As generative AI proliferates, it’s beginning to reach the ads we hear on podcasts and the radio.
Startup Adthos this week launched a platform that uses AI to generate scripts for audio ads — and even add voiceovers, sound effects and music. Customers can optionally go in and edit the ads using a browser-based, multi-track editor, yielding a “fully produced” ad that can be played out via integrations with radio automation systems and ad servers.
Adthos CEO Raoul Wedel pitches the platform as a way for advertisers to “efficiently self-service” and “instantly” produce high-audio ads — and for publishers to “support long-tail revenue streams” by creating self-service ad-buying platforms.
“The real game changer is that a small advertiser can run a great-quality ad in a minute or two,” Wedel said, pointing out the potential for localization and personalization.
But the platform — like many embracing generative AI tech — raises ethical questions.
Adthos’ platform is made up of a mix of in-house tech and third-party APIs. To generate ad scripts, Adthos leverages OpenAI’s recently released GPT-4 text-producing model. And on the voice side, Adthos compiled a library of hundreds of synthetic voices, including — according to Wedel — “Emmy Award-winning” voice actors and talents.
Wedel says that the voice actors are “fully aware” of the platform and paid either a flat fee or a licensing fee for the use of their voice, based on their preferences. “Adthos has already paid out hundreds of thousands in royalties to voice actors,” he added.
But what’s unclear is whether all of those voice actors chose to be included in Adthos’ library in the first place.
Recently, Vice reported that actors are being asked to sign away rights to their voices so that clients can use AI to generate synthetic versions that could eventually replace them. The contract language tends to be ambiguous, the voice actors say, with some actors reportedly being told that they can’t be hired without agreeing to clauses that give away the rights to use their voices for synthetic training.
Wedel didn’t reveal the contractual terms of Adthos’ arrangements. But he reaffirmed that voice actors on the platform have signed consent and licensing agreements.
“We use real-life voice actors for the training of synthetic voices. We train and select them based on our experience in markets and ad creation,” Wedel said. “We have created our own training datasets and recording software for the talent to record their voice.”
Wedel added that actors are afforded some control over how their voices are used on the platform, like the choice to preclude them from use in certain ad categories, such as religious or political ads.
Asked about what steps, if any, Adthos takes to moderate the content generated by its platform, Wedel said that it leaves the task of reviewing ads before they’re finalized and broadcast to customers and publishers.
Generative AI, particularly text-generating AI, has a tendency to go off the rails. For example, OpenAI’s AI-powered chatbot, ChatGPT, has been found to make up facts and reinforce ethnic and gender stereotypes. If an Adthos customer was to totally automate the ad generation process, they might run the risk of producing toxic, offensive and ultimately harmful ads — an undesirable outcome for any brand, surely.
Wedel notes that OpenAI performs some filtering at the API level. But — recognizing that this alone isn’t sufficient, perhaps — he says that Adthos plans to add content moderation “as the platform develops,” including rules about words or terms that can’t be used in ads.
“It’s up to the publisher to have a human in the loop,” Wedel said. “But we have specific requirements, like disclosing the AI nature of the ads and only using voiceover talent with their consent.”
In any case, Adthos claims to have several paying customers, including Dutch media conglomerate Talpa, and advertisers and agencies like GroupM, Dentsu and Sportradar.
I asked Wedel point blank if he was concerned that Adthos’ platform could cost marketing and advertising agencies and companies jobs. He quoted Sam Altman, the CEO of OpenAI, who said it was a “conscious decision” to start conversations in society early about what’s coming where it concerns AI.
“It will undoubtedly cause jobs to be lost. But the people that prepare and embrace AI will thrive and be more productive,” Wedel said. “The ones that don’t will be the ones losing jobs.”
Adthos uses generative AI to completely automate audio ads by Kyle Wiggers originally published on TechCrunch