ChatGPT
“Sky” sounds too similar: Scarlett Johansson takes action against AI voice
Scarlett Johansson made her voice available to an AI years ago – in the film “Her”
© Imago Images
A sensational live conversation with ChatGPT causes a dispute between the AI company OpenAI and Hollywood great Scarlett Johansson. The actress thinks the artificial voice sounds too much like hers.
Does ChatGPT speak like Scarlett Johansson?
The voice of ChatGPT, which was recently heard at a demonstration, was so “uncanny similar” to hers that even her closest friends and journalists could not tell the difference, said the actress. OpenAI counters: The basis for the AI voice were sentences spoken by another actress – and she was never meant to sound like Johansson.
The dispute, which has now become public, explains why OpenAI has now announced that it will temporarily phase out the voice called “Sky”. Johansson described that Altman approached her in September last year with the offer to use her characteristic voice as the basis for ChatGPT’s AI voice. After careful consideration, she declined for personal reasons.
The offer has a history: a decade ago in the film “Her,” Johansson took on the speaking role of an AI software with which the protagonist played by Joaquin Phoenix falls in love. A few days ago, a ChatGPT demonstration caused a stir because the AI was able to hold a conversation. Altman then wrote on the online platform X that the software was like AI from movies – and added the word “here” to his message.
“Protection of the image, work, identity”
Johansson subsequently responded by saying she was “shocked and angry” when she heard the AI voice from the OpenAI demonstration. Clarification of the circumstances is important. Especially in times when society is confronted with deceptively real deepfakes, transparency and effective laws are needed to protect the personal rights of individuals. It’s “about protecting our image, our own work, our own identities.”
“Sky” was introduced to ChatGPT back in September 2023 along with four other voices, “Breeze,” “Cove,” “Ember,” and “Juniper.” Only with the new AI model GPT-4o should the chatbot be able to have a truly flowing conversation. The five voices are based on sentences that people have spoken for OpenAI. They were chosen from more than 400 applications from film and stage actors, the company said in a blog post. OpenAI does not give their names “to protect their privacy,” as it explains.