61
votes
Scarlett Johansson says she is 'shocked, angered' over new ChatGPT voice
Link information
This data is scraped automatically and may be incorrect.
- Title
- OpenAI pulls AI voice that was compared to Scarlett Johansson in the movie 'Her'
- Authors
- Bobby Allyn
- Published
- May 20 2024
- Word count
- 685 words
Sam Altman's behavior in this whole incident is morally and ethically problematic. You sure don't want such a guy to be in charge of the frontier breaking technology like AI, especially with the threat of deep fakes looming everywhere. Microsoft did the wise thing to not further the relationship with him for technology collaboration.
Sad part is that I don't think hoisting Altman out would do much, just revolve in the next profit driven morally gray CEO "trailblazing the future". At some point it becomes a societal issue.
It's a shame because the tech is really cool, but I do hope legislation comes down hard on this. The US was way overdue for these tech companies lobbying around established laws (assuming they don't stop them to begin with).
What do you mean?
They had actually hired him for a brief time, then something didn't work out and he went back to Open AI again as CEO.
That seems like a misunderstanding of what happened. Sam Altman won the OpenAI board struggle and went back to OpenAI. Microsoft still retains a close relationship with OpenAI, it just went back to how it was before the board struggle.
That’s not how I read the situation at all. OpenAI was being intransigent so Microsoft brought Altman on as a stopgap, but then they convinced OpenAI to come to heel and they reinstalled Altman plus put Microsoft on the board so they wouldn’t be caught unaware of the board did something against their interest in the future.
As I see it, Microsoft is all in on Altman. In fact, I see it like Microsoft was about to torch their entire relationship with OpenAI and build a new division around Sam Altman if OpenAI refused to reinstate him.
It wasn't so much that it didn't work out as it was OpenAI's board had some upheaval and wanted him to come back. He didn't go to Microsoft on his own, OpenAI pushed him out and one or some of the board members changed their minds after the fact.
I feel that Scarlet Johansson has a very generic (white female) American voice. A clear low tone and a bit of vocal fry? Those are common voice qualities among modern American women.
I've always interpreted Cove and Sky as the generic white American voices, and Ember and Juniper as the generic African American voices. And then Breeze is the overly enthusiastic undergraduate voice.
edit: I just pulled up one of ScarJo's interviews as a sanity check. Her voice doesn't strike me as very distinctive. If I were to listen to it without having known it was hers, I wouldn't be able to pin it down.
Oh yeah. Her voice is near perfect generic. If they hadn't been courting her for months to license the voice it would have been fine, but...
Perhaps, given the context, it's a little suspicious?
No they very clearly were trying to reference her performance in Her. Altman’s tweet announcing it even hung a lampshade on it, as did the coverage in the press about the release.
I would say it’s more of a rasp than a vocal fry. Her way of speech is also sultry in a way that generic American females don’t talk in.
They also were trying to get her because she voiced Samantha in Her.
I am curious how one makes an AI human sounding voice that sounds like nobody on the planet. Reminds me of Deadmaus accidently recreating Darude Sandstorm in his studio.
Also the American midwestern accent is one of the closest accents to the mid-atlantic accent that has been modified for use by news casters because it is one of the easiest English language accents to understand. So if you want an easily understandable English language AI voice, odds are it's gonna sound like some random Jack or Diane from Indiana (suckin on a chili dog).
It's not that they made it sound like "anyone", it's that they made it sound like Johansson specifically, after trying and failing to get her involved, and then publicly leaning into the association (Altman tweeted the word "her" and one of their engineers changed their cover image to a screenshot from the film).
Ooh yeah, that's bad. And creepy. I thought it was less direct like "we like this movie AI and actress, use it as inspiration", not "make me an AI sex doll replica voice of Johansson".
To be clear, they claim to have used human voice actors for the training process starting before they ever approached Johansson (as opposed to cloning her voice directly using AI), but thanks to California's "likeness" laws the fact that they clearly were trying to evoke her performance could still be a legal problem for them.
Clearly OpenAI is trying to evoke Samantha, the virtual character she played in the movie. They're not trying to evoke Scarlett Johansson the real-life public figure. I'm not familiar enough with the laws or precedence to know if that is sufficient for a defense. My gut feeling is that Spike Jonze, the writer and director, might have more of a case than Johansson the voice actor. Does an actor have ownership claim over their performance of a character belonging to someone else?
I believe actors own their likenesses, but not their characters. This makes sense when you think about it: it allows new actors to portray franchise characters in sequels when the original actor is unavailable or doesn't want to continue the role, and allows cartoon adaptations of IPs, provided the characters look and sound substantially different from the live action portrayals. A prime example that my brain won't let go of is the 80s cartoon adaptation of "Ghostbusters," in which none of the main characters really look anything like the actors in the live action films. I don't know how much that reference resonates today, some 35+ years on, but like I said, my brain found that and couldn't look past it.
So, supposing OpenAI had a settled arrangement with whoever owns the "Her" IP, they could name their AI voice "Samantha" and give it certain character traits that appear in the movie, but the voice likeness rights would remain with Johansson. There's a hard line that needs to be drawn here, as we begin to cope with the ability of algorithms to replicate extant human likenesses. Unfortunately, drawing hard lines is famously difficult in IP law, so God only knows where we'll end up.
Discussion with additional context and timeline: https://news.ycombinator.com/item?id=40421225
Gross
I don't typically read HN so I scrolled through for a while and did laugh that my least favorite AI argument was present - but seemed "downvoted" (I am probably getting the terminology wrong and I don't care)
"But what if a baby had grown up hearing the same voice and spoke like SJ because of that." (Paraphrased) I personally think this is the worst argument for any sort of AI. And it's usually used to cover theft. Anyway I hit my limit on HN so thanks
The entire purpose of copyright laws is to use artificial means to create a market for artists. It’s not there to enable tech billionaires to get richer. There’s no social interest in enabling that, regardless of whatever overwrought analogies they want to make. If it’s not enabling a livelihood for artists by maintaining a functioning market for their work it doesn’t matter.
The waffling about copyright here seems very weird to me -- this case has absolutely nothing to do with copyright (aside from OpenAI also taking a very blasé attitude towards the law there). Likeness/publicity rights are what's at play here, and they're an entirely different type of intellectual property that has a completely different legal standard than copyright. Being incidentally similar to something else is a defense for copyright infringement, but even a human impressionist can be a violation of someone's right to their likeness if they're presented in a way that's intended to confuse people for the real thing -- there's plenty of precedent there.
I don't have a clue what this has to do with my comment? I just think "what if a human had learned from X and done Y" is a stupid argument for AI.
My comment is explaining why it’s a stupid argument.
Got it, it came off as if you were explaining something I wasn't getting and then I wasn't understanding why you would since I do know why it's stupid. I follow now
It's also stupid because AI isn't a person, and it isn't learning naturally or in this case "just happening to grow up to sound like ScarJo". But I see this argument about LLM and AI art apps on Tildes and elsewhere too. And I also think it's stupid then too.
I don't understand why you're explaining this situation to me. What indication have I given that I don't understand this situation? I'm not even talking about ScarJo. Why are you doing the same thing I just expressed irritation about?
Which argument is actually addressing blanket legislation? The idea of a human child? I don't actually think it is. I think it's really easy and simple to distinguish between a person looking at a piece of art, and drawing something themselves vs an "AI" iterating based on a bunch of material scraped off the internet. I don't think copyright should be eased for the purposes of "AI" use.
I will continue to think the comparison to be a stupid argument. And if people want to "address blanket legislation" they can do that.
Copyright is wholly irrelevant here even if they used ScarJo's own voice afaik, it would be issues about the use of her likeness regardless. You can't copyright your face or the sound of your voice. So copyright law would be no help regardless. And when it comes to the actual likeness rights, it literally wouldn't matter if someone was born sounding exactly like ScarJo -- hiring them to confuse people into thinking it was ScarJo would still be a violation of her likeness rights.
Bingo on the original situation. But also very confusing to me in this particular sub thread.
I was trying to respond to the implication that copyright law would somehow be more help here without this weird "what if a child" argument. I think that argument is usually bad for copyright too (and betrays a real lack of understanding of AI tbqh) but it's just utterly detached from reality in the current case.
Your last paragraph, starting with "In this case" was fully explaining the current situation. Literally you did explain it. Please don't tell me you didn't.
I recognize you don't think it's stupid. I don't agree. I also think the current "AIs" inherently do run off copyright infringement/piracy and/or theft depending on whether you talk about how the material was acquired or how they're trained to respond. I think comparing these programs to people is contributing to the deliberate misinformation on the part of the companies and individuals selling these products to everyone under the sun as a cost saving solution. I don't like them. I'm not going to feel differently about the argument.
In this case, the argument was made to defend OpenAI from the allegations. I don't know whether it was in "good faith" or not. That doesn't make it a better argument, in my opinion.
I'm not a lawyer, but it sounds like the similarity alone is potentially sufficient under California law in particular and based on the precedent set by Midler v whomever it was. Similarly had there been no AI, but just an impersonator and all the same evidence, that would also have fallen under the same law. So no, a child would not have made any difference here. The evidence for this case is particularly helpful because it's so blatant
But yeah, a lot of the time the law fails to protect people against corporations. A lot of people get fucked over, even rich and famous people. That's a problem with the protection of our civil rights, and privacy laws in general not being more robust. We're all royally fucked there.
I will not claim our US copyright law has no problems, it does. But the comparison to a child learning is absolutely a comparison to the AI, not a question of an adult impersonator, and thus this is all a side discussion unrelated to the "human learning comparison" argument I mentioned.
If you would prefer, I can call the argument weak and unconvincing. In every instance I've seen it, it has been so.
I'm sorry my comment hit a nerve, but yours has hit mine, particularly in you claiming not to explain the current case immediately after doing it.
I am still not sure whether they want her voice as something like a ironic joke, or if they genuinely watched the movie and thought this was a future worth aspiring to.
Am I allowed to reference "Don't Create the Torment Nexus" here?
I've been thinking the same. I can't take it any other way, they're genuinely embracing this stereotype, monetizing on it. Handwaving the risks as just fiction, yet blind to their realistic equivalents.
I wouldn’t say ironic so much as they just think it’s cool. I don’t necessarily think it’s part of that meme of “hey we made the evil thing from the sci fi novel/movie.” I think Her portrays these human/machine relationships in a positive light, even if at the end of the day it admits it can’t really replace human relationships.
I think they’re just taking advantage of the male loneliness thing and using a pop cultural touchstone as like an incentive.
Her has more ambiguity for sure than The Terminator but if OpenAI called a project Skynet it would at least be more clearly in the "haha"-category. Here they have clearly chosen a flirty submissive female voice and using that to exploit the loneliness market isn't exactly admirable. And that they think it is cool, makes me wonder if they really don't see the problematic undertone of that movie.
Her isn’t really dystopian though. It’s a bit melancholy but it’s more a reflection on social isolation and our relationships to technology than the specific technology being showcased. You could have written a similar story a generation prior about falling in love with a Blogger.
It doesn't exactly end on a happy note. Not for the humans, anyway. It's no Torment Nexus but not really something companies should be racing to bring to market either. Or at least not deliberately comparing their products to.
At the end the human characters do start talking to each other to share their experiences of losing their AI friends. So that’s not the worst thing. And it’s sort of implied that the experience expands Joaquin Phoenix’s capacity to relate to people once he gets over it. Samantha sort of functions as a post divorce “rebound girlfriend” that way.
Plus they basically created a super intelligent, spiritually evolved being that attaints enlightenment at the end.
OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show (Washington Post)
…
…
…
This is exactly why I find the tech industry so disgusting. I have refused to work in it again because the tech industry is like capitalism on steroids. The CEO's vision of """innovation""" at whatever cost to whoever.