Last week, OpenAI presented the latest update of ChatGPT: the chatbot can now "speak" and "converse" with users in various voices, complete with "human" filler words ("um", "uh", etc.), pauses, simulated breathing, laughter, and giggles. This new version of the chatbot sounded exactly like Scarlett Johansson, who voiced an AI about 10 years ago for Spike Jonze's film "Her," a melancholic character study about the consequences of emotional attachments to AI systems. During the presentation, OpenAI CEO Sam Altman posted a single word: "Her."
Today, Scarlett Johansson released a statement announcing legal action, and OpenAI subsequently removed the "Sky" voice option from ChatGPT. It’s all a misunderstanding and that the resemblance to Scarlett Johansson's voice was unintentional. Sure thing. As Johansson's statement reveals, OpenAI and the actress had been in negotiations for months, and just two days before the presentation, Altman tried to persuade the actress to lend her voice to his "Her" version. Johansson refused, and OpenAI still used a voice that was extremely similar to hers.
OpenAI using the voice of maybe the most famous actress in the world who also portraied the voice of an AI in a (misunderstood) movie about AI is not a coincidence.
In a recent piece at Aeon, Bryan Norton wrote about the philosophy of Bernard Stiegler, who differentiated the terms "technics" and "technology" to show how our fascination with the surface of technology — the shiny gadgets, the colorful interfaces, the sleek designs, and today: the seductive synthesized voices of famous actresses — obscures our actual relationship to the tools and their true impact on us as humans in a societal social fabric.
Stiegler developed the concept of "pharmacology", according to which new media technologies affect us similarly to pharmaceuticals, potentially having both beneficial and harmful effects. Stiegler's notion of technics points to these "pharmaceutical" effects of technology: the "side effects" that lie beneath the surface of interfaces and mere application results.
Sam Altman gave the latest version of ChatGPT the voice of a famous actress with the explicit goal of realizing the fictional chatbot from Spike Jonze's film. The question now is, in Stiegler's terms, what "pharmacological" effects lurk behind this Scarlett Johansson interface. Two things come to mind.
Firstly, a recent study confirmed the psychoactive effect of AI chatbots: an experiment with over 2000 conspiracy theory believers showed that a conversation with a chatbot could reduce their beliefs by about 20%. This sounds great first, but it essentially means that, yes, indeed: a chatbot can influence people to change fundamental attitudes. AI systems are actually persuasive apparently — and using the familiar voice of Scarlett Johansson creates a direct connection between the auditory, now also Hollywood-glamorous interface surface of ChatGPT technology, and the seductive power of the speech-simulating technics beneath.
(This is why I signed the open moratorium letter a year ago: I don’t care much about alleged AI doomsday scenarios. But the seductive power of mimetic and anthropomorphic AI systems, which suggest that there’s “personality” behind simulated language, and the unpredictable psychological consequences if these "personalities" lead to a synthetic theory of mind, where AI systems occupy a real and effective place as "social actors" in our minds — this worries me more than any robotic paperclip-juggling AI. Another similar example would be the psychoactive effect of romantic chatbots, which fulfill all our desires on command, potentially creating unpredictable and unfulfillable expectations in real human relationships.)
By (non-consensually) using the voices of famous actresses and making them emphatically "human" (with filler words and ohs and ahs and giggles), OpenAI is deliberately trying to get "under the skin" of its users (to quote the title of another film starring Scarlett Johansson). With this AI-mimetic seduction tactic, OpenAI obscures both the psychoactive effect of its product and the fact that we pay dearly for this synthetic seduction-by-chatbot in the form of gigantic carbon footprints and a water consumption from data centers that increased by 30% for Microsoft in 2022 alone.
These are just some of the possible and actual effects of ChatGPT's technics, which OpenAI's techno-mimetic masks are already obscuring today. More will follow.