How to Declare Your Content’s Origin in the Age of AI
More than a year into the modern AI Gold Rush, grim predictions about the end of creative content jobs happen every day. The IMF has reported that AI is bad news for workers in general, as many jobs are too easily replaced by AI agents. While the strikes by both the WGA and SAG/AFTRA pushed back on AI for jobs, predictions of 200,000 Hollywood jobs lost to AI are still floating around. Some of these are just safe strategic decisions by the movers and shakers in the industry. Recently, movie mogul Tyler Perry canceled an $800 million dollar expansion of his Atlanta studio after he saw the initial output from OpenAI’s text-to-video app Sora. How many millions of untold stories and jobs will be lost due to AI being used as a solution for content creation instead of the tool for talented creators to use in their creative process?
The backlash is already here, too. Last year, Marvel’s worst television show to date, Secret Invasion, opened with an AI-generated title sequence that drew the ire of fans, film industry members, and critics alike. Despite a ludicrous response that the theme of the poorly-received show somehow justified use of the technology, no one was buying it. People want to know if content is created by human beings or by machines because it might affect their willingness to buy or consume it.
Did You “AI” This?
This is why we need to figure out how to classify content and make sure those labels are applied to creative work across the creative world. Modern consumers are far more likely to make an informed decision about what to buy than in the past, if only for the fact that this information is much easier to obtain and even automate into your buying cycles with the advent of modern technology. While Boomers and Gen-X folks took what they got and liked it in the past, Millennials and Gen-Z consumers are more likely to hold the parties that produce products responsible for their actions. This is, I might need to point out, a good thing.
Much like buying organic food, I believe content consumers are soon going to be looking for word on whether the article they read, the art exhibit they are going to view, the game they are going to play, or the song they are going to hear has been made by people or, as I often say, “by the robots.”
Yet, there’s a bit of nuance here, so I’d like to recommend how to classify Content Origin since we do just that at Credtent, our new AI-industry utility focused on ensuring properly-credited and credible content is used in AI training processes. As a technology researcher and content maven who has been involved with generative AI projects like LyricStudio for years before ChatGPT came out, I’ve been thinking about this for a long time.
Is it real or is it AI?
This section title is a riff on the old ad “Is it real or is it Memorex,” back when duplication of music was a concern for the music industry, but the technology companies doubled-down about how wonderful their reproduction was instead of music performed live. There was a conflict there and the tech won out, as it always does.
With generative AI, it’s a bit more of a battle though, because reproducing a song is quite different than using that song to make new songs. My cassette tape might record a song off the radio (dating myself here…) but hearing that could lead me to future purchases of the music from that artist, concerts, t-shirts, and generally becoming a fan. The magical duplication took on the role of marketing material for the song. With generative AI, we get none of that. An artist’s work is mashed up in a goulash served in what we call a ‘black box’ solution, meaning we have little to no insight into how these outcomes are generated.
If I used generative music AI to make a song I like, I can’t truly know how much of my new work came from the creative output of Tame Impala, Neutral Milk Hotel, or the Throwing Muses. I like the sound of that mashup, but if I use a generative AI music tool to build it, it’s hard to pick out the specific origin of the components of the work. Worse, I don’t know how to compensate these talented artists for their contributions. No AI company gives out credits — in fact, at least OpenAI’s CTO seems unwilling to discuss their sources when asked directly. Adobe, Bloom and some other LLMs say they are only using licensed and public domain (not ‘publicly available’) work. Do we know that for sure?
Even so, how do we talk about these people/AI hybrid creations?
Content Origins Defined
Plenty of startups are out there trying to break the code on determining the difference between AI-created content and human-created content…by using AI tools. That’s not what we’re doing here. Instead, we’re defining Content Origin so creators of all types can disclose what went into putting out their latest post, book, song, video, etc. in the age of AI. By setting these standards, Credtent’s hope is that consumers will have more information when they choose to buy, consume, and support the work of artists or AIrtists, if you will.
AI-Created Content (ACC)
This one is easy to define. ACC is simply content created by an AI that is marginally or not edited at all by a human creator. This content should always have a watermark from the AI tool used, and the creator should disclose its origin so that consumers can make their decision about supporting the content a creator has produced or not if they don’t want to experience ACC.
ACC may be factually correct, but without any kind of human connection to a creator, it tends to ring as hollow as a spreadsheet. In that way, we should consider ACC potentially useful if it is also produced by a credible source, but any artistic value will be from chance at best. Sure, the art created by the humans whose content helped train the AI could spark something in a reader. However, without intention from a creator who understands how to express an idea in a human way, the result is likely to be bland and lack human depth. Will things get much better one day? Yes, I think they will. Regardless of that, it is our belief at Credtent that the people whose work went into training the AI tool that was used to create this content should be compensated appropriately or allowed to exclude their work from this process entirely.
AI-Assisted Content (AAC)
In the middle of ACC and purely human-created work is AAC, which uses generative AI for building structure for the content, but it is rewritten, recomposed, edited, changed, and otherwise made the creator’s own through their hard work. In this case, the original song, poem, story, artwork, or other creative piece that was generated by the AI is unrecognizable in the edited form.
Thus, an AI did assist in composition of the work, but mostly for basic scaffolding, organization, or brainstorming. The final product is the creator’s own. That last point is right on the line, because genuinely human-created content could be inspired by something AI generated so let’s talk about that in the next section. Even so, ideally we disclose whenever a generative AI tool like ChatGPT, Claude, Gene, Gemini, etc. has been used to create some part of the work. Maybe you edited a lot of it away, but if you used a generative AI tool to build your work, we believe it should be disclosed.
Human-Created Content (HCC)
While this seems easy to classify— particularly for anything created before the advent of modern computers — there’s a bit more to talk about for content creation in the 2020's. Keep in mind that AI tools have been autocorrecting (and sometimes ruining) our messages and documents long before ChatGPT popularized generative AI and changed the world.
So, how much AI can you use before you can no longer consider your content entirely HCC? Well, let’s be reasonable here. Autocorrect is fixing typos and, unless you are composing novels like William S Burroughs, it is not going to affect your creative output. I think you can call work that has been autocorrected HCC without too much worry.
Let’s push that a bit further. AutoCrit is a wonderful application I’ve used when reviewing my book-length work. This app will score your content for clarity, readability, and a variety of other parameters to help you improve your work. AutoCrit will tell you when you overuse a word, switch tenses when you may not have meant to do so, and other useful things that a good line-editor would also do. In technology, we like to say that AI is supposed to be for the three D’s, right — the Dangerous, the Dirty, and the Dull. Line-editing is the Dull in a nutshell — it’s more about format, not creativity or aesthetics (yes, there are exceptions). But most people using AutoCrit are not using it to fix creative problems — they are just trying improve the writing for clarity or readability. I call this Insightful AI because they do not make direct changes to your work. Instead, they offer insight to the creator who can choose to make a change…or not. Maybe I wanted to repeat the word “Quixotic” seventeen times on one page to make some kind of point. Autocrit warns me that it’s not an ideal editorial choice, but it doesn’t suggest something new.
This isn’t just about prose and poetry. You might use Photoshop to touch up something without using its new super-powerful generative AI features. You just might have just wanted to brush out a few lines of dust on a picture or perhaps some photobomber in the background. Or a musician remixing a track might want to bring up an immortal bass line. These are the tools of the trade and they aren’t about adding aesthetic value that is false or from someone else. They are used by creators to fix an imperfection they perceive or to more accurately express themselves with an editorial change. This goes beyond Insightful AI into Corrective AI. Once again, cleaning up an image is not the same as telling Photoshop to add apples to the grass and diamonds in the sky of the image you created.
Thus, HCC can include works made with these traditional tools that would not be considered generative AI. These tools are used to modify the creation humans have made based on minor corrections or insight, so their use should not force content into the realms of AAC that admits direct generative tools being used.
What about Generative AI for ‘Sketching’ and Inspiration?
Can generative AI ever be used for Human-Created Content (HCC)? Sure, but not in the final work. When I started talking to more creatives about their use of generative AI, I spoke to my friend, the extraordinarily talented artist Vivien Mildenberger, who has developed a gorgeous style of her own that has graced everything from the covers of hit children’s books to articles in the New Yorker. Vivien noted to me that she likes using generative AI products like Midjourney to sketch out her ideas. She could explore concepts quickly to see how they might look, consider the perspective, and other factors that she’d normally sketch out on her iPad.
For her, gen-AI is just another tool for thinking about the work she wants to create. It’s using gen-AI in a way that is not about creating an aesthetic or reproducing the work of others whose work was part of, probably unwillingly, the training data. No, she’s just using technology to get sketches done quicker.
Similarly, more than a few writers I spoke with noted that they would use gen-AI to explore ideas, asking the tools to suggest concepts for, in one case, reasons why a character might be discharged from the military. Her story was not about the discharge, it was about the character and what happens to him when he has to go home and explain to his family why he is home. Is this much different from a Google search for the same information? She didn’t ask the gen-AI product to write her story’s plot. She did research.
For these uses, where the product of the gen-AI’s prompt does not appear in the work, gen-AI can still be used for HCC by our reckoning at Credtent.
What are you doing with this information?
Credtent.org has designated these three types of content for a simple reason: To enable creators to clearly label their work so people know what they are buying and consuming. Credtent is a platform that helps artists and creatives register their content so they can decide if they are willing to license it for AI training or not.
By registering your work with Credtent, a Public Benefit Corporation, you can confirm the type of content you have created and choose to exclude it from AI training or make passive income by getting annual licensing fees from the AI companies that use your work.
Use our Content Origin Badges to disclose how your content is created so people know what they’re getting. While Credtent encourages creators to stick to a specific type of content and to display the badge on their website or social profiles, creators who use a variety of techniques can use Content Origin badges at the individual content level as well.
For example, perhaps a songwriter writes most of her music on her own, but she used LyricStudio’s brilliant songwriting tech to help her complete a song for commercial release. She should disclose that the song was AAC since generative AI was used in its composition even if she changed is significantly to make it her own. Another song that she wrote herself but used some mixing apps to complete would be HCC because the tools used didn’t make creative decisions for her.
Content Origin is for creators to disclose and we hope to encourage it as best practice going forward. While our visual badges are the best way to make sure people know what they are getting, we also believe these labels should go into the metadata of each content piece.
What about AI stealing all that Content for Training?
Content creators have been getting a raw deal from Big Tech for a very long time. First, Google scraped the web to bring us search and made hundreds of billions on ads to get to our content. Then, Facebook connected us and claimed our content as their own, forced traditional media to pay to be on their platform, and now they making hundreds of billions on ads.
Now, Big AI is omnivorously consuming so much content that they’re running out. How much are they paying creators? In most cases, they pay nothing. When they do make an offer, it’s a pittance and a one-time payment.
When I worked at a content curation startup called PublishThis, we used to believe that Fair Use allowed us to reuse content as long as we sent eyeballs to the original work. But hiding behind Fair Use to steal the essence of someone’s work so you can produce new work is very different indeed.
It’s time for creators to rise up and demand subscription payment for their hard work making content
Letters and lawsuits are fine but Credtent is taking a more pragmatic approach because something needs to be done NOW.
Credtent has been built by creators, scientists, and technologists to orchestrate the process of ensuring a creator’s wishes are respected by AI.
Opt-Out Just Once
AI tools mostly expect creators to opt out of their training sets, as if everyone has time to contact tens of thousands of AI systems. Instead, register once with Credtent and we’ll get you excluded from AI tools across the industry. We’ll solve your AI problem in one step.
Did an AI tool already consume your work? We’ll audit them to make sure guardrails are in place to not allow someone to create work like yours.
Or Opt-In for Passive Income
You can also opt to license your work for subscription passive income. Credtent’s goal is to give YOU control of your work again so Big AI can stop stealing your content for their models.
Credtent, as a certified Public Benefit Corporation, is also committed to helping AI tools do a better and more respectful job of picking the content on which they train, including avoiding disinformation or harmful work reposted as misinformation.
Don’t delay because AI is moving at lightspeed and the sooner we start, the more AI tools we can prevent from stealing your work for their own purposes.
Join the Movement — Sign Up Now at Credtent.org
P.S. This article is HCC. I do like to use AI for structure sometimes when I’m writing, but I always rewrite the article entirely. Maybe AIs will eventually get good enough to write better than I do, but I don’t know if it’ll ever get to a place where I will prefer its writing to my own.
