AI, IP, vocal likeness, and Scarlett Johannson

Basic question up front: To what extent should/does* someone own the likeness of their voice?

Background: Scarlett Johannson and OpenAI appear on a litigation collision course over the similarity of the latter’s “Sky” AI chatbot voice to the former’s voice as used in “Her,” a film in which Scarlett voiced … wait for it … an AI chatbot.

Links below, but the basic facts (which are in dispute) are that OpenAI developed its chatbot, found an actress who sounds similar to Scarlett, coached her and the Chatbot to sound even more like Scarlett, contacted Scarlett to attempt to license her voice, failed to reach agreement, then deployed the Chatbot with a Scarlett/“Her”-esque voice nonetheless.** Along the way, the OpenAI CEO Ex-Tweeted a reference to “Her” as well.

Very interested in perspectives on this. As a lawyer with limited IP practice, I’m currently on the side that OpenAI violated Scarlett’s intellectual property but open to alternative viewpoints given the novelty of the issue.

*Emphasizing I’m interested in opinions based on legal precedent and what the law ought to be.

**The sequence of events is disputed by the parties, at least in media reporting.

EDIT: Here is OpenAI reading Scarlett’s statement on the matter should you want to hear the vocal similarity.

Link: www.youtube.com/watch?v=jsoUBkYxXpw

So, name, image, and likeness (NIP) are kind of a vague area of intellectual property law under what is termed personality rights or “right of publicity”. You cannot copyright a given name, or trademark an image or likeness of an actual person, but of course you can copyright the name of a fictional personage, and trademark created unique images and likenesses even if they are produced in the obvious form of a real person. The United States federal code says nothing about rights of publicity so it defaults to state jurisdictions and common law in terms of what is permitted.

It is generally understood that journalistic and media agents may freely use images and names in the guise of news (which includes paparazzi, celebrity tours, and parody) but not to use the image or reputation to sell other products or services, or in a way that unduly harms the reputation of the individual or suggests an unwarranted endorsement. What poses a ‘likeness’ can be pretty vague; if I make a champaign label and call it ‘Frankie Blue Eyes’, with an image of a sharp-faced smiling guy in a hat, a jury will likely recognize it as an obvious facsimile of Frank Sinatra and (correctly) infer that this is an effort to sell my product using the reputation or implied endorsement of the famous singer. But if I leave off the image, or just call it ‘Blue Eyes’, or otherwise don’t make the reference obvious, it is a much more difficult case to make.

‘Vocal likeness’ seems like a far stretch to make in general; there is no such thing in law as an objectively unique vocal signature (whether voices are truly uniquely identifiable is somewhat debatable), but not only does Ms. Johansson have a voice that is immediately recognizable by much of the public but OpenAI is also using it in a context that is clearly making a homage to her role in the movie Her (a film about an AI agent embedded in a smartphone with a sexy voice), and furthermore reportedly contacted Ms. Johansson repeatedly in an effort to get her to provide her vocal talents to their program.

Assuming they didn’t actually sample her vocal patterns and just found other voice actors who could provide a similar sounding voice, I don’t think she would have a case for being exploited or having work taken from her, but depending on how it is used she might be able to make an argument that this has diminished her ability to get new work or her reputation as a unique vocal talent, although she would have to demonstrate some material loss. Apparently OpenAI turned some dials and changed the voice enough that it clearly isn’t Scarlett Johansson (or an obvious imitation of the role), so at least their lawyers were concerned enough to push for that, but I think it would be an uphill battle to argue for actual damages to earnings or reputation.

Stranger

We have an informal maxim in the legal community that bad facts make bad law. Here, the (albeit disputed) facts certainly appear to establish that OpenAI intended to (mis)appropriate Scarlett Johannson’s vocal likeness. And a strong argument could be made in court that she has damages to her earnings, especially if OpenAI turns a large profit, on the basis that the familiarity and brand of her vocal likeness made the product more desirable to consumers.

That said, I continue to struggle with the broader question: to what extent does someone own their vocal likeness? Set aside the admittedly bad facts (for OpenAI) in this case as needed. But many artists have distinctive voices. As a hypothetical, imagine if a science documentary used AI-generated voices indistinguishable from William Shatner or Leonard Nimoy in a commercial film. There is a premium associated with their voices in that genre, yes?

I remember some radio ads (I can’t remember the product) that had “celebrities” endorsing the product. But the ad made a point to say these were celebrity impersonators and not the actual celebrities (who were not named).

So it appears this company and its ad agency recognized that a vocal likeness is legally protected, at least in theory.

On the other hand, it also argues that OpenAI might be able to follow a similar strategy. If it makes sure to explicitly state that its Chatbot is voiced by an impersonator, it might be able to defend itself in court.

In the late 80s, Ford used a voice similar to Bette Midler in advertisements after she declined. Midler supposedly won a sizeable settlement and set some new legal standards regarding use of voice. But I Am No Lawyer.

I think that is an open question. It is certainly permissible to imitate another person’s voice for the purpose of parody, even if that is for profit. I’m not aware of any specifically law regarding the appropriation of ‘vocal likeness’ on its own (although adopting other identifiable characteristics of speech and manner have precedent) although doubtless there is something but not enough to provide protection for musicians, voice actors, and public speakers whose speech can now be indistinguishably imitated by generative AI.

But it is an even broader question than voice; as generative AI improves (and despite my criticisms of how the purported reasoning abilities and essential factuality are being exaggerated, it will) it will be able to accurately reproduce the vocal manner, speech patterns, facial likeness, and other characteristics to become essentially indistinguishable from the person being imitated, and filtering out ‘deepfakes’ will become enormously challenging if not impossible. How do we handle the legal situations when everybody has the ability to control a simulacrum with the same computing power as a digital editing station or a game console, and not just from the standpoint of repetitional damage or loss of revenue, but when we literally can’t distinguish fact from fiction?

Here is the commercial:

Honestly, it doesn’t sound particularly identifiable as Bette Midler to me so I’m surprised she managed to get a settlement, but it may have just been a desire for a wealthy automaker to avoid bad publicity.

Stranger

Established caselaw is quite clear. In addition to the Better Midler story already named, there was an example with Tom Waits, where he declined and they went ahead with an imitator and he sued and won.

If Johansson can establish a paper trail showing intent to infringe on her identity, the suit is a slam dunk.

Sounds like a real slippery slope. As long as it was not actually the real voice of Johansson herself, I would argue there is no infringement. Otherwise, anyone who has a voice that resembles anyone else’s would be in trouble - what if someone’s voice just happens to sound like Obama’s?

As long as that person doesn’t actually imply that they are, in fact, Obama, there’s no problem, is there?

I’m even more interested in the ‘disclaimer’ question from upthread: forget imply — what happens if there’s explicit mention that, oh, hey, X isn’t Y, okay?

Well, I believe the huge „wink, wink, nudge, nudge“ makes a difference, as they’re clearly trying to evoke a familiar voice/character. In this case, it’s not as if Scarlett Johannson is just suing out of the blue.

Well, granted. But — speaking of “out of the blue” — what I was wondering about was this exact bit posted upthread: “What poses a ‘likeness’ can be pretty vague; if I make a champaign label and call it ‘Frankie Blue Eyes’, with an image of a sharp-faced smiling guy in a hat, a jury will likely recognize it as an obvious facsimile of Frank Sinatra and (correctly) infer that this is an effort to sell my product using the reputation or implied endorsement of the famous singer.”

But what if there’s no implied endorsement, and no chance to infer that it’s actually Johansson, because they flatly state that it’s not Johansson? Sure, there’s a wink and a nudge and a wink and a nudge, but what if they slowly and patiently explain that it’s not her voice you’re hearing and she hasn’t given an endorsement — even adding that they offered her money, and she said ‘no’ because she wanted nothing to do with this — and then, having made all of that absolutely clear, then: what?

FWIW, I had my ChatGPT set to voice as Sky and I didn’t even realize that was supposed to be Scarlett. Then again, I’m not always the best at recognizing voices, and I had never seen “Her.” I checked today and Sky is no longer an option. Oddly, I didn’t even notice when they changed her voice on me to Juniper.

Then, I would guess, it becomes a question of intent. If the likeness to a famous person is truly completely accidental, then your internal communication (emails, texts, memos) as well as press releases or commercials, etc. should reflect that. That’s what courts are for, to settle difficult edge cases that are not clearly defined. What happens in either case - well, I suppose we’ll see.

But what if they don’t make that claim about intent? What if they don’t argue it was accidental? Oh, there’s nothing accidental about the resemblance, they say at each step. That’s the reason for the explicit disclaimer, they add, every time.

Say they make clear — both in public statements and internal communications — that, because they don’t want anyone inferring that it’s her or that it was done with her endorsement, they’re putting that explicit disclaimer out there rather than risk implying the opposite. What follows?

I forgot about the Tom Waits suit, but I’ll point out that it wasn’t just based upon similarity of voice but because the imitation of Waits’ singing style (which the defendants admitted to) provided the perception that Waits was endorsing the defendant’s products, which formed the basis for harm to reputation and other damages. Since OpenAI isn’t explicitly using a voice similar to Ms. Johansson’s to sell itself or another product (although it is pretty obviously leveraging off of the appeal to similarity to her voice role in Her), a similar claim is a little more squishy, although it was obviously substantial enough to first ‘detune’ the voice and then remove it entirely from their interface.

The party in question can explicitly claim that it is not an imitation but if a judge or jury finds that there is sufficient similarity and evidence of intent to warrant an award damages for a plaintiff, they’re essentially deciding that the disclaimer is false. In the case of OpenAI, I think any reasonable person who is familiar with the film role and story would conclude that OpenAI is quite intentionally invoking a similarity between Ms. Johansson’s vocal role and their use of a breathy voice with characteristic vocal fry and laugh. Whether there are any actual damages to reputation, earnings, or otherwise is a separate question but OpenAI claiming that they aren’t trying to reference that film character is obviously bullshit.

BTW, I was wrong in my earlier statement that there are no laws specifically regarding appropriation of a vocal likeness; Tennessee passed a law earlier this year called the Ensuring Likeness Voice and Image Security (ELVIS) Act specifically to protect vocal artists from AI imitations and ‘deepfakes’. How well that will stand up to legal challenge (especially in music and television when musicians are portrayed by other actors) but there is now at least some statute law regarding the ‘cloning’ of vocal signatures.

Stranger

Maybe I should apologetically bow out of this thread instead of hijacking it, but I’m genuinely not getting this and it seems directly on point: what happens if the party in question admits that it’s an imitation, and explains that, look, we know that it’s so good an impression that plenty of people who hear it would falsely believe that it’s the celebrity’s voice unless we do a disclaimer — and, they add, that’s why we did the disclaimer.

You seem to be talking about a situation where the disclaimer is there to back up a “claim that it is not an imitation”. But I’m asking: what if the disclaimer is that it’s an imitation?

Right. And their motive for intentionally trying to reference that film character seems equally clear: to increase the appeal of their product, gain market share, win out over competitors, etc. They are attempting to profit financially from use of an imitation of Johansson’s voice.

Whether or not such use materially affects Johansson’s own career seems secondary. A real issue, but not the only issue by far. It seems reasonable for Johansson to ask: why should these people be able to make money this way?

No doubt hundreds of people in the public eye who have distinctive voices are watching this situation with great interest.

Well, Bette Midler denied permission to use her voice, then Mercury used the soundalike anyway, so it was an obvious attempt to get around that. And Midler didn’t “manage to get a settlement,” she won the case and then won the appeal.

As for whether the imitation sounds like the original, listen for yourself. Not just her voice but the entire arrangement of the music.

It is possible to trademark a given name. Harlan Ellison did so in 2001, and the Asimov and Heinlein estates did it in 2000 and 2011 respectively.

Trademark law is different from copyright law, to be sure. And Johnson appears to be threatening suit using likeness law, which is a third thing. There is currently no federal law on this; only a variety of state laws protect name, image, and likeness. The estate of George Carlin has already launched a suit because of an AI-imitation of his voice, claiming both infringement of copyright and violation of California’s NIL law.

These and many other interesting and pertinent facts are to be found here. Artificial Intelligence Prompts Renewed Consideration of a Federal Right of Publicity, by the Congressional Research Service.