When Algorithms Meet Artists: Semantic Compression and Stake-holder Marginalisation in Public AI-Art Discourse (2013-2025)

Imagine you are walking into a giant, noisy town square where everyone is shouting about a new invention: Artificial Intelligence. In this square, there are two main groups of people. One group is the "Builders," the tech companies and scientists who are making the AI. The other group is the "Users," the artists whose work the AI is learning from. The Builders are loud and clear, talking about how amazing, fast, and revolutionary the new machines are. They are the ones holding the megaphones.

But here is the tricky part: How do we know what the Users actually think? In science, especially when studying how people interact with technology, researchers often look at "public discourse." This is just a fancy way of saying "what people say in the news, in legal court papers, in podcasts, and in research articles." They use a tool called "topic modeling," which is like a super-smart sorting machine. It reads thousands of sentences and groups them into buckets based on what they are about. If the news is mostly talking about "AI as a cool tool" and "AI as a scary monster," the machine puts those sentences in two big buckets.

The big question this paper asks is: If we listen to the whole town square, do we actually hear the Users? Or does the noise of the Builders drown them out? The researchers wanted to see if the complex, messy, and varied opinions of real artists were getting a fair share of the microphone, or if their voices were being squashed into a tiny, unrecognizable corner of the conversation.

---

The Great Voice Squeeze

In this study, two researchers decided to play a game of "semantic tag" to see what was really happening in the world of AI and art between 2013 and 2025. They gathered a massive pile of public talk—1,736 chunks of text from news articles, podcasts, court cases, and research papers. They used their sorting machine to organize this pile into 20 different topics. Some topics were about the history of the technology, some were about the philosophy of creativity, and some were about the legal rights of artists.

Then, they brought in the real artists. They took the answers from 252 practicing artists in the US (painters, photographers, writers, musicians, and more) who had been surveyed about their feelings toward AI. These artists had very specific, nuanced opinions. Some thought AI was a great tool but a bad business model. Some wanted to be paid for their work; others were willing to donate it for free. Some were scared; others were curious. In total, these artists expressed 70 unique viewpoints (or "frames") across five main areas: Threat, Utility, Ownership, Transparency, and Compensation.

The researchers then took these 70 unique artist viewpoints and tried to "project" them into the same 20-topic map they had built from the public news. They wanted to see: Where do the artists land on the map?

The Shocking Result: A Tiny Island in a Big Ocean

The answer was startling. The public discourse map was huge and varied, filled with 20 different topics. But when the researchers dropped the artists' voices onto this map, they didn't spread out. They didn't land in the "Legal Rights" bucket or the "Fair Pay" bucket. Instead, 94.9% of all the artists' concerns got squashed into just two topics.

Think of it like this: Imagine a giant library with 20 different rooms. The public is talking about everything from the history of paintbrushes to the math behind the AI. But when the artists try to speak, they are all forced into Room 1 and Room 2.

• Room 1 is labeled "AI as Creative Collaborator."
• Room 2 is labeled "AI Art Authenticity and Human Creativity."

The other 15 rooms in the library, which make up 58.2% of the entire conversation, are completely empty of artist voices. These empty rooms are where the public talks about "AI Copyright and Legal Protection," "Artist Defense Tools," and "AI Art Authorship Debates." The public is having a huge conversation about artists in these rooms, but the artists themselves are not there. It's like a town hall meeting about your neighborhood where the mayor, the police, and the developers are all talking, but the actual residents are locked out of the building.

It's Not Just a Style Mismatch

You might think, "Well, maybe the artists just sound different. Maybe they use short, simple sentences in surveys, while the news uses long, fancy paragraphs. Maybe that's why they don't match up."

The researchers checked this. They took the public news and chopped it up into short, simple sentences that sounded exactly like the artists' survey answers. They tried to match the style. But even when the style was identical, the artists still didn't fit in. The gap wasn't about how they spoke; it was about what they were talking about. The public discourse was simply ignoring the specific, actionable demands of the artists.

The Four Ways the Squeeze Happens

The paper found that this "compression" happens in four sneaky ways:

1. Topical Exclusion: The artists are simply not invited to the party. As mentioned, 15 out of 20 topics have zero artist voices, even though those topics are literally about artist rights and legal defense.
2. Frame Redirection: This is the most confusing part. When artists talk about Ownership (who owns the art?) or Utility (is this tool useful?), their words get rerouted. Instead of landing in a topic about "Money" or "Rights," their words get dumped into the "Authenticity" room. So, an artist saying, "I want to own my work," gets grouped with a philosopher asking, "Is this even real art?" The specific demand for ownership gets lost in a debate about aesthetics.
3. Binary Simplification: The public loves a simple story: Is AI a "Threat" or a "Tool"? The researchers found that the public discourse forces artists into these two boxes. If an artist says, "AI is a threat to my job," the public conversation puts them in the "Threat" box. But if they say, "AI is not a threat," they get put in the "Authenticity" box. The nuance is gone. The public conversation doesn't have a place for an artist who says, "It's a threat to my job, but I also use it to make cool stuff."
4. Voice Collapse: This is the trickiest one. The public discourse groups people together just because they sound similar, even if they mean opposite things. For example, the "Creative Collaborator" room contains artists who love AI and artists who hate it, simply because they both use the same "concerned artist" tone. The public sees a group of artists talking, but they can't tell that half of them want to ban the tech and the other half want to sell it.

Why This Matters

The researchers are very clear: this isn't just a weird statistical quirk. It's a structural problem. When policymakers, judges, and companies look at the news to decide what artists want, they are looking at a distorted picture. They see a simple debate about "AI vs. Human" or "Art vs. Tech." They don't see the complex reality where artists have 70 different, specific demands about how they should be paid, how their data should be used, and what their rights are.

The paper suggests that this "semantic compression" means the primary people affected by AI (the artists) are being pushed to the edge of the conversation. They are the subject of the talk, but they are not the ones doing the talking. The public sphere has captured the idea of AI and art, but it has failed to capture the reality of the artists' concerns. The result is a governance system that might make laws based on a simplified, compressed version of the truth, missing the actual needs of the people it is supposed to protect.