In the Disney fairy tale “The Little Mermaid,” the main character Ariel is tricked into losing her voice by the sea witch Ursula. Only when her voice is separated from her body does Ariel realize how important it is to her persona.
This week, Hans Christian Andersen’s classic looks like a parable for our times, as movie star Scarlett Johansson opens up about who owns her voice and what she’d do if someone imitated hers for commercial gain question.
When Johansson declined a request to voice OpenAI’s latest generative artificial intelligence tool, ChatGPT-4o, company CEO Sam Altman offered up another actor who sounded very similar to her. Altman denies trying to make GPT-4o sound like Johansson. But his denial rang hollow, because on the same day OpenAI launched GPT-4o, he tweeted – “Her” – the title of the 2013 sci-fi movie in which Johnson stars as a woman Voiced by artificial intelligence assistant Samantha.
Whatever the case may be, Johnson’s situation is a warning to all of us. No matter how much we value our voices, artificial intelligence has created a world where that ownership feels threatened. Whether we express our voices freely or have them stolen, artificial intelligence can make us sound like we are saying things we never said.
audio box and me
I got a little taste of it myself. This week at Vivatech in Paris, I tried out Audiobox, Meta’s generative AI speech tool, which the company first launched last summer. The tool works by listening to a recording of your voice and synthesizing it so that it can be used to read text aloud, as if you were speaking the words yourself. You can try it here.
I used my iPad to record a few seconds of my voice, then typed a phrase in the hope that the tool would read it aloud to me. In less than a minute, it read this sentence to me in my voice. It was an incredible experience to hear my voice coming out of the iPad, reading a sentence I’d never actually said out loud.
Meta’s artificial intelligence tools are just one of many.
Meta’s privacy policy means that your AI Audiobox voice is yours to use, but a number of companies are emerging, including ElevenLabs and Speechify, capable of creating AI versions of your voice. I freely gave my voice for this experiment, but it could just as easily have been recorded from radio, television, YouTube videos, or podcasts and made into an AI without my knowledge or permission Copy. People whose voices are in the public domain (such as celebrities) are the easiest targets, but from voice notes to phone recorders, no one is immune.
Remember the election robocalls that mimicked President Joe Biden’s voice?
In the age of artificial intelligence, we have no choice but to accept that our voices will never truly be immune to deception. Just as we are increasingly wary of AI-generated images, we must also accept the reality that not everything we hear can be believed.
When nothing can be taken at face value, trust is more important than ever. The people who create our technology need to show us that they can be trusted by the things they are designed for (us) to do the right thing. Companies like Meta and Google are now familiar with people’s ability to publicly scrutinize their trust and safety policies and enforce those policies. Even if we still feel uneasy, most of us understand these efforts and know where we stand.
OpenAI needs to build trust, not break it
New companies like OpenAI, without leaving behind a legacy of trust and security, still need to earn their stripes. This week, Ultraman’s company hit its first hurdle.
OpenAI may not have trained ChatGPT-4o on Johnson’s voice, but by finding a way to imitate it regardless of the actor’s wishes, the company made it clear that it considers people’s vocal similarities to be fair game. Given that she refused to lend her voice to OpenAI, the company made its tools sound like her, in any case signaling that it was not interested in respecting her refusal to consent.
Altman rejected that notion, saying in a statement that OpenAI cast the actress to voice the chatbot Sky before contacting Johnson. “Out of respect for Mr. Johnson, we have suspended the use of Sky voices in our products,” he added. “We are sorry to Ms. Johnson that we did not communicate better.” OpenAI did not immediately respond to a request for further comment.
Consent is also at the heart of copyright lawsuits filed against OpenAI and Microsoft over text used as training material for their large-scale language models.
Dario Amodei, CEO of artificial intelligence company Anthropic, said at VivaTech that artificial intelligence has made the issue of copyright and ownership of intellectual property “a little blurry.” Amodei, who previously worked at OpenAI and later founded Anthropic, the maker of the Claude chatbot, has been critical of his former employer. It is precisely because of the complexity of these problems that humans have so far stuck to text rather than introducing “other modalities,” Amodei said.
Amodei said that as artificial intelligence becomes more intelligent and capable, we as a society need to work hard to recognize that artificial intelligence will infringe on human capabilities and sometimes be disturbing.
I think we’re already there.
Like other technologies before it, artificial intelligence entered our lives before guardrails were in place. Governments are now scrambling to correct the problem and provide tech companies with a rulebook to follow. For example, earlier this year, the Federal Communications Commission banned the use of artificial intelligence to copy voices in automated calls.
As this debate unfolds, we may not be able to completely lock down our own voices, but we can learn how to use them in the brave new world of artificial intelligence by remembering the voices we hear on television, radios, phones, PCs, and other devices. Protect yourself from the world.
Editor’s note: CNET used an artificial intelligence engine to help create dozens of stories and tagged them accordingly. The annotation you are reading accompanies an article that deals with the topic of artificial intelligence in substance, but is written entirely by our expert editors and writers. For more information, please see our artificial intelligence policy.
