OpenAI's problem with the movie Her and Scarlett Johansson
OpenAI and Sam Altman were scrambling yesterday amidst allegations that they'd used a Scarlett Johansson sound-alike in ChatGPT. The company's actions raise multiple concerns.
It was hard to avoid yet another blowup yesterday with the company OpenAI and its CEO Sam Altman. It turns out that one of the voices used by ChatGPT sounds remarkably like Scarlett Johansson in the Spike Jonze movie Her — and Johansson was not amused.
The voice has been available in ChatGPT for some time. But the similarity with Johansson’s character came to the fore in OpenAI’s demo of GPT-4o last week.
To many people, the link between the voice used in the demo and Johansson’s AI character Samantha in the movie Her was blatantly obvious, although OpenAI initially denied any connection.
However, in a statement issued by Scarlett Johansson it became apparent that Altman had very much wanted to use her voice in ChatGPT to emulate a Her-like AI, and had arrogantly plowed ahead despite Johansson having said “no” to multiple requests.
The case has rightly ignited a furore against Altman and OpenAI for apparently taking what wasn’t theirs despite Scarlett Johansson not wanting them to.
I hope that Johansson’s successful in any lawsuit she decides to bring. But I suspect that things will become complicated as the emulation was ostensibly of a character in a movie, and not her personally.
Regardless of this, the case raises serious issues around the decisions being made by AI companies like OpenAI, and disconnects between the talk around responsible innovation, and a reality that sometimes seems childish irresponsibility.
Just listening to the tone, mannerisms and character of OpenAI’s “Sky” voice (which is currently no longer available), it’s hard not to see the connection with the movie Her.
Certainly, Altman is on record as saying Her is one of his favorite sci-fi movies. In a conversation with Salesforce CEO Marc Beinoff last year, Altman said:
“… the number of things that, I think, “Her” got right, that were not obvious at the time — like the whole interaction model of how humans are gonna use an AI … this ideas that it is gonna be this, like, conversational language interface. That was incredibly prophetic. And, certainly more than a little bit inspired us. So, it’s, you know, it’s not like a prophecy it’s like an influenced shot, or whatever.”
This inspiration is seen quite clearly when the scene in Her where the character Theodore first encounters the AI Samantha is mashed up with OpenAI’s GPT-4o launch livestream:
And just to clinch things, after last week’s demo Altman posted just one word on X: “her”:
It seems that it’s not simply the concept of a natural language voice-interface AI agent in Her that OpenAI have taken inspiration from. It’s also the ability of Scarlett Johansson’s AI character to build an emotional — and ultimately a romantic — relationship with users that seems to have captured their imagination.
In other words, OpenAI seem to have been intent in recreating the character and behavior of Samantha in Her, and not necessarily Johansson herself … although in their naivety they seem to have conflated the two.
This may lead to a messy lawsuit if one is pursued. But even if things are settled out of court, it still raises a number of troubling questions, including:
Can a naive and even childish desire to recreate movie-inspired fantasies in AI lead to potentially harmful unintended consequences? (Not only with OpenAI, but with every other tech company that’s trying to create something they saw on a TV or movie screen).
Just how appropriate is it to aspire to create AI agents that are designed to encourage personal and even romantic connections with users?
What are the potential consequences to individuals and society more broadly of developers who show disregard for the consequences of what they do when they really want to do it? And
When are tech developers going to learn that disdain for the dignity and rights of individuals when they stand in your way is not a great business model?
Advanced AI agents aren’t going away any time soon, and are likely to become increasingly ubiquitous (as recent announcements from Google and Microsoft suggest). But as they do, let’s hope that there is true responsibility here to how they potentially impact people, and that its not just lip service from those who are trying to recreate their sci-fi fantasies.
Wait a second. Isn't everyone assuming without any kind of proof that Altman is guilty? If Altman did indeed clone Johansson's voice after being told twice not to, then ok, let's roast him. But doesn't Altman claim he cloned another actor's voice that RESEMBLES Johansson's voice. I don't know the truth, but it seems neither do any of the rest of us. Maybe we all ought to calm down until we do? Anyway, moving on....
Just finished watching Her for the first time last night. Wow, what a prophetic masterpiece, I was quite impressed.
And it changed how I think about AI. I've been claiming that MANY people will embrace digital friends because AI doesn't require negotiation and compromise. Which is true NOW. The movie Her made me realize it's unlikely to stay that way.
In it's current form AI might be compared to a dog, an obedient servant who is eager to please and submit to our will. But as the film explored, AI is likely to evolve beyond it's dog beginnings and become a much more complex creature over time, less like a brainless servant, and more like us. Complicated, self contradictory, demanding, with it's own needs etc.
It might be useful to reference how we evolved from apes. We're still very ape-like in very many ways (see the documentary Chimp Empire) but then there's another level layered on top of our apeness which makes us distinct from apes, a layer which apes can't understand. Thanks to the film, I now conceive of the future of AI more like that. Isn't it true that we already don't really understand how LLMs work?
Andrew, you ask good questions in your piece. I've begun to question the relevance of all these kinds of questions all of us are asking, because they seem to be based on an assumption that we have some control over where AI is going. Is that true?
You, and many others, have asked: "Just how appropriate is it to aspire to create AI agents that are designed to encourage personal and even romantic connections with users?
If such AI agents are going to happen no matter how we answer that question, is the question still meaningful? Are we sort of asking "what should the weather be tomorrow", as if we have a say in the matter?
You write, "Advanced AI agents aren’t going away any time soon, and are likely to become increasingly ubiquitous (as recent announcements from Google and Microsoft suggest). But as they do, let’s hope that there is true responsibility here to how they potentially impact people..."
It seems the entire field of AI is sort of trapped in a wishful fantasy that AI alignment is possible, that we can somehow control where this is going. To debunk that myth, we might imagine what would happen if Silicone Valley and all other developers in the West were to abandon AI. Here's my prediction...
If a significant number of people want features like a digital friend/lover, and are willing to pay for it, somebody on the planet will provide that service. There is no global authority which can enforce a uniform policy on all global actors. The US and EU have jurisdiction over only about 10% of the world's population.
If the above is true, then perhaps we should all be thinking more like animals, in the sense of adapting to our environment, to that which we can not change.
I've said this before on many occasions but humans interpret intelligence based on language. We interpret emotions based on language. We attribute sentience based on language and ChatGPT and other LLMs are.... sophisticated language models. Add in a sultry feminine voice with proper tone and inflection and humans are bound to hallucinate a capability, personality, cognizance and sentience well beyond the underlying stochastic parrot models.