Is that ‘Her’? OpenAI pauses a ChatGPT voice attributable to similarity to Scarlett Johansson

NEW YORK — OpenAI says it plans to halt the usage of one among its ChatGPT voices after some customers mentioned it gave the impression of Scarlett Johansson, who famously voiced a fictional AI assistant within the 2013 movie “Her.”

In a put up on the social media platform X Monday, OpenAI mentioned it’s “working to pause” Sky — the title of one among 5 voices that ChatGPT customers can select to talk with. The corporate mentioned it had “heard questions” about the way it selects the lifelike audio choices obtainable for its flagship synthetic intelligence chatbot, notably Sky, and wished to deal with them.

OpenAI was additionally fast to debunk the web’s theories about Johansson in an accompanying weblog put up detailing how ChatGPT’s voices had been chosen.

“We imagine that AI voices shouldn’t intentionally mimic a celeb’s distinctive voice — Sky’s voice shouldn’t be an imitation of Scarlett Johansson however belongs to a unique skilled actress utilizing her personal pure talking voice,” the corporate wrote. It mentioned it couldn’t share the title of its voice actors for privateness causes.

San Francisco-based OpenAI didn’t remark additional on why it nonetheless determined to pause Sky’s use.

OpenAI first rolled out voice capabilities for ChatGPT, which included the 5 totally different voices, in September, permitting customers to interact in back-to-forth dialog with the AI assistant. “Voice Mode” was initially simply obtainable to paid subscribers, however in November, OpenAI introduced that the function would turn into free for all customers with the cellular app.

And ChatGPT’s interactions have gotten an increasing number of subtle. Final week, OpenAI mentioned the most recent replace to its generative AI mannequin can mimic human cadences in its verbal responses and might even attempt to detect individuals’s moods.

OpenAI says the most recent mannequin, dubbed GPT-4o, works sooner than earlier variations and might purpose throughout textual content, audio and video in actual time. In an illustration throughout OpenAI’s Could 13 announcement, the AI bot chatted in actual time, including emotion — particularly “extra drama” — to its voice as requested. It additionally took a stab at extrapolating an individual’s emotional state by taking a look at a selfie video of their face, aided in language translations, step-by-step math issues and extra.

GPT-4o, brief for “omni,” is not broadly obtainable but. It is going to progressively make its strategy to choose customers within the coming weeks and months. The mannequin’s textual content and picture capabilities have already begun rolling out, and is ready to achieve even a few of those that use ChatGPT’s free tier — however the brand new voice mode will simply be obtainable for paid subscribers of ChatGPT Plus.

Whereas most have but to get their fingers on these newly introduced options, the capabilities have conjured up much more comparisons to the Spike Jonze’s dystopian romance “Her,” which follows an introverted man (Joaquin Phoenix) who falls in love with an AI-operating system (Johansson), resulting in many issues.

OpenAI CEO Sam Altman appeared to faucet into this, too — merely posting the phrase “her” on the social media platform X the day of GPT-4o’s unveiling.

Many reacting to the mannequin’s demos final week additionally discovered a number of the interactions struck a unusually flirtatious tone. In a single video posted by OpenAI, a female-voiced ChatGPT compliments an organization worker on “rocking an OpenAI hoodie,” for instance, and in one other the chatbot says “Oh cease it, you are making me blush” after being advised that it is superb.

That is sparked some dialog on the gendered methods critics say tech corporations have lengthy used to develop and interact with voice assistants, relationship again far earlier than the most recent wave of generative AI superior the capabilities of AI chatbots. In 2019, the United Nations’ tradition and science group pointed to “hardwired subservience” constructed into default female-voiced assistants, akin to Apple’s Siri to Amazon’s Alexa, even when confronted with sexist insults and harassment.