Opinion | What Scarlett Johansson vs. ChatGPT says about AI

This op-ed has been updated.

Sam Altman has promised creators that ChatGPT comes in peace. This week, he showed us what he really thinks.

The OpenAI chief executive recently introduced a new audio chatbot equipped with five “personas” capable of deciphering confusing questions, telling bedtime stories and engaging in plain old small talk. One thing, however, didn’t seem new at all: the voice of the “Sky” persona. It sounded an awful lot like Scarlett Johansson.

Her fans, her friends, her family – even Johansson herself – agreed.

The company claimed that the throaty, flirty voice “belongs to a different professional actress using her own natural speaking voice.” The Post’s later reporting backs up this story. But the thought that the startling similarity was just a coincidence is more fantastical than the robot apocalypse.

Advertisement

Johansson says Altman asked her in September to lend her voice to his soon-to-be released tool. She considered the offer but said no. Two days before the technology’s demonstration, Altman asked again, but before she responded, the system was released.

As if that weren’t enough grounds to cry “thief!” the 2013 Spike Jonze film “Her” features Johansson as a super-intelligent AI assistant called “Samantha” who – or that – enamors a lonely and love-starved Joaquin Phoenix. (Spoiler: The affair ends poorly.) And there’s more: Altman counts the flick as his favorite movie.

Johansson’s lawyers have fired off a letter threatening ChatGPT with legal action, and the company has put access to Sky on pause. It maintains it did nothing wrong.

There are many methods by which OpenAI could conceivably have crafted Sky. The worst? Training its system on snippets of Johannson’s professional or otherwise public output. The best, and nearest to OpenAI’s narrative? Picking a different actress with style and syntax eerily similar to a movie robot who, it turns out, is romancing thousands of oblivious users at the same time. None of the imaginable in-betweens — synthetic voice manipulated to imitate Johannson; real voice altered by AI to resemble a Hollywood-generated male fantasy — flatters OpenAI, either.

Advertisement

When ChatGPT first emerged, the key question was about what the new technology would mean for blue- and white-collar workers who perform the kinds of rote tasks that AI could likely learn to do far more efficiently. Next came worries about the effects on more specialized areas, such as writing legal briefs or research papers.

The Johansson saga taps into something perhaps more visceral: a fear that large language models such as ChatGPT will take the very thing that defines us as human: our creativity. Until now, our ability to dream something up in our minds or souls and express it in the real world was unique to us. This sentiment motivated the Writers Guild of America strike that plagued studios all last summer — or, less noticed, the uprising among fanfiction authors when they discovered their opuses riffing on popular works were included in massive AI training data sets.

The makers of AI technology understandably want to assure our anxious species that these fears are unwarranted. So they’ve been saying they have no intention of hoovering up human creations to turn AI systems into hyper-optimized creators in their own right: OpenAI said on Monday that “we believe that AI voices should not deliberately mimic a celebrity’s distinct voice.” Altman has said before that AI shouldn’t be anthropomorphized at all. But what they are doing, it seems, is just the opposite.

Advertisement

By so brazenly borrowing Samantha to build Sky, OpenAI proved not only that it can indeed replicate a celebrity’s distinct voice but that it will — with or without consent. Either OpenAI’s leaders lacked even the awareness of the reaction they’d elicit or they didn’t care. Altman was supposed to send the message that AI is for humans to harness, not to be harnessed. Instead, he sent the message that our nightmares are about to come true.

As a technology we don’t and can’t understand encroaches on daily life, we’re relying on the integrity of the people who are building it to do so for our benefit. Historically, this has been a bad bet: Silicon Valley’s darlings such as Facebook and Google, after all, proved in scandal after scandal to be more devoted to their bottom lines than their lofty missions of improving the world.

Worse still, dystopia is exactly what Altman’s beloved film depicts. He told on himself in more ways than one last week when he tweeted a single word: “her.”

ncG1vNJzZmivp6x7uK3SoaCnn6Sku7G70q1lnKedZLyxtc2ipqerX2d9c4COaWxoamJkwKSt0aWcraxdn7yprc2sqqimXZi1osDGqatmmZypuqK6jJqgZqubrnqpsdFo