Episode 18: Ethics and AI (A Creative Industry's Dilemma)

AI use is everywhere. But what does that mean exactly? Did someone use a tool to correct grammar? Did they have it look for redundancy in their writing or did they use it to create a cover, write AND narrate their book? 

In this episode, I break down the current AI categories to know in publishing and outline ways that narrators can not just protect their performances, but their moral code, too. 

Sources and show notes

1. The Atlantic, “Is This What Comes After AI Slop?” — reporting on the disputed Daggermouth allegations, the unpublished study, Wolfe’s denial, and Simon & Schuster’s response.
2. Pozotron, “Audiobook Proofing Service” and “Script Preparation” — the company’s descriptions of its AI-assisted proofing, preparation, and pickup tools.
3. Amazon Kindle Direct Publishing, “Content Guidelines: Artificial Intelligence Content” — definitions and disclosure requirements for AI-generated versus AI-assisted material.
4. U.S. Copyright Office, “Copyright and Artificial Intelligence” and “Registration Guidance for Works Containing AI-Generated Material” — human-authorship and disclosure guidance.
5. ACX, “Voice Replica Beta Help for Producers” — ACX’s description of voice replicas, narrator control, compensation, and product labeling.
6. SAG-AFTRA, “No Contract, No Work Order for Audiobooks” — union restrictions concerning audiobook digital replicas and AI training.
7. National Association of Voice Actors, “Synthetic Voice/AI Rider” — contractual resource for voice actors.
8. Authors Guild, “AI-Related Model Publishing Contract Clauses” — model provisions covering AI narration, artwork, translation, and other publishing uses.

Want to keep in touch? Me, too! 

Brenda Reads Audiobooks Linktree

and Check out my website: Brendascottwlazlo.com

For service inquiries, use the contact me form through my website. 

Transcript

SPEAKER_00 0:00

Now, if you have not been hearing about the controversy surrounding the book Daggermouth by H. M. Wolf, I'm not gonna say you're living under a rock. I'm gonna say you have a much healthier relationship with your social media than I do, because it has taken over all of my platforms. And it was the inspiration for this week's topic because I think it brings up a ton of additional conversations that need to be had in addition to just the one on is this AI generated or not? So I'm gonna start out telling you about it. The book started out as an independently published dystopian romance, became a massive viral hit, and was eventually picked up by Simon and Schuster in a reported seven-figure deal, Go HM Wolf. And now it's been placed at the center of a much bigger conversation about artificial intelligence in publishing because of a reporting in the Atlantic, where researchers were studying more than 14,000 self-published Kindle books and used an AI detection tool called Pangram and classified Daggermouth as approximately 60% AI-generated or AI-assisted when they ran it through. The researchers also claimed they found phrases that appeared in other books suspected of containing AI-generated text. Now, H. M. Wolfe, the author, categorically denies using generative AI to write the book. Her statement says, I wrote this book myself, and Simon and Schuster have said they will stand behind the book. And the study hasn't been peer-reviewed yet, and AI detectors aren't infallible. So I'm not bringing this up because I've decided whether HM Wolf used AI. I don't have access to her computer or drafts or her process. I'm bringing it up because the controversy is revealing a much bigger problem, and we've reached a point where creative workers can't treat AI like some hypothetical thing that like may enter our industry someday because it's already here. And it's coming at us in a lot of different ways. So I want to break down what those ways are. If you work in audiobooks and you encounter AI-generated or AI-assisted content, you can know about it before accepting the job. And there are already books appearing on ACX that seem to have been written and translated or illustrated by AI, and authors are using it for covers, marketing copy, brainstorming. Now, I know my stance, and I want you to have a stance on AI too, but I don't want to stand here and issue some kind of decree about what every narrator is or isn't allowed to do. I'm not the AI police. My own moral compass points toward art that is led by human beings and writers and performers and illustrators. I believe people should be paid for creative labor. Like I believe consent matters, and I don't want somebody taking my voice or my work and using it to build something that's designed to replace me. Those are my pillars, but you've got to identify yours. And to do that, I think we've got to stop treating AI use like it's this one giant, featureless category monolith, and start looking at what the technology is actually doing. And the first category is administrative assistance. This includes things like drafting routine emails, or organizing your schedule, or summarizing meeting notes, or sorting information in spreadsheets. You may actively choose it, or you may actually find out that it has been, like me, quietly added to the programs that I was already paying for and now crashes your computer because it takes up so much memory. But for some people, any use of generative AI is a hard no because of concerns about training data and environmental cost. That is a valid line. And other people decide that using it to organize their inbox and to send them highlighted to-do lists based on priorities is meaningfully different from asking it to write a novel or perform an audiobook. That can also be a line, but it should be a conscious one. The second category is production assistance. Now, this is where Positron is the example. Positron describes its software as AI-powered, and it can scan a manuscript for difficult words, help organize character information for narrators or editors or preppers or proofers. It compares the recorded narration of the text with the audio and flags missing or added words to help generate these pickup reports that we get. And the company says its purpose is to reduce repetitive production work so human performers and proofers can concentrate on accuracy, tone, and performance. Now that's different from generating the performance itself. But some narrators may still object to it. You may not want to upload an unpublished manuscript or your raw audio to an AI-powered platform. You may have questions about how that data is used and who has access to it and how it is used to train. And those are completely reasonable questions. It helps with proofing, quote unquote, doesn't mean we should stop reading the terms of service in these different programs. Now the third category is creative assistance. And this is where AI websites, particularly geared toward authors, are used to brainstorm plots or develop characters or rewrite passages. Sometimes they support dialogue, they generate marketing copy. Now, Amazon's own publishing guidelines distinguish between AI-assisted and AI generated content. If the author creates the content and uses AI to brainstorm and refine it or error check it, Amazon classifies that as AI assisted. If the tool produces the actual text, image, or translation, even if a person edits it afterward, Amazon calls it AI generated. And important to us, KDP requires disclosure to Amazon of AI-generated content, but not AI-assisted content. So maybe you're comfortable narrating a human-written book whose author used an AI grammar tool. Or maybe you're comfortable with AI-generated promotional copy, but not an AI-generated cover. Maybe the cover's your line because an illustrator, an artist could have been hired. You're allowed to draw the line differently from Amazon, but you gotta draw the line. The fourth category is complete content generation. We're talking a machine producing large portions of a book, an illustration, a translation, or a narration? This is where my own concerns get much sharper. It's raising an artistic question. What exactly are we being hired to perform? And is that performance led from human content or computer-generated content? And this is why it's important to know your own stance. If I accept a book because I believe I'm collaborating with another creative person and I learn that manuscript was substantially generated by a machine, I may feel like the nature of the job was misrepresented to me. That doesn't mean every narrator is going to make the same decision, but it does mean we should probably be talking about disclosure before that contract is signed, not after a controversy erupts. Now, the fifth and biggest controversial category is synthetic narration and voice replication. This isn't the same thing as proofing a human voice. This is the technology that's generating the performance itself or is using recordings of a performer to create a digital clone of their voice. ACX now describes a voice replica beta in which participating narrators can create computer-generated replicas of their voices. According to ACX, the narrator chooses which projects to audition for, reviews the final production, and receives payment or royalties. Audiobooks created this way are supposed to be labeled as narrated by the narrator's voice replica. Now, some narrators may view a consent-based replica as another service they can control. Other narrators, including me, may see synthetic narration as a direct threat to the human craft and labor of audiobook performance. Those positions aren't identical. Far from it, but the baseline should still be consent, control, disclosure, and compensation. Now, if you're screaming, what about the union? SAGAVDRA has issued a no-contract, no work order for its members regarding audiobook companies seeking to create or use digital voice replicas to train audiobook AI systems unless that company becomes a union signatory. Now the National Association of Voice Actors also offers an AI writer that's designed to address synthetic voice use in contracts. Whether or not you're in the union, there is a practical lesson there. Protect your voice in writing. Don't assume that a standard confidentiality clause or ownership clause protects you. Look for language about machine learning, model training, digital replicas, synthetic voices, derivative voices, reuse, sublicing, and technologies that may be developed in the future. Those are very specific terms, but those are the ones you need. You also need to ask whether your audition, your raw files, your pickups, your outtakes, or your finished performance can be used for anything beyond producing and marketing that specific audiobook. And that can be hard, right? You can be approached as a newer narrator by a company you've never worked for. They want you to audition, and you're just happy to be in the room, right? So you send off that audition, and then you never hear back. And then you find out later that they're gathering all those auditions and putting them into an AI tool. That's trans perfect. That's a scam that's currently going on right now. So if the answer is vague, get clarification before you record. Any valid company will be happy to answer questions and clarify for your own business protection? So, how do you create your own ground rules? I begin with these five questions. One, what part of the process did the AI perform? Did it organize a schedule? Did it check a manuscript? Did it rewrite anything or generate a cover? Did it create the actual book or create the narration? Second, was the material used to build the system obtained with consent? I may feel differently about a tool that's trained on licensed material than I do on one that was built by scraping creative work without permission. Third, is the use transparent? Does everyone involved know what was generated, what was assisted, and what's still human created? A small non-narrator example of this is HIPAA compliance requires one or two specific approved apps for doctors to have to record their patient interactions so that they can have notes of that later. They have to ask, in the state of Illinois, they have to ask consent by the patient every single time they want to use it. That's transparency. Fourth, who's being displaced or deprived of payment by using this? If the tool is helping a proofreader locate possible errors, is it being used so the proofreader or actor or writer or engineer doesn't have to be hired at all? Or are they using it as a tool to be more efficient? And fifth, what happens to my work? Is my voice being stored? Can it train a model? Can it be cloned? Can this company sell or sublicense it? Can I revoke consent? Will I be told about every new use and will I be paid? Your answers can become a short personal policy. I'll give you an example. Maybe yours says, I'll use AI-assisted administrative tools, but I won't use generative tools for creative work. I'll narrate AI-assisted manuscripts if the author discloses the use, but I won't narrate substantially AI-generated books. I'll use automated proofing tools only when the manuscript and audio aren't retained for training, and I never consent to voice cloning. You can put that on your website. Or you can also have your policy simply be no generative AI anywhere ever in my business. Whatever your line is, write it down, revisit it as the technology changes, put protections into your contracts and ask questions early. And recognize that you may occasionally learn something after the fact that you couldn't have reasonably known. So maybe we should also have some grace for the people around us that might have that happen to them as well. Encountering undisclosed AI doesn't make you complicit in something you were never told about. What matters is what you do once you have the information. Now, if you know me at all, I am adamantly opposed to replacing human narrators with virtual voices. I'm disappointed when authors choose synthetic narration, and I'm, you know, a little heartbroken by it, but at this moment, it is still their prerogative. I also believe it can cost them trust and connection in the community and the opportunity to create something richer with another artist, but I can't build my career entirely around policing other people. My time is better spent protecting my own work and supporting human creators and becoming exceptionally good at the things a generated voice can't bring into the booth, like instinct and experience and collaboration and risk and humor and timing and oh my god, breath, and all those strange, beautiful choices that happen when a human being meets a story. AI is going to keep appearing in our tools and our contracts, our inboxes, our auditions, and some of the books that are sent to us. We don't have to pretend it isn't happening, and we don't have to surrender our standards just because avoiding it has gotten complicated. We can look directly at that complication in the eyes. We can decide where our line is and we can put our energy into building the kind of creative world and the kind of creative career that we actually want to live in. I hope this resonates with you in some way. If it does, please comment, like, subscribe, continue to share this podcast. It is such a passion of mine. And in the meantime, take care of yourself. And I will talk to you in the next one.