“We never go out of style…”
I had a great chat with Kaya Yurieff over at The Information yesterday about the dispute between OpenAI and Scarlett Johansson.
Quick Background
OpenAI reportedly tried to negotiate with Johansson to license her voice to train and power an AI-assistant, similar to the character Johansson played in the movie Her. She declined.
Fast-forward to earlier this month and OpenAI launches an AI-assistant tool that sounds surprisingly similar to Johansson’s voice. They pulled the service and Johansson issued a statement about the matter.
Here’s a quick breakdown of the legal issues with Scarjo’s open letter to OpenAI and the questions raised around voice vs. style.
Two Key Issues:
- Is a generic style of voice (e.g., a soothing female computer) protectable under the right of publicity, as opposed to a unique and distinctive voice? (key terms: style vs. voice)
- Does this matter if the underlying materials used to train a model directly or heavily rely on the actual voice of the person that is being artificially recreated? What if it’s just instructions given to another voice actor?
Legal Background
- Voice actors develop a unique style, including rhythm, pace, inflections, and accents, which becomes their industry signature.
- This style, while not copyrightable, is possibly protected under the Right of Publicity.
- Unauthorized imitation of a voice actor’s style could infringe upon their right of publicity, especially if used without consent in a commercial context (including advertisements or product promotion).
Right of publicity is a distinct right (usually based in state privacy law) different than copyright.
Copyright would cover the artistic expression embodied in a specific painting. As you know, “style” has been a big focus in the copyright space for gen AI tools that allow prompts to utilize the name of a particular artist to produce an output in a similar style.
For rights of publicity (or name, image, and likeness) rights to apply or get triggered, there usually needs to be a violation—typically a direct or sometimes indirect commercialization—of a well-known individual’s name, image/likeness, or voice.
What Next?
In her open letter, Johansson is arguing that voice actors, like other performers, have a vested interest in protecting their unique style under federal law.
OpenAI might argue, as it seems to be doing, that it is pursuing a generic style of voice, which was created separately from Johansson’s acting in the movie Her.
Will these actions by OpenAI lead to more regulatory scrutiny? Or, will it call into question the ability to use gen AI tools in a way that mimics well-known vocal styles?
What do you think? 🤔
3 responses to “OpenAI vs. Scarlett Johansson: The Voice Imitation Controversy Explained”
[…] OpenAI vs. Scarlett Johansson: The Voice Imitation Controversy Explained […]
It may be that their big mistake was in negotiating with her first…
But ultimately, this sounds very close, factually, to the Midler v. Ford and Waitts v Frito Lay cases , except that the imitator isn’t a sound-alike vocalist, it’s an AI system. (and ChatGPT isn’t an Advertisement)
I’d also ask whether a Lanham Act §43(a) claim might apply here, if ‘consumer’s were confused/misled that it’s actually Ms. Johansen… Maybe we don’t need a federal Right of Publicity law.
The constitutional issues are significant. Is the voice of the AI just talking to users a commercial use that can even be regulated? Would such regulation even survive intermediate scrutiny?
Thanks for the writeup, Franklin!
Right! The Bette Midler and Tom Waits cases offer a great historical perspective on this, despite the differences you point out. AI is giving all of us so much to think about and process as developments continue to roll out.