Well, here are some things that aren't really being disputed:
* OpenAI wanted an AI voice that sounds like SJ
* SJ declined
* OpenAI got an AI voice that sounds like SJ anyway
I guess they want us to believe this happened without shenanigans, but it's bit hard to.
The headline of the article is a little funny, because records can't really show they weren't looking for an SJ sound-alike. They can just show that those records didn't mention it. The key decision-makers could simply have agreed to keep that fact close-to-the-vest -- they may have well understood that knocking off a high-profile actress was legally perilous.
Also, I think we can readily assume OpenAI understood that one of their potential voices sounded a lot like SJ. Since they were pursuing her they must have had a pretty good idea of what they were going after, especially considering the likely price tag. So even if an SJ voice wasn't the original goal, it clearly became an important goal to them. They surely listened to demos for many voice actors, auditioned a number of them, and may even have recorded many of them, but somehow they selected one for release who seemed to sound a lot like SJ.
Clearly an SJ voice was the goal, given that Altman asked her to do it, asked her a second time just two days before the ChatGPT-4o release, and then tweeted "her" on the release day. The next day Karpathy, recently ex-OpenAI, then tweets "The killer app of LLMs is Scarlett Johansson".
Altman appears to be an habitual liar. Note his recent claim not to be aware of the non-disparagement and claw-back terms he had departing employees agree to. Are we supposed to believe that the company lawyer or head of HR did this without consulting (or more likely being instructed by) the co-founder and CEO?!
They hired the actor that did the voice months before they contacted SJ. The reaction on this site to the news that this story was false is kind of mindbending.
> Are we supposed to believe that the company lawyer or head of HR did this without consulting (or more likely being instructed by) the co-founder and CEO?!
Yes this is pretty typical. The CEO doesn’t make all decisions. They hire people to make decisions. A company’s head of legal could definitely make decisions about what standard language to use in documents on their own.
It could have simply been the other way around: they auditioned some unknown voice actors, then someone noted that one of them sounded like Scarlett Johansson. They optimistically contacted SJ, assuming she would agree, but then had to back off.
Sky does not really sound like SJ though if you listen side by side. According to OAI's timeline, they intended to have Sky in addition to SJ. OAIs voice models including Sky predate the GPT4o voice assistant. Also:
"In a statement from the Sky actress provided by her agent, she wrote that at times the backlash “feels personal being that it’s just my natural voice and I’ve never been compared to her by the people who do know me closely.”"
It did not seem like an issue before and the Sky voice was public many months before GPT4o. I don't believe SJ can claim to own all young, attractive woman voices whether they are used as a voice assistant or not. It seems like the issue is being blown out of proportion. It does make a good story. The public perception of AI right now is generally negative and people are looking for reasons to disparage AI companies. Maybe there are good reasons sometimes, but this one is not it.
> It seems like the issue is being blown out of proportion.
It kinda feels like its on purpose. Someone in a previous thread mentioned that this might have been a cynical marketing ploy and I'm warming up to the theory. After they recorded the Sky VA, they figured out a whole marketing campaign with SJ to promote the voice feature. After she turned them down (twice), they released with just enough crumbs referencing the movie to goad SJ into committing a first degree Streisand.
With the slow roll out, everyone would have forgotten about the feature the day after the announcement but now it's been in the news for a week, constantly reminding everyone of what's coming up.
I'm also curious, legally speaking, is it an issue even if Sky's actress does sound like Scarlett? What if OpenAI admits they intentionally chose someone who sounded like Scarlett? Does it matter whether she was using her natural speaking voice or intentionally mimicking Scarlett's voice and mannerisms?
This seems similar to the latest season of Rick and Morty. Whether justified or not in that particular case, it rubs me the wrong way a bit in principle to think that a production can fire someone only to hire someone else to do a near-perfect copy of their likeness. If (as in the OpenAI case) they'd gone further and trained an AI on the impressions of Justin's voice, would that have been considered an AI impersonation of Justin with extra steps?
All of which is to say, this seems like a pretty interesting legal question to me, and potentially broader than just AI.
"SJ can't own all female AI voices" is attacking a straw man version of the complaint, which is much narrower. The question is whether OpenAI deliberately fostered the impression of an association between their product and her performance, which she had so far refused.
To your point, there have many female assistant voices on the market, including Sky -- but what might have tripped the line of impersonation was the context this particular one was presented and marketed. I don't know where exactly that line should be, but you can certainly reject this kind of marketing without stifling anybody's legitimate career.
Regardless of the moral implications, "sounds almost exactly the same" is not copyright infringement. Perhaps it could be trademark infringement if she had trademarked her voice like Harley-Davidson attempted (and failed) to trademark the sound of their motorcycles, but "sounds alike" is a pretty hard case to prove, and it's completely blown away if they can demonstrate that another human sounds indisputably similar.
People do celebrity impressions all the time, and that's not infringement either, because it's not actually copying that person's voice.
I'm sympathetic to SJ in this matter, especially after the Disney Black Widow debacle, but it sounds like she had the opportunity to write herself a nice check, and she turned it down.
On the basis of this article, it sounds like she doesn't have the cause of action that she had believed she had; I imagine that her legal team are now advising a fast settlement, but OpenAI's legal team might prefer to milk the free publicity for as long as they can, especially if they are fairly certain they would prevail at trial.
It isn't about copyright, it's about passing-off, it's described elsewhere in detail in these threads what it means. It's about intention and what the customer believes. If customers might believe its SJ, due to samas tweets, general likeness in voice, and the context (voice assistant), the public info about them trying to get SJ to do this - that's passing-off, even if it wasn't training on her voice per se. There are numerous law cases about this.
> it sounds like she had the opportunity to write herself a nice check, and she turned it down.
If I were SJ, I'd turn it down too. Shes in no need of money, and selling her voice to OpenAI would make most of creators and every single voice actor hate her (not to mention the Twitter mob).
In majority of creative circles, the current social norm is to hate AI, so touching AI in any way is too risky for reputation.
It probably is worth paying attention to the water WaPo is carrying for OpenAI here next to their publisher's announcement about prioritizing the use of AI in their newsrooms.
It doesn't seem like you'd need "shenanigans" for this. Lots of voice actors are capable of doing voices that sound like other people, and some even have a natural voice that happens to sound very similar to a particular more noteworthy celebrity. AFAIU, the rights to your likeness only apply to your likeness, not to the likeness of someone else who happens to look or sound a lot like you.
For a case that doesn't involve AI at all, consider situations where a voice actor in a cartoon is replaced (sometimes while still alive) by someone who can perform a voice that sounds the same. Decisively not illegal. Most people don't even find it immoral, as long as the reason for getting rid of the original voice actor wasn't wrong on its own (e.g. Roiland).
> For a case that doesn't involve AI at all, consider situations where a voice actor in a cartoon is replaced (sometimes while still alive) by someone who can perform a voice that sounds the same. Decisively not illegal.
Because there are contractual clauses. Do you think Hank Azaria owns the voice of 'Homer Simpson'? Or does Fox own that? It would be crazy to develop a show and then be held hostage to your voice actors for all future shows - what if they get hit by a car?
The article clearly disputes this. They hired and worked with the voice actor for Sky months before the first time SJ was contacted, and the voice actor used for Sky never had the movie Her or SJ's name mentioned to her a single time
The Movie Her predates all of this by years, and Sam Altman even tweeted "her"! The OpenAI team are clearly well aware of Scarlett's voice (its inconceivable the majority of the team at OpenAI haven't at least seen part of the film that almost defined their industry). The movie predates all of this by years - of course they knew.
When auditioning actors "months before" they can still look for an actor who guess what? Sounds like SJ, even "before the first time SJ was contacted".
As the actor - I'd likely also be looking to emulate SJ in Her - its clearly what the client was looking for.
I would say that OpenAI wanted something that sounded like her which in turn sounded like Scarlett Johannson.
I also think the "sounded like" is less clear than you think. Is it similar, yes. But how similar I am not sure what the line is but for sure I didn't think it was Scarlett Johannson. By saying it is Scarlett Johannson and relating it to her our brains will make the association though. That is marketing.
Let's say I'm making a movie. I have an old wizard character similar to Gandolf in Lord of the Rings, so I contact the guy who played Gandolf in Lord of the Rings. He says no, so I hire a different actor who also fits the "old wise wizard" archetype.
> I guess they want us to believe this happened without shenanigans, but it's bit hard to.
Right. And the question is, did they actually used SJ's voice as part of their training data? Because there's a lot of that available given all her works.
There's a reason why they wanted 'her', specifically. What reason is that? If they could just work with a noname voice actress (likely, for far cheaper), why not just do that from the get go? It could be a marketing gimmick and maybe they wanted her name more than just the voice to add to the buzz. If it is not that, then the sequence of events doesn't make sense.
> In a statement from the Sky actress provided by her agent, she wrote that at times the backlash “feels personal being that it’s just my natural voice and I’ve never been compared to her by the people who do know me closely.”
This isn’t the timeline though? The actor for Sky was hired and cast before they even reached out to SkarJo. The idea that they wanted to literally reproduce “Her” feels like motivated reasoning to me.
That would require a step 3 where they get in a time machine:
> But while many hear an eerie resemblance between “Sky” and Johansson’s “Her” character, an actress was hired to create the Sky voice months before Altman contacted Johansson, according to documents, recordings, casting directors and the actress’s agent.
If you say "yes we can" as your corporate announcement of that person who sounds like Obama, and one of your employees (or rather ex-executives) says "the secret ingredient in AGI is Obama", it actually can be illegal. The main issue in NIL rights (as with trademarks) isn't similarity - it's brand confusion.
- the original report by Scarlett said she was approached months ago, and then two days prior to launch of GPT-4o she was approached again
Because of the above, my immediate assumption was that OpenAI definitely did her dirty. But this report from WaPo debunks at least some of it, because the records they have seen show that the voice actor was contacted months in advance prior to OpenAI contacting Scarlett for the first time. (also goes to show just how many months in advance OpenAI is working on projects)
However, this does not dispel the fact that OpenAI did contact Scarlett, and Sam Altman did post the tweet saying "her", and the voice has at least "some" resemblance of Scarlett's voice, at least enough to have two different groups saying that it does, and the other saying that it does not.
I don't know, to me, it's just sounds like they know how to cover all their bases.
To me, it sounds like they had the idea to make their AI sound like "her". For the initial version, they had a voice actor that sounds like the movie, as a proof of concept.
They still liked it, so it was time to contact the real star. In the end, it's not just the voice, it would have been the brand, just imagine the buzz they would have got if Scarlett J was the official voice of the company. She said no, and they were like, "too bad, we already decided how she will sound like, the only difference is whether it will be labelled as SJ or not".
In the end, someone probably felt like it's a bit too dodgy as it resemblance was uncanny, they gave it another go, probably ready to offer more money, she still refused, but in the end, it didn't change a thing.
Agreed — seems like they had a plan, and probably talked extensively with Legal about how to develop and execute the plan to give themselves plausible deniability. The tweet was inadvisable, and undoubtedly not part of the actual plan (unless it was to get PR).
A more charitable scenario might be that they hire the voice actor and it sounds a bit like her. Someone suggests why don't we just get Scarlett to do it properly, wouldn't that be cooler? They reach out and she says no. They decide to continue with the one that sounds a bit like her.
> In the end, someone probably felt like it's a bit too dodgy as it resemblance was uncanny
What if it wasn’t a computer voice model but rather a real-life voice actress that you could pay a few cents to try to imitate Scarlett Johansson’s voice as best as she could?
That’s effectively what’s happening here, and it isn’t illegal.
It guess it also leads to the bigger question: do celebrities own their particular frequency range? Is no one allowed to publicly sound like them? Feels like the AACS DVD encryption key controversy all-over again.
> they gave it another go, probably ready to offer more money, she still refused, but in the end, it didn't change a thing.
That's not what she said happened. She said they released it anyway before she and Sam could connect, after Sam had reached out, for the second time, two days prior to the release.
> In the end, someone probably felt like it's a bit too dodgy as it resemblance was uncanny, they gave it another go, probably ready to offer more money, she still refused,
That was just a few days before launch, right? What was their plan if she said yes at that point? Continue using the "not-her" voice but say it was her? Or did they also have her voice already cloned by then and just needed to flip a switch?
A plausible alternative explanation for asking Johansson:
(1) They cast the current actor to test the technology and have a fallback. The actor sounds somewhat different from Johansson but the delivery of the lines is similar.
(2) They then ask Johansson because they want to be the company that brought “Her” to life. She declines.
(3) They try again shortly before the event because they really want it to happen.
(4) They proceed with the original voice, and the “her” tweet happens because they want to be the ones that made it real.
Asking shortly before the release is the weakest link here. It’s possible they already had a version trained or fine tuned on her voice that they could swap in at the last minute. That could explain some of the caginess. Not saying it’s what happened or is even likely, but it feels like a reasonable possibility.
My unsubstantiated theory: They have a voice trained on Johansson's body of work ready to go, but didn't release it because they didn't get her permission. This explains why they were still asking her right up to the ChatGPT-4o release. Then people (including Johansson) associate this Sky voice with Johansson and Her. OpenAI realizes it looks bad, despite not being intentional, so they pull Sky for PR reasons.
Yes, but it changes the narrative from “they couldn’t get Scarlett to record the voice, so they copied her voice” to something much less malicious. Contacting Scarlett, when you already have voice recordings ready but would prefer someone famous, isn’t that bad of a thing imho.
> Yes, but it changes the narrative from “they couldn’t get Scarlett to record the voice, so they copied her voice” to something much less malicious.
I don't think it's less malicious if they decided to copy her voice without her consent, but just didn't tell her until the project was underway, then continued even after she said no.
There's legal precedent that hiring a copycat is not OK, so it's not like proving it was a copycat salvages their situation.
I wouldn't be surprised if the real reason they hired a copycat early is because they realized they'd need far more of Johansson's time than she'd be willing to provide, and the plan was typical SV "ask forgiveness not permission, but do it anyway regardless."
If the goal was to make the voice sound like the one from Her, then it's still illegal.
Same way you can't get someone who sounds like a famous celebrity to do voice in a commercial and just let people think it's the famous celebrity when it's not
Unless they can clearly demostrate reproducing the voice from raw voice actor recordings, this could be just a parallel construction to cover their asses for exactly this sort of case.
When discovery happens and there’s a trail of messages suggesting either getting ScarJo or finding someone that sounds enough like her this isn’t going to look good with all the other events in timeline.
I'm not sure if that's enough to protect OAI, it feels like they wanted SJ, found a similar voice actor as a version 1, tried to "officially" get SJ's voice, and when it failed instead of pulling it continued on. It still feels quite a deliberate move to use her likeness, and the "contact 2 days before" sounds like they really wanted to get her okay before using the other VA's voice.
Sounds more plausible that someone pointed out to them internally they could be in a heap of trouble if Scarlett objected after they released it. It doesn’t matter if it was actually her voice or not it matters if people think it was her voice. If someone pointed this out late in the process than yeah there would have been a mad scramble to get Scarlett to sign off. When she didn’t then that put them in a bad spot.
> Is it a crime for voice actors to sound similar to, say, James Earl Jones?
And the answer is, of course: It depends. For one thing, it depends on whether the company using the sound-alike's voice are in a business closely related to the theme of Star Wars, and whether they market whatever it is they're marketing by referring to Jones' iconic performance as Vader. ("<PANT> ... <PANT>") If they do that, then yes, it most likely is.
Yeah, sus af because of the call 2 days before they released it to the world. And they were just asking for it when they tweeted the frickin "her". I mean, come on.
I never comment on HN I’ve just always been a long time lurker but I feel like I’m going crazy here reading comments.
SJ is not the “AI” portrayed in the movie her. And AFAIK she does not in fact have all the same idiosyncrasies and tones in real life as the voice does in the movie because she was in fact directed to act like that.
Not only that but the voices are not the same because there was another actress for sky as we have seen.
To me It seems as if the case for SJ is DOA unless it comes out somehow that they in fact trained on her voice specifically. But since that doesn’t seem like the case I have no idea how SJ can legally own all voices that sound like hers.
It would obviously be a different story if OpenAI were saying that sky was SJ but that’s not the case. To me the question should be is “can the studio own the character in her that openAI was copying and any similar things”. Which given that systems like SIRI were already out there in the world when the movie came out and we knew this tech was on the way. The answer should be no but IANAL.
I’m not a huge fan of OpenAI anymore and I think they deserve criticism for many things. But this situation isn’t one of them.
Clarification: Of course if it turns out that they in fact trained on SJ or altered the voice to be more like hers then I’d think differently. I still think the studio has more of a claim though look from the outside and not being a lawyer.
It's not a question of owning all voices that sound like her, it's a question of "are customers deceived into thinking it is her" and "does it affect SJ negatively to be associated with this sound alike" when her income comes partly from her distinctive voice (much like Morgan Freeman). Sam Altman tweeting "Her" right before the announcements is what builds the case for SJ.
Imagine we hired a Leo Messi look alike and made him play football badly or something worse, if viewers can clearly tell it's not him it falls under parody but if we use camera trickery to keep a fooling doubt, we could be in legal trouble.
I think Morgan Freeman is a useful comparison to make. Imitations of his voice have been used in a lot of political campaign videos (not sure how many of them got permission). An imitation of his voice was also used in a UK "morethan" advert where they did seek permission and pay him. Another highly popular AI voice would be David Attenborough, used in any number of videos.
It is not about the voice. Rather using the fame of known actress to boost the product. If your inner motive is to sound like her because she is well-known, differences in voice does not matter much.
So if I have a company that sells, say, manure, I can search and hire a voice actress that sounds exactly like Scarlett to promote me in radio ads? And write a tweet that vaguely implies that it's really her?
Exactly. I barely know who this actress is. To me, it sounds like the tens of thousands of other white american voices. How is the remotely too similar?
Well as the article states, there’s legal precedent protecting actors from having their voices “impersonated” by other actors. The fact that Altman tweeted “her” and contacted Johansson can make the case for the intent to impersonate.
> because she was in fact directed to act like that
It's still Scarlet Johansons voice and acting. The same role with the same lines read by different actors would be very different. Imagine for example that they would have cast Tilda Swinton for Samantha. Even with the same script it would probably end up a very different character. Actors aren't interchangeable.
It's very clear that OpenAI was trying to make ChatGPT sound like Samantha from Her. Whether they used Scarlet Johansons voice to train, or excerpts from the movie, or had writers come up with typical responses that sound similar to Samantha are details, and it's up to the lawyers to figure out whether this is legal or not.
But the undisputable fact is that OpenAI took heavy inspiration from a movie, and did so without permission. You could argue that taking inspiration from a popular movie is fair game, but I'm not sure where the line is between "inspiration" and a blatant rip-off.
But is that the point? Here is a relevant precedence, for instance, that may or may not change your mind:
Tom Waits is a singer known for his raspy singing voice. Back in the late 1980s, Frito-Lay, Inc., the makes of Doritos, thought it was a great idea to run an ad in which the music had the atmosphere and feel of a Tom Waits song. Except the professional singer they hired for that got the job done a bit too well: the sounds of his voice in the commercial was so close to Tom Waits' work (he had for ten years sang in a band covering Tom Waits songs) that in November 1988, Waits successfully sued Frito-Lay and the advertising company Tracy-Locke Inc., for voice misappropriation under California law and false endorsement under the Lanham Act [1].
Now, when you hear Tom Waits speak in interviews, I find that his voice does not sound nearly as raspy as in his performances. But the point is that it does not matter so much whether OpenAI used the actual voice of Johansson or hired someone to imitate her performance.
Given the fact that Johansson was initially contacted by OpenAI to provide her voice and declined, we can surely assume that the selection of the particular voice actress they ended up using was no coincidence.
>Given the fact that Johansson was initially contacted by OpenAI to provide her voice and declined
This order is wrong according to the article, the VA was contracted before ever reaching out to SJ.g
Additionally here is a relevant anecdote, for instance, that may or may not change your mind?
>In a statement from the Sky actress provided by her agent, she wrote that at times the backlash “feels personal being that it’s just my natural voice and I’ve never been compared to her by the people who do know me closely.”
It would suck to be blacklisted from your career because your voice may sound too similar to another famous person, if viewed from a certain light.
The point is : for a little podcast to use some AI to make a couple of ephemeral jokes about real people should absolutely be allowed and might be one of the few moral use cases of AI. (see dudesy podcast humor like george carlin standup and tom brady standup)
But for a massive tech company, to fuck over an individual artist in such a blatantly disrespectful way is hugely different.
When I first used ChatGPT's voice assistant's I was like "Wow, this one is clearly Scarlett Johansson from Her, they even copy her mannerisms."
No amount of unverifiable "records" (just pieces of paper provided by somebody who has a multimillion dollar incentive to show one outcome) will change my mind.
But if they can produce the actual voice artist I'd be more open-minded.
Funny, I'm the opposite. I saw clips from the film after the controversy (it's been ten years since I saw the film itself) and Sky sounds nothing like Johansson to me. No amount of unverifiable "records".
1. The sky voice currently available in the app is a different model from the one they presented (one is pure TTS, the new one in GPT-4o is a proper multi modal model that can do speech in and out end to end)
And then there's me, and I'm somewhere in the middle. When I first heard that voice, I didn't really think anything of it. But retrospectively given the media reporting from Sam Altman tweeting about the movie and the reports of approaching Scarlet Johansson, I can make that connection. But I would not have without the context. And without real reporting I would have dismissed it all as speculation.
They voice artist put out a statement through her lawyer. She also stated her voice has never been compared to Scarlett in real life by anyone who knows her.
that's because scarlett's voice is pretty generic white upper middle class woman with a hint of vocal fry, and a slight hint of california (pretty typical given pervasiveness of media from california).
She's not exactly gilbert gottfried or morgan freeman.
I'd like to hear her raw voice compared to the polished product. Listen to famous singers' acoustic vs. heavily audio-engineered final cuts. Big difference. I think if you played this OpenAI "Sky" voice to a sample population and said it was a famous person's voice, SA would come up frequently.
This shows how bad it is. If you're proactively sharing a package of docs with the Washingington Post, you're toast.
Altman's outreach, his tweet, and the thousands of tweets and comments talking about how similar Sky is to ScarJo is enough to win the case in California.
Then we can add this to the long list of insane lawsuits going the wrong way in California.
They asked SJ, she said no. So they went to a voice actor and used her. Case closed, they didn't use SJ's voice without her permission. That doesn't violate any law to any reasonable person.
I don't think the mannerisms of a performance something that's copyrightable though. It sounded like they used a voice actor who was instructed to speak with a similar intonation as Her, but Scarlet Johansson's voice is more raspy, whereas Sky just sounds like a generic valley girl.
For a case to the contrary: Midler v. Ford -- a case in which Ford hired one of Bette Midler's ex-backup singers to duplicate one of her performances for an ad (after trying and failing to get Midler herself). Ford never said this was actually Midler -- and it wasn't -- but Midler still sued and won. https://law.justia.com/cases/federal/appellate-courts/F2/849...
"Her" is one of my favorite movies of all time, and not once while watching the demo did I think that it sounded specifically like ScarJo. The whole concept, of course, made me think of "Her", but not the voice itself.
As a non-American I only hear Scarlett Johansson's voice in the examples I've heard, to me it clearly is an impersonation. Maybe state-side that specific voice sound is more common and thus less recognisable as Scarlett Johansson's.
This whole thing is starting to feel like another Sam Altman spotlight production. There's enough evidence to show no wrongdoing, but it was handled in a way to make people think there was a scandal. Maximum spotlight for relatively low risk. I wonder if people will get tired of being jerked around like this.
I'm genuinely not sure what you're trying to say here. Are you claiming that this was somehow engineered by Altman on purpose to draw attention, because all publicity is good publicity? Or engineered by his enemies to throw mud at Altman, because if you throw enough some of it will stick?
Occam's Razor argues that Sam simply wanted ScarJo's voice, but couldn't get it, so they came up with a legally probably technically OK but ethically murky clone.
> they came up with a legally probably technically OK but ethically murky clone.
Isn't what OpenAI does all the time? Do ethically murky things, and when people react, move the goal posts by saying "Well, it's not illegal now, is it?".
I would like to think that a normal person, having not been able to hire voice work from a specific well-known actor, and wanting to avoid any image of impropriety, would use a completely different voice instead. Sam isn't dumb, he knew the optics of this choice, but he chose it anyways, and here we all are, talking about OpenAI again.
It's not a clone. What is ethically murky about it?
You want Brad Pitt for your movie. He says no. You hire Benicio Del Toro because of the physical resemblence. Big deal.
Having seen "Her" and many other Scarlet Johansson movies, I didn't think for a second that GPT-4o sounded like her. On the contrary, I wondered why they had chosen the voice of a middle aged woman, and whether that was about being woke. It wasn't until social media went hysterical, I realized that the voices were sort of similar.
This was rocket fuel for activists trying to get a nationwide personality rights law on the books. That would almost certainly increase costs for OpenAI.
I don't think you understand. It's extremely well established in law, you can't approach someone to voice an advert for you, get told no, and then hire an impersonator to do it. Take all the AI hype bullshit and the cult of personality bullshit out of it. What Altman did is very standard and very clearly not allowed. He will end up paying for this in monetary terms, and paying further for it in the reputation damage - in that no one can trust OpenAI to conduct business in good faith.
This assumes that the voice actress was an impersonator. By her own statements, no one who knows her has said that her voice sounds like Scarlett Johansson (personally, I agree). And she was auditioned and hired before SJ was even approached. I don't think that this falls under the "very standard" scenario you reference.
If I didn't much care for my critics, then letting them invent a lot out of story I can rebut easily is worth waiting a few days, knowing full well I can publish it widely whenever I want.
An ordinary person worries all the time about dealing with the legal system. A big company does it all the time.
I mean clearly having Scarlett Johansson on board was plan A.
Bringing the voice offline and then revealing it was a recording of someone else who coincidentally sounded exactly the same is definitely plan B or C though.
I don't understand how you can trust OpenAI so much to think it was all an accident.
> I honestly don't understand how delusional you have to be to think OpenAI wanted this to happen.
(1) I've become tired of the "I honestly don't understand" prefix. Is the person saying it genuinely hoping to be shown better ways of understanding? Maybe, maybe not, but I'll err on the side of charitability.
(2) So, if the commenter above is reading this: please try to take all of this constructively. There are often opportunities to recalibrate one's thinking and/or write more precisely. This is not a veiled insult; I'm quite sincere. I'm also hoping the human ego won't be in the way, which is a risky gamble.
(3) Why is the commenter so sure the other person is delusional? Whatever one thinks about the underlying claim, one would be wise to admit one's own fallibility and thus uncertainty.
(4) If the commenter was genuinely curious why someone else thought something, it would be better to not presuppose they are "delusional". Doing that makes it very hard to curious and impairs a sincere effort to understand (rather than dismiss).
(5) It is muddled thinking to lump the intentions of all of "OpenAI" into one claimed agent with clear intentions. This just isn't how organizations work.
(6) (continuing from (5)...) this isn't even how individuals work. Virtually all people harbor an inconsistent mess of intentions that vary over time. You might think this is hair-splitting, but if you want to _predict_ why people do specific irrational things, you'll find this level of detail is required. Assuming a perfect utility function run by a perfect optimizer is wishful thinking and doesn't match the experimental evidence.
If it wasn't for us being biased by the surrounding circumstances I don't think people would have confused their voices. Their voices are not that similar. I probably personally know people with a voice as similar to SJ as Sky's. You probably do too.
The voice actress says the same: "I’ve never been compared to her by the people who do know me closely."
But then suddenly a story emerges and their voices are indistinguishable. All of these extra details shouldn't have even mattered.
Because of "Her" I imagine and all the memes about GPT-4o. The "Her" and GPT-4o memes I always thought of them as "they both have a real-time charming female AI assistant" and not like they have the same voice.
the tweet, plus all of the last minute trying to get scarlett to sign off on it signals an intent to try and make it sound like "her"
why be an apologist?
OpenAI was given an opportunity by Scarlett to prove that they did not intend to make it sound like her, and instead of responding their choice was to take down the voice. (yet another signal)
I think you'd have to be willfully ignorant to believe there wasn't some intent here.
Whether or not it's legal to copy someone's likeness in this fashion is another story.
I never made the connection between the Sky voice and Scarlett Johansson's. I've seen many of her movies. She has an extremely distinctive voice that has a certain huskiness to it and the Sky voice totally lacks that.
Some voices are sexy and both of them fall into that category -- but that's beside the point.
That aside, it is genuinely pleasant to have a conversation with chatGPTo and some of that has to do with the voices. There's a kind of irony here because people generally imagined that AI would be cold, logical, unempathetic, etc. But LLMs are the opposite; they're extremely polite and deferential. Meanwhile they aren't that good at logic!
I don't find any of the OpenAI voices sexy or deferential. They sound fake happy to me, like a customer service phone menu or an elementary school teacher, and reek of Bay area vocal fry [1] and lilt. I wish there was a greater diversity of accents and speaking patterns available, such as can be seen on the Speech Accent Archive [2].
Readit News
* OpenAI wanted an AI voice that sounds like SJ
* SJ declined
* OpenAI got an AI voice that sounds like SJ anyway
I guess they want us to believe this happened without shenanigans, but it's bit hard to.
The headline of the article is a little funny, because records can't really show they weren't looking for an SJ sound-alike. They can just show that those records didn't mention it. The key decision-makers could simply have agreed to keep that fact close-to-the-vest -- they may have well understood that knocking off a high-profile actress was legally perilous.
Also, I think we can readily assume OpenAI understood that one of their potential voices sounded a lot like SJ. Since they were pursuing her they must have had a pretty good idea of what they were going after, especially considering the likely price tag. So even if an SJ voice wasn't the original goal, it clearly became an important goal to them. They surely listened to demos for many voice actors, auditioned a number of them, and may even have recorded many of them, but somehow they selected one for release who seemed to sound a lot like SJ.
Altman appears to be an habitual liar. Note his recent claim not to be aware of the non-disparagement and claw-back terms he had departing employees agree to. Are we supposed to believe that the company lawyer or head of HR did this without consulting (or more likely being instructed by) the co-founder and CEO?!
Yes this is pretty typical. The CEO doesn’t make all decisions. They hire people to make decisions. A company’s head of legal could definitely make decisions about what standard language to use in documents on their own.
"In a statement from the Sky actress provided by her agent, she wrote that at times the backlash “feels personal being that it’s just my natural voice and I’ve never been compared to her by the people who do know me closely.”"
It did not seem like an issue before and the Sky voice was public many months before GPT4o. I don't believe SJ can claim to own all young, attractive woman voices whether they are used as a voice assistant or not. It seems like the issue is being blown out of proportion. It does make a good story. The public perception of AI right now is generally negative and people are looking for reasons to disparage AI companies. Maybe there are good reasons sometimes, but this one is not it.
It kinda feels like its on purpose. Someone in a previous thread mentioned that this might have been a cynical marketing ploy and I'm warming up to the theory. After they recorded the Sky VA, they figured out a whole marketing campaign with SJ to promote the voice feature. After she turned them down (twice), they released with just enough crumbs referencing the movie to goad SJ into committing a first degree Streisand.
With the slow roll out, everyone would have forgotten about the feature the day after the announcement but now it's been in the news for a week, constantly reminding everyone of what's coming up.
This seems similar to the latest season of Rick and Morty. Whether justified or not in that particular case, it rubs me the wrong way a bit in principle to think that a production can fire someone only to hire someone else to do a near-perfect copy of their likeness. If (as in the OpenAI case) they'd gone further and trained an AI on the impressions of Justin's voice, would that have been considered an AI impersonation of Justin with extra steps?
All of which is to say, this seems like a pretty interesting legal question to me, and potentially broader than just AI.
To your point, there have many female assistant voices on the market, including Sky -- but what might have tripped the line of impersonation was the context this particular one was presented and marketed. I don't know where exactly that line should be, but you can certainly reject this kind of marketing without stifling anybody's legitimate career.
People do celebrity impressions all the time, and that's not infringement either, because it's not actually copying that person's voice.
I'm sympathetic to SJ in this matter, especially after the Disney Black Widow debacle, but it sounds like she had the opportunity to write herself a nice check, and she turned it down.
On the basis of this article, it sounds like she doesn't have the cause of action that she had believed she had; I imagine that her legal team are now advising a fast settlement, but OpenAI's legal team might prefer to milk the free publicity for as long as they can, especially if they are fairly certain they would prevail at trial.
If I were SJ, I'd turn it down too. Shes in no need of money, and selling her voice to OpenAI would make most of creators and every single voice actor hate her (not to mention the Twitter mob).
In majority of creative circles, the current social norm is to hate AI, so touching AI in any way is too risky for reputation.
For a case that doesn't involve AI at all, consider situations where a voice actor in a cartoon is replaced (sometimes while still alive) by someone who can perform a voice that sounds the same. Decisively not illegal. Most people don't even find it immoral, as long as the reason for getting rid of the original voice actor wasn't wrong on its own (e.g. Roiland).
Because there are contractual clauses. Do you think Hank Azaria owns the voice of 'Homer Simpson'? Or does Fox own that? It would be crazy to develop a show and then be held hostage to your voice actors for all future shows - what if they get hit by a car?
When auditioning actors "months before" they can still look for an actor who guess what? Sounds like SJ, even "before the first time SJ was contacted".
As the actor - I'd likely also be looking to emulate SJ in Her - its clearly what the client was looking for.
> never had the movie Her or SJ's name mentioned to her a single time
How do you know that?
I also think the "sounded like" is less clear than you think. Is it similar, yes. But how similar I am not sure what the line is but for sure I didn't think it was Scarlett Johannson. By saying it is Scarlett Johannson and relating it to her our brains will make the association though. That is marketing.
Let's say I'm making a movie. I have an old wizard character similar to Gandolf in Lord of the Rings, so I contact the guy who played Gandolf in Lord of the Rings. He says no, so I hire a different actor who also fits the "old wise wizard" archetype.
Is any of that illegal?
Right. And the question is, did they actually used SJ's voice as part of their training data? Because there's a lot of that available given all her works.
There's a reason why they wanted 'her', specifically. What reason is that? If they could just work with a noname voice actress (likely, for far cheaper), why not just do that from the get go? It could be a marketing gimmick and maybe they wanted her name more than just the voice to add to the buzz. If it is not that, then the sequence of events doesn't make sense.
If you literally use SJ's image or voice, then you're in trouble.
If it's an SJ lookalike or soundalike (and you don't claim otherwise), there's no problem.
Right? What's the "shenanigans?"
This isn't true. At least with respect to "soundalike" see, e.g., Waits v. Frito-Lay 978 F.2d 1093 and Midler v. Ford Motor Co. 849 F.2d 460.
* OpenAI wanted an AI voice that is SJ's voice
* SJ declined
* OpenAI got an AI voice from another person that sounds like SJ
> But while many hear an eerie resemblance between “Sky” and Johansson’s “Her” character, an actress was hired to create the Sky voice months before Altman contacted Johansson, according to documents, recordings, casting directors and the actress’s agent.
- the original report by Scarlett said she was approached months ago, and then two days prior to launch of GPT-4o she was approached again
Because of the above, my immediate assumption was that OpenAI definitely did her dirty. But this report from WaPo debunks at least some of it, because the records they have seen show that the voice actor was contacted months in advance prior to OpenAI contacting Scarlett for the first time. (also goes to show just how many months in advance OpenAI is working on projects)
However, this does not dispel the fact that OpenAI did contact Scarlett, and Sam Altman did post the tweet saying "her", and the voice has at least "some" resemblance of Scarlett's voice, at least enough to have two different groups saying that it does, and the other saying that it does not.
To me, it sounds like they had the idea to make their AI sound like "her". For the initial version, they had a voice actor that sounds like the movie, as a proof of concept.
They still liked it, so it was time to contact the real star. In the end, it's not just the voice, it would have been the brand, just imagine the buzz they would have got if Scarlett J was the official voice of the company. She said no, and they were like, "too bad, we already decided how she will sound like, the only difference is whether it will be labelled as SJ or not".
In the end, someone probably felt like it's a bit too dodgy as it resemblance was uncanny, they gave it another go, probably ready to offer more money, she still refused, but in the end, it didn't change a thing.
What if it wasn’t a computer voice model but rather a real-life voice actress that you could pay a few cents to try to imitate Scarlett Johansson’s voice as best as she could?
That’s effectively what’s happening here, and it isn’t illegal.
It guess it also leads to the bigger question: do celebrities own their particular frequency range? Is no one allowed to publicly sound like them? Feels like the AACS DVD encryption key controversy all-over again.
That's not what she said happened. She said they released it anyway before she and Sam could connect, after Sam had reached out, for the second time, two days prior to the release.
That was just a few days before launch, right? What was their plan if she said yes at that point? Continue using the "not-her" voice but say it was her? Or did they also have her voice already cloned by then and just needed to flip a switch?
Dead Comment
Dead Comment
I don't think it's less malicious if they decided to copy her voice without her consent, but just didn't tell her until the project was underway, then continued even after she said no.
There's legal precedent that hiring a copycat is not OK, so it's not like proving it was a copycat salvages their situation.
I wouldn't be surprised if the real reason they hired a copycat early is because they realized they'd need far more of Johansson's time than she'd be willing to provide, and the plan was typical SV "ask forgiveness not permission, but do it anyway regardless."
Same way you can't get someone who sounds like a famous celebrity to do voice in a commercial and just let people think it's the famous celebrity when it's not
When discovery happens and there’s a trail of messages suggesting either getting ScarJo or finding someone that sounds enough like her this isn’t going to look good with all the other events in timeline.
If it goes to court, they’ll settle.
> Is it a crime for voice actors to sound similar to, say, James Earl Jones?
And the answer is, of course: It depends. For one thing, it depends on whether the company using the sound-alike's voice are in a business closely related to the theme of Star Wars, and whether they market whatever it is they're marketing by referring to Jones' iconic performance as Vader. ("<PANT> ... <PANT>") If they do that, then yes, it most likely is.
Deleted Comment
Dead Comment
SJ is not the “AI” portrayed in the movie her. And AFAIK she does not in fact have all the same idiosyncrasies and tones in real life as the voice does in the movie because she was in fact directed to act like that.
Not only that but the voices are not the same because there was another actress for sky as we have seen.
To me It seems as if the case for SJ is DOA unless it comes out somehow that they in fact trained on her voice specifically. But since that doesn’t seem like the case I have no idea how SJ can legally own all voices that sound like hers.
It would obviously be a different story if OpenAI were saying that sky was SJ but that’s not the case. To me the question should be is “can the studio own the character in her that openAI was copying and any similar things”. Which given that systems like SIRI were already out there in the world when the movie came out and we knew this tech was on the way. The answer should be no but IANAL.
I’m not a huge fan of OpenAI anymore and I think they deserve criticism for many things. But this situation isn’t one of them.
Clarification: Of course if it turns out that they in fact trained on SJ or altered the voice to be more like hers then I’d think differently. I still think the studio has more of a claim though look from the outside and not being a lawyer.
Edit: clarification
Imagine we hired a Leo Messi look alike and made him play football badly or something worse, if viewers can clearly tell it's not him it falls under parody but if we use camera trickery to keep a fooling doubt, we could be in legal trouble.
SJ doesn’t get to own the voice rights to everyone that sounds at all like her just because she is famous.
Sounds like: "Eh nevermind, we are going to use it anyway and BTW, I'm going to tweet 'HER' "
You don't think that will have no weight whatsoever in a lawsuit?
The article doesn't state that. It says that about singers. Very different.
It's still Scarlet Johansons voice and acting. The same role with the same lines read by different actors would be very different. Imagine for example that they would have cast Tilda Swinton for Samantha. Even with the same script it would probably end up a very different character. Actors aren't interchangeable.
It's very clear that OpenAI was trying to make ChatGPT sound like Samantha from Her. Whether they used Scarlet Johansons voice to train, or excerpts from the movie, or had writers come up with typical responses that sound similar to Samantha are details, and it's up to the lawyers to figure out whether this is legal or not.
But the undisputable fact is that OpenAI took heavy inspiration from a movie, and did so without permission. You could argue that taking inspiration from a popular movie is fair game, but I'm not sure where the line is between "inspiration" and a blatant rip-off.
Tom Waits is a singer known for his raspy singing voice. Back in the late 1980s, Frito-Lay, Inc., the makes of Doritos, thought it was a great idea to run an ad in which the music had the atmosphere and feel of a Tom Waits song. Except the professional singer they hired for that got the job done a bit too well: the sounds of his voice in the commercial was so close to Tom Waits' work (he had for ten years sang in a band covering Tom Waits songs) that in November 1988, Waits successfully sued Frito-Lay and the advertising company Tracy-Locke Inc., for voice misappropriation under California law and false endorsement under the Lanham Act [1].
Now, when you hear Tom Waits speak in interviews, I find that his voice does not sound nearly as raspy as in his performances. But the point is that it does not matter so much whether OpenAI used the actual voice of Johansson or hired someone to imitate her performance.
Given the fact that Johansson was initially contacted by OpenAI to provide her voice and declined, we can surely assume that the selection of the particular voice actress they ended up using was no coincidence.
[1] http://law2.umkc.edu/faculty/projects/ftrials/communications...
This order is wrong according to the article, the VA was contracted before ever reaching out to SJ.g
Additionally here is a relevant anecdote, for instance, that may or may not change your mind?
>In a statement from the Sky actress provided by her agent, she wrote that at times the backlash “feels personal being that it’s just my natural voice and I’ve never been compared to her by the people who do know me closely.”
It would suck to be blacklisted from your career because your voice may sound too similar to another famous person, if viewed from a certain light.
But for a massive tech company, to fuck over an individual artist in such a blatantly disrespectful way is hugely different.
Dead Comment
No amount of unverifiable "records" (just pieces of paper provided by somebody who has a multimillion dollar incentive to show one outcome) will change my mind.
But if they can produce the actual voice artist I'd be more open-minded.
2. Look at these images and tell me they didn't intend to replicate "Her": https://x.com/michalwols/status/1792709377528647995
She's not exactly gilbert gottfried or morgan freeman.
Altman's outreach, his tweet, and the thousands of tweets and comments talking about how similar Sky is to ScarJo is enough to win the case in California.
They asked SJ, she said no. So they went to a voice actor and used her. Case closed, they didn't use SJ's voice without her permission. That doesn't violate any law to any reasonable person.
Yes, this discussion is about right of publicity, not copyright.
Copyright is not the whole of the law.
That's a pretty low bar for "produced the actual voice artist".
Is my my google-fu failing me and I'm not looking in the right place?
Occam's Razor argues that Sam simply wanted ScarJo's voice, but couldn't get it, so they came up with a legally probably technically OK but ethically murky clone.
Isn't what OpenAI does all the time? Do ethically murky things, and when people react, move the goal posts by saying "Well, it's not illegal now, is it?".
You want Brad Pitt for your movie. He says no. You hire Benicio Del Toro because of the physical resemblence. Big deal.
Having seen "Her" and many other Scarlet Johansson movies, I didn't think for a second that GPT-4o sounded like her. On the contrary, I wondered why they had chosen the voice of a middle aged woman, and whether that was about being woke. It wasn't until social media went hysterical, I realized that the voices were sort of similar.
This was rocket fuel for activists trying to get a nationwide personality rights law on the books. That would almost certainly increase costs for OpenAI.
And every one of it's competitors. I think regulatory capture would be just as much, if not more, of a victory for OpenAI.
Can you explain and/or cite the legal basis here? What cases? What law?
An ordinary person worries all the time about dealing with the legal system. A big company does it all the time.
Bringing the voice offline and then revealing it was a recording of someone else who coincidentally sounded exactly the same is definitely plan B or C though.
I don't understand how you can trust OpenAI so much to think it was all an accident.
(1) I've become tired of the "I honestly don't understand" prefix. Is the person saying it genuinely hoping to be shown better ways of understanding? Maybe, maybe not, but I'll err on the side of charitability.
(2) So, if the commenter above is reading this: please try to take all of this constructively. There are often opportunities to recalibrate one's thinking and/or write more precisely. This is not a veiled insult; I'm quite sincere. I'm also hoping the human ego won't be in the way, which is a risky gamble.
(3) Why is the commenter so sure the other person is delusional? Whatever one thinks about the underlying claim, one would be wise to admit one's own fallibility and thus uncertainty.
(4) If the commenter was genuinely curious why someone else thought something, it would be better to not presuppose they are "delusional". Doing that makes it very hard to curious and impairs a sincere effort to understand (rather than dismiss).
(5) It is muddled thinking to lump the intentions of all of "OpenAI" into one claimed agent with clear intentions. This just isn't how organizations work.
(6) (continuing from (5)...) this isn't even how individuals work. Virtually all people harbor an inconsistent mess of intentions that vary over time. You might think this is hair-splitting, but if you want to _predict_ why people do specific irrational things, you'll find this level of detail is required. Assuming a perfect utility function run by a perfect optimizer is wishful thinking and doesn't match the experimental evidence.
If it wasn't for us being biased by the surrounding circumstances I don't think people would have confused their voices. Their voices are not that similar. I probably personally know people with a voice as similar to SJ as Sky's. You probably do too.
The voice actress says the same: "I’ve never been compared to her by the people who do know me closely."
But then suddenly a story emerges and their voices are indistinguishable. All of these extra details shouldn't have even mattered.
If it sounds nothing like her, and there was no intent to make it sound like her, why would he tweet "The killer app of LLMs is Scarlett Johansson"?
Many people made the comparison right after it was released https://x.com/search?q=scarjo%20until%3A2024-05-14&src=typed... and https://x.com/search?q=johansson%20until%3A2024-05-14&src=ty...
why be an apologist?
OpenAI was given an opportunity by Scarlett to prove that they did not intend to make it sound like her, and instead of responding their choice was to take down the voice. (yet another signal)
I think you'd have to be willfully ignorant to believe there wasn't some intent here.
Whether or not it's legal to copy someone's likeness in this fashion is another story.
Sounds like a female. Sounds like a white person. Sounds like between 30 and 40. It must be SJ.
Some voices are sexy and both of them fall into that category -- but that's beside the point.
That aside, it is genuinely pleasant to have a conversation with chatGPTo and some of that has to do with the voices. There's a kind of irony here because people generally imagined that AI would be cold, logical, unempathetic, etc. But LLMs are the opposite; they're extremely polite and deferential. Meanwhile they aren't that good at logic!
1. https://www.youtube.com/watch?v=Q0yL2GezneU
2. https://accent.gmu.edu/browse_language.php?function=find&lan...
> Meanwhile they aren't that good at logic This statement is contentious; depends on what level of abstraction you're looking at.
> Extremely polite and deferential This is a setting that can be turned off btw.