was sorted by AI. The people who trained the AI deserve solid pay but not copyrights.

-- Cheryl Miller, Pahrump, Nev

.

Open to all

AI should be open source, with no possibility of copyright for its output.

-- Nicholas J. Rao, Roswell, Ga

If AI becomes sentient. . .

If the AI is not sentient, then copyright should attach to the person who worked with the AI (even if it was just a request to create or think about an item). If AI becomes sentient, then copyright should belong to the AI -- since it would be a separate, self-aware intelligence.

-- John Parziale, Weedsport, N.Y.

Amended copyright

I believe that when an artificial Intelligence service produces an article, play, story, etc., for an end user without charge, then the entity providing the AI service could retain partial ownership and be compensated if that work becomes a financial success. If, however, the user pays an upfront fee to use the AI service then the AI (owner) would be due no additional compensation.

I would suggest a series of questions in the user agreement that the consumer would be required to navigate through that highlight the expectations of the AI service provider and the potential financial impact on the end user. For example: "Do you realize that XYZ AI Inc. will retain 50% of the copyright for any work you sell that XYZ AI Inc. helps you create?"

-- Gregory Pyne, Sandy Springs, Ga.

There's nothing really new

I believe that in the big scheme, all combinations of thoughts were once already thought of. So my input, also, is not new.

Hence my answer must be a no to copyrighting anything by AI. Any "story" is only rehashed and reassembled. As such, it is difficult enough, if not impossible, to come up with a new story line. Let us not make it worse by allowing AI to protect "its" property. Someone would just let the computer crank away and "write" all possible iterations -- similar to a chess computer figuring all possible moves.

-- Tim Haake, Charlotte, N.C.

---

Demetria Gallegos is an editor for The Wall Street Journal in New York. Email her at demetria.gallegos@wsj.com." [1]

1. C-Suite Strategies (A Special Report) --- How Copyright Should Work in the Age of AI: Readers Weigh In: New tools raise all sorts of complicated, legal questions. Most prominently: Should people be allowed to copyright AI output? And if so, who gets to be the owner? Gallegos, Demetria.  Wall Street Journal, Eastern edition; New York, N.Y.. 29 Jan 2024: R.6.