The future of the Internet is here and many fear it. Will Generative AI take the bread out of content creators’ mouths?

Disinformation, manipulation and deepfake

Angie Holan, director of the International Fact-Checking Network, an initiative of the Poynter Institute (USA), which promotes fact-checking by creating standards in the industry, believes that we don’t yet fully understand how AI technology will change the media landscape. The problem arises because this technology can be used for good as well as bad purposes.

“We are seeing people who are doing political campaigns that use misinformation and use AI. So, we’re seeing things like deepfake audio. There have been multiple cases that have been covered in the mainstream media about people making fake audio messages that sound like you know, Joe Biden. So, they’re false messages with different purposes. Some of them tend to be like false endorsements, some of them have been like false information on voting days and times. So, just basically in any context where you might find misinformation, you can also find AI-generated misinformation”, says Holan.

A recent Recorder article shows that the AI-fueled deepfake phenomenon is also rife in Romania. Numerous fake ads created with AI are circulating on social networks, imitating the voices of celebrities, politicians or business people and using video snippets from various shows or podcasts to promote various get-rich-quick schemes and investment scams. Basically, the journalists demonstrate, using AI has created complex criminal networks that lure thousands of gullible users and leave them without money in their bank accounts.

Angie Holan recounts that there are also AI bots that post information at a very fast pace, and this creates a problem for news organizations, which can’t keep up with automated systems.

Even when there are no dishonest intentions at play, AI can still misinform the public. AI systems sometimes “hallucinate”, that is, certain algorithm errors provide information that is not real (most AI developers claim they try to minimize these software errors).

“From my point of view, one of the big problems with AI is that it does not have good safeguards for accuracy built into it. A Language Learning Model (note: LLM – the “learning” algorithm on which AI systems are built) works by language prediction, and when the AI models run into some sort of knowledge gap, they just flow in words. And so, we see, just recently, Google tried to roll out an an AI assistant and it told people to eat pizza with glue on it”, Holan adds.

Lack of clear policies against fake AI content harms writers

Jane Friedman, who has 25 years of experience in the publishing industry, is a writer and editor of The Hot Sheet, a newsletter about the industry. In the summer of 2023, she was contacted by a reader. He had discovered that a number of the author’s books were on sale on Amazon. Friedman didn’t know what the reader was referring to, so she checked the platform and discovered titles that bore her name but that she hadn’t written.

“I informed Amazon that someone was using my name on books I didn’t write. But since using AI to generate a book isn’t against the law, and there are many people with my name, Amazon didn’t immediately take any action” Jane Friedman says, in an interview for Panorama via Google Meet.

“It was only after some persistence that they finally took the books down, when it was clear that someone was attempting to confuse consumers and perpetuate fraud”, she adds.

Her books were listed not only on Amazon, but also on the author page on Goodreads, an Amazon-owned site that is a kind of social network dedicated to books and reading enthusiasts.

“So, it didn’t match the sort of writing I typically do, it didn’t match my voice or style. I don’t think that anyone took any care to make it look or sound like me. It was just creating content that could feasibly be associated with me, and then they put my name on it”, she also explains in the interview for Panorama.

Jane Friedman says she doesn’t know who used her name for the AI-generated books because Amazon didn’t give her that information and she would have had to file a lawsuit to get it, which would have been costly. The author limited herself to official channels to complain to Amazon, on which occasion she found that the platform has no procedure in such situations.

“They don’t have takedown requests that relate to this sort of fraud. They have copyright infringement and trademark infringement claims that you can make, but what happened to me is neither a copyright violation, nor a trademark violation. So, even though they did ultimately take it down, it was dispite of that, it wasn’t because of it”, Friedman says.

She also explains that the platform probably took down the books more likely as a result of the pressure and the fact that the issue had gotten into the media, which was hurting Amazon’s image, but says she received no further explanation from the company.

The future of cultural industries is uncertain in the AI era

As technology develops, AI systems will become more and more “chiseled” and capable of producing content that almost perfectly mimics human-created content. That worries The Authors Guild, an American organization with a century-long tradition of defending the rights of writers and authors.

Umair Kazi, director of advocacy at The Authors Guild, says one of the biggest concerns about Artificial Intelligence and its impact on content authors is the dilution of the content market.

“Generative AI can produce books and long-form works. They’re not always perfect, but they’re going to get better as Generative AI gets better. But what we’re concerned about is that those AI-generated works that are very quickly be able to produce will enter the market and they will displace works written by our members, human authors, in the sense that they’re quicker and cheaper to produce. A writer might take like a year to write a well-researched nonfiction book. But, with Generative AI, you could write a book”, Kazi says for Panorama, in a conversation via Google Meet.

He says the organization works with entities like Amazon, so he’s seen malicious users use AI-generated content on the platform. The main problem in this situation, Kazi points out, is that regular users don’t always realize that they’ve been tricked with this content. And even if the quality of AI-generated work is now not comparable to that of work generated by real writers, this could change in the future.

Another problem is that writers’ databases and content are being used to train AI models, especially since this kind of content is high quality for LLM systems, Kazi details. However, similar to news articles, writers often did not give their consent for their works to be used to train AI models.

“The books were just taken. They were not paid anything. And the result is you have a technology that can then basically kill the entire profession of writing, the craft of writing”, Kazi also says.

AI systems can be useful to authors if used correctly, and Kazi notes that he knows of publishers already using them to automatically write marketing copy.

Regulation and honest collaboration are the way to balance

Similar to what happens in the literary industry, the media industry also needs protection against abuses committed with the help of Artificial Intelligence. That’s why Courtney Radsch has created a framework for protecting and compensating journalists and media organizations for the use of their content by AI systems.

“The framework basically looks at the inputs into AI systems, the processing, and the outputs. And so, the framework looks at where is value created in journalism with respect to different parts of the AI system”, she says.

“It looks at archives, at data that is created in real time, and how those can be useful for different parts of AI systems, for example, developing and training Large Language Models or other foundation models relies on access in part to quality information”, Radsch continues.

As technology evolves, the expert says, quality content will become more and more valuable, and media organizations should assess the value of the content archive they have, as well as the content they currently produce.

Another part of the framework developed by Radsch concerns the “hallucinations” that AI systems can produce. In order to avoid this risk, AI systems need access to real-time or up-to-date information to allow fine-tuning of LLM models. This benefit of access to high-quality content can be provided by media organizations to technology companies.

“But then you also want to look at how do we establish the value. Is this about a one-time licensing fee? Is this about royalties? How can they think about the value, for example, of using these tools in their newsrooms?”, Radsch asks, in the conversation with Panorama.

What media organizations should do, she believes, is assess not only what benefits AI systems bring to newsrooms, but also what the newsrooms are offering in training these systems and how that value can be monetized, through the creation of licenses or usage rights.

Two other essential elements included by Radsch in the framework are regulatory and environmental protection. Clear policies and rules related to intellectual property rights and copyright, as well as intermediary liability, which does not currently apply to AI companies, would help avoid monopolies, abusive situations and environmental impacts.

For now, the European Union is at the forefront of regulating Artificial Intelligence. On August 1, 2024, the European Artificial Intelligence Act (AI Act) came into force, which aims to “encourage the responsible development and deployment of Artificial Intelligence in the European Union”, says the European Commission’s press release.

Essentially, the AI Act gives AI developers and implementers clear rules and obligations on how AI technology can be used. The set of rules also provides a way to assess AI systems based on criteria related to the risk to the public (from the lowest, such as AI systems that send spam messages, to the highest, such as AI systems that stand in the way of human rights by abusive use by various entities).

How to make peace with AI

In terms of the future of the relationship between booming AI and content creators, Peter Erdelyi believes there is a policy problem in general. Over-regulation, he says, will get in the way of finding an innovative solution, and the friction between the big AI companies and the media industry is moving far too fast.

Although no one knows exactly how the situation will evolve, Erdelyi thinks it’s important to create policies so that content creators and publishers can be rewarded and collaboration can be more constructive.

Courtney Radsch says AI will continue to be integrated into newsrooms through partnerships between tech companies and publishers. “I worry about these partnerships and these deals, because I think it’s once again creating a dependency handful of big tech platforms”, she explains.

One solution could be to work with smaller companies, start-ups or companies that honestly pay for licenses.

“I think they (note: the newsrooms) need to really think about the choices they make about the products that they integrate and how they reflect the values of public interest, and do not further their dependency on a handful of platforms, the way that their choices during the social media age did”, Radsch adds.

The future of content creators and media entities, Angie Holan believes, lies in their ability to secure sustainable financial models, especially as the quality of online content generally declines.