There Are No Girls on the Internet - Scarlett Johansson’s Open AI voice fight shows the need for consent in tech

00:00:00.000 This is an iHeart Podcast.

00:00:02.580 Guaranteed human.

00:00:05.140 Get Bell Pure Fiber Internet with Crave, Netflix, and Disney Plus

00:00:08.720 from $94 a month.

00:00:10.160 Price guaranteed for two years on Internet

00:00:12.040 with a two-year term and auto-pay credit.

00:00:14.400 Visit bell.ca for details and to check availability.

00:00:17.180 Bell. Connection is everything.

00:00:19.620 Run a business and not thinking about podcasting?

00:00:21.880 Think again.

00:00:22.520 More Americans listen to podcasts

00:00:23.980 than ad-supported streaming music from Spotify and Pandora.

00:00:26.680 And as the number one podcaster,

00:00:28.060 iHeart's twice as large as the next two combined.

00:00:30.820 Learn how podcasting can help your business.

00:00:32.640 Call 844-844-IHEART.

00:00:34.920 If your bookshelf and your For You page

00:00:36.860 are equally important to your personality,

00:00:39.100 welcome home.

00:00:40.700 Pro Society is a weekly podcast

00:00:42.560 that's part book club, part group chat

00:00:44.480 for anyone who thinks Pride and Prejudice

00:00:46.180 and Love Island deserve the same level of discourse.

00:00:48.920 Each week, we're connecting the dots

00:00:50.360 between books, the internet, and pop culture

00:00:52.460 with your favorite writers, book talk creators,

00:00:54.980 and plenty of overthought opinions.

00:00:57.080 Yeah, I'm obsessed. I'm obsessed.

00:00:59.020 Listen to Pro Society on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.

00:01:08.860 There Are No Girls on the Internet is a production of iHeartRadio and Unbossed Creative.

00:01:16.960 I'm Bridget Todd, and this is There Are No Girls on the Internet.

00:01:19.640 Welcome to There Are No Girls on the Internet, where we explore the intersection of technology, social media, and identity.

00:01:28.420 And all of those things intersected in a major way this week with Scarlett Johansson and open AI.

00:01:34.680 So let's talk about it.

00:01:35.580 Full disclosure, I was already working on putting together an episode re-watching Spike Jonze's 2013 movie Her, starring Scarlett Johansson's voice as an AI assistant.

00:01:46.560 I really wanted to compare and contrast what the movie thought AI integration with our life would be like and what it actually has been like 10 years later.

00:01:54.360 I'm really excited that the movie Her is part of the public conversation right now because it's one of my favorite movies.

00:02:00.320 If you haven't seen it, I don't want to give too much away, but Scarlett Johansson is the voice of Joaquin Phoenix's AI software.

00:02:07.480 The movie imagines a future where AI is less like Siri and more like a real human.

00:02:12.780 People in the Herd universe fall in love with AI.

00:02:16.080 They have friendships and real meaningful relationships with AI.

00:02:19.900 And that's partly because AI sounds like a real human person who speaks to you and behaves like a person would, not like a robotic voice.

00:02:28.900 And as I was preparing for that episode, the whole thing with Scarlett Johansson really blew up.

00:02:34.960 And the more I thought about it, honestly, the madder I got.

00:02:38.280 Last night, I was getting ready for bed and I was sort of angrily brushing my teeth.

00:02:42.020 And I found myself thinking about this yet again.

00:02:45.120 And the kind of chorus in my mind that I kept saying over and over to myself was that these tech guys just think they own whatever woman they want.

00:02:54.200 Because to me, this is not even really about Scarlett Johansson.

00:02:57.820 It is about what happens when consent in technology is violated again and again and again.

00:03:02.980 And how it erodes the trust that we should be able to count on being at the center of our tech experiences.

00:03:08.700 and how it reinforces that the most powerful companies in our world who are shaping our

00:03:14.360 collective futures consistently demonstrate that they cannot be trusted to simply respect people,

00:03:20.080 especially when those people are women. Okay, so here's what's going on. OpenAI, 0.94

00:03:25.300 the company that makes ChatGPT and a major player in the AI space, has been flirting with integrating

00:03:31.040 voice technology to ChatGPT since around last year. But last week, OpenAI finally revealed a

00:03:37.180 new conversational interface for ChatGPT that they called Sky. Yep, just like a lot of voice

00:03:43.720 technology, Sky has the voice of a woman. But Sky also has a voice that is really similar to the

00:03:49.580 one that Scarlett Johansson used to play the AI assistant called Samantha in the movie Her. But

00:03:55.120 then OpenAI suddenly disabled this feature over the weekend. Grand opening, grand closing. And

00:04:01.500 this comes after OpenAI's head, Sam Altman, who you might remember we made an episode about

00:04:06.980 he was fired for something. We don't totally know what, but it seemed to be related to his

00:04:11.780 lack of honesty. And then he was rehired and is now basically doing whatever the hell he wants.

00:04:17.020 Well, Sam Altman was talking up this integration and comparing it to the movie Her and talking

00:04:21.960 about how we'd finally have AI that felt like a real human that you could be friends with,

00:04:26.100 which is a plot line right out of the movie. Which, spoiler alert, I do think that some of

00:04:30.460 these tech geniuses might actually be low-key misunderstanding the takeaway from the movie.

00:04:36.940 But anyway, so shutting down this new voice technology after Sam Altman was driving so much anticipation about it, everybody, myself included, was like, what is going on?

00:04:47.380 Like, what's the story there?

00:04:48.420 So then on Monday, we get the real tea, which is that Scarlett Johansson told Wired in a statement that OpenAI actually reached out to her to ask her to be the actual voice of their new conversational interface.

00:05:01.560 And she declined twice.

00:05:03.520 And that OpenAI basically just used her voice anyway, or at least a voice that sounds a lot like her voice.

00:05:10.760 And OpenAI's Sam Altman even tweeted a reference to her work in the movie Her when announcing that new chat-JPT voice interface.

00:05:18.720 So there isn't really a ton of plausible deniability on his part even.

00:05:22.380 Okay, so this is what Sky, OpenAI's, not Scarlett Johansson's, voice integration sounds like.

00:05:28.740 I don't have a personal name since I'm just a computer program created by OpenAI, but you can call me assistant. What's your name?

00:05:38.100 And here is Scarlett Johansson as the voice of the AI Samantha from the movie Her.

00:05:43.700 Well, right when you asked me if I had a name, I thought, yeah, he's right. I do need a name, but I wanted to pick a good one.

00:05:49.120 So I read a book called How to Name Your Baby, and out of 180,000 names, that's the one I like the best.

00:05:54.220 Wait, you read a whole book in the second that I asked you what your name was?

00:05:57.300 In two one hundredths of a second, actually.

00:06:00.200 Wow.

00:06:01.540 It sounds pretty similar to me, and ScarJo agrees.

00:06:06.200 Here's what she told Wired in a statement.

00:06:08.340 Last September, I received an offer from Sam Altman, who wanted to hire me to voice the current ChatGPT 4.0 system.

00:06:15.440 He told me that he felt by my voicing the system, I could bridge the gap between tech companies and creatives

00:06:21.220 and help consumers to feel comfortable with the seismic shift concerning humans and AI.

00:06:26.480 He said he felt my voice would be comforting to people.

00:06:29.240 After much consideration, and for personal reasons, I declined the offer.

00:06:33.460 Nine months later, my friends, family, and the general public all noted how much the

00:06:38.240 newest system named Sky sounded like me.

00:06:41.620 When I heard the release demo, I was shocked and angered and in disbelief that Mr. Altman

00:06:46.280 would pursue a voice that sounded so eerily similar to mine that my closest friends and

00:06:51.160 news outlets could not tell the difference.

00:06:53.660 Mr. Altman even insinuated that the similarity was intentional,

00:06:57.800 tweeting a single word, her,

00:06:59.800 a reference to the film in which I voiced a chat system, Samantha,

00:07:02.980 who forms an intimate relationship with a human.

00:07:05.720 Two days before the ChatGPT 4.0 demo was released,

00:07:09.440 Mr. Altman contacted my agent, asking me to reconsider.

00:07:13.080 Before we could connect, the system was out there.

00:07:16.300 As a result of their actions, I was forced to hire legal counsel,

00:07:20.120 who wrote two letters to Mr. Altman and OpenAI,

00:07:22.740 setting out what they had done

00:07:24.420 and asking them to detail the exact process

00:07:26.880 by which they created the sky voice.

00:07:29.100 Consequently, OpenAI reluctantly agreed

00:07:31.200 to take the sky voice down.

00:07:33.200 In a time where we are all grappling with deepfakes

00:07:35.500 and the protection of our own likeness,

00:07:37.560 our own work, our own identities,

00:07:39.900 I believe these are questions

00:07:41.240 that deserve absolute clarity.

00:07:43.260 I look forward to resolution

00:07:44.420 in the form of transparency

00:07:45.540 and the passage of appropriate legislation

00:07:47.840 to help ensure that individual rights are protected.

00:07:51.260 So I really applaud Johansson here.

00:07:53.420 And I think this is the first time

00:07:54.760 that there has been a legal dispute over a soundalike

00:07:57.080 that is, as far as we know, not AI-generated.

00:08:00.700 And I think it could set a precedent

00:08:02.300 for this kind of thing going forward,

00:08:04.120 especially for voice actors and creative professionals

00:08:06.960 who can't afford lawyer's fees or a big lawsuit

00:08:09.900 if their likeness or voice is used this way

00:08:12.480 without their consent.

00:08:13.740 Her statement is also just a good reminder

00:08:15.540 that Johansson has been here before.

00:08:17.840 She is one of the most targeted celebrity figures

00:08:19.980 for AI deepfaked images. 0.51

00:08:22.140 So finding out that OpenAI actually asked Scarlett Johansson

00:08:26.000 to work on this twice, and when she said no,

00:08:29.120 they just found a sneaky workaround to do it anyway,

00:08:32.260 enrages me.

00:08:33.720 It enrages me as a voice professional,

00:08:35.840 it enrages me as a creative,

00:08:37.640 and it enrages me as a woman.

00:08:39.760 You know, when I say on the show

00:08:41.240 that the exploitation of women is baked into technology 0.85

00:08:44.260 in a lot of ways from the ground up,

00:08:46.100 that these are features and not bugs,

00:08:47.960 this is a great example of what I mean. It matters that a company like OpenAI would build their

00:08:53.940 anticipated voice system in a way that has the exploitation of a woman baked into its earliest

00:08:59.280 foundation. And this is not happenstance. It colors how they see women and other marginalized

00:09:05.060 people as just available to take from in service of them making money to create their vision,

00:09:11.020 a vision that by design ignores and exploits us. Like, don't these people understand that no

00:09:17.560 means no? I should say that OpenAI says that they did not actually steal her voice. But I also want

00:09:23.840 to say that I 100% do not believe them at all. Here's OpenAI's statement. We support the creative

00:09:29.760 community and worked closely with the voice acting industry to ensure we took the right steps to cast

00:09:34.980 ChatGPT's voices. Each actor receives compensation above top of market rates, and this will continue

00:09:41.020 for as long as their voices are used in our products. We believe that AI voices should not

00:09:45.660 deliberately mimic a celebrity's distinct voice. Skye's voice is not an imitation of Scarlett

00:09:50.280 Johansson, but belongs to a different professional actress using her own natural speaking voice.

00:09:55.400 To protect their privacy, we cannot share the names of our voice talents.

00:09:59.160 So here's my opinion about what's actually going on. I believe that they probably did work with a

00:10:04.840 human voice actor, and they probably intentionally picked a voice actor that sounded a lot like

00:10:10.140 Scarlett Johansson. And I think they had this person ready to go, whether or not Scarlett

00:10:15.080 Johansson agreed to do this or not. Like, I don't think they really cared about actually having

00:10:19.740 Scarlett Johansson's permission, and they were going to either use this soundalike or use Scarlett

00:10:25.020 Johansson's real voice. Because in addition to his single word, her tweet, Sam Altman, the head of

00:10:30.480 OpenAI, also said that the new AI voice technology, quote, feels like AI from the movies. OpenAI's

00:10:37.620 chief technology officer, Mira Mirati, said that that was all a coincidence. But even still,

00:10:43.240 it's like they want to have it both ways.

00:10:45.580 They obviously want us, the public,

00:10:47.780 to be associating their new technology

00:10:49.920 with the AI in the movie Her.

00:10:52.520 And they're clearly trying to capitalize on that

00:10:55.240 for this rollout.

00:10:56.800 But they want to have all of that

00:10:58.900 and benefit from all of that

00:11:00.280 without actually having the consent 0.94

00:11:02.500 from the real human woman

00:11:04.140 behind the voice in the movie that they're referencing.

00:11:07.020 As Bethany Frankel might put it,

00:11:08.940 it is a cheater brand.

00:11:13.240 Let's take a quick break.

00:11:20.320 Get Bell Pure Fiber Internet with Crave, Netflix, and Disney Plus from $94 a month.

00:11:25.380 Price guaranteed for two years on Internet with a two-year term and auto pay credit.

00:11:29.620 Visit bell.ca for details and to check availability.

00:11:32.400 Bell, connection is everything.

00:11:34.900 Run a business and not thinking about podcasting?

00:11:37.580 Think again.

00:11:38.560 More Americans listen to podcasts than ad-supported streaming music from Spotify and Pandora.

00:11:42.860 And as the number one podcaster, iHeart's twice as large as the next two combined.

00:11:47.400 So whatever your customers listen to, they'll hear your message.

00:11:50.380 Plus, only iHeart can extend your message to audiences across broadcast radio.

00:11:54.600 Think podcasting can help your business?

00:11:56.480 Think iHeart.

00:11:57.520 Streaming, radio, and podcasting.

00:11:59.600 Call 844-844-iHeart to get started.

00:12:02.740 That's 844-844-iHeart.

00:12:05.200 If your bookshelf and your For You page are equally important to your personality, welcome home.

00:12:11.120 This is Prose Society, the weekly podcast that's part book club, part group chat for thought daughters, pop culture obsessives, and anyone who thinks Pride and Prejudice and Love Island deserve the same level of discourse.

00:12:23.600 I'm Eli Rallo, and every week we're connecting the dots between books, the internet, and the conversations everyone can't stop having.

00:12:30.700 I'm going to have to look up this story.

00:12:32.600 I'm obsessed.

00:12:33.340 So obsessed.

00:12:33.840 From best-selling authors and your favorite book talk creators to the latest pop culture moments, nothing is off the table.

00:12:40.240 It's like if you can hide some real messages inside compelling characters, then that is a Trojan horse.

00:12:48.140 Whether you're looking for literary deep dives, smart pop culture conversations, or a community of readers who love to think a little too much, you're in the right place.

00:12:57.320 Listen to Pro Society on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.

00:13:03.340 See you between the pages.

00:13:10.560 And we're back.

00:13:13.980 So I'm not an attorney, so I can't really speak to the legal frameworks at play here.

00:13:17.960 But I do think it is a good opportunity to talk about consent in technology.

00:13:22.260 So in my opinion, looking at the facts and the timeline, I believe OpenAI picked a voice actor that sounded a lot like Scarlett Johansson to be able to complete this vision that they had been hyping up that merges creativity and technology for their voice tech.

00:13:37.920 And I think they were betting that legally they were in the clear because they were using a soundalike. 0.86

00:13:43.100 But beyond what might happen in a lawsuit, to me, this is the classic story of someone trying to cover their ass after they weaseled out of getting consent.

00:13:52.680 Like the question that OpenAI is asking is probably, would we lose a lawsuit? 0.88

00:13:57.280 The questions that they're not asking is, is this ethical?

00:14:00.480 Are we crossing the boundaries of what this person said they don't want to do?

00:14:03.960 Or questions like, are we further damaging the trust between the technology that we make and the people we want using it?

00:14:10.200 And these are important questions that they should be asking because once that trust is damaged, it is not easy to win back. 0.54

00:14:16.640 And the entire thing makes me incredibly uncomfortable because I can't not see that we're talking about a woman whose body is being used to financially benefit people who are mostly men.

00:14:27.200 And these people who are mostly men have appointed themselves the architects of our future and have shown themselves to be the kind of people who are not interested in clear, affirmative consent.

00:14:38.640 I can't help but think that they feel as entitled to the body and likeness of a woman as they do to everyone's data and information online that they use to train their AI without consent for their own profit.

00:14:50.380 It's about entitlement.

00:14:52.140 And why would anybody watching trust these people?

00:14:55.100 If OpenAI thinks, like their statement says, that this was done working with the creative

00:14:59.820 voice talent industry and meant to be done in a respectful way, I think it's fair to

00:15:04.560 say that as an industry, we are not buying it.

00:15:07.860 We're smart enough to see these tricks and know that if this is how you're moving, that

00:15:12.060 none of us should trust you.

00:15:13.480 And I think that OpenAI really needs to understand and internalize why we are all so skeptical.

00:15:20.140 It's because of their own behavior.

00:15:22.220 This is a natural consequence of their actions.

00:15:25.300 Scores of writers and artists and creative professionals

00:15:28.200 have already had to sue OpenAI specifically

00:15:31.440 for using their intellectual property to train their AI

00:15:34.720 without their permission.

00:15:36.140 And a big bulk of the Hollywood writer strikes

00:15:38.960 was about AI stealing the work of creative professionals

00:15:41.960 and then being used to replace

00:15:44.120 those same creative professionals.

00:15:46.340 Like, OpenAI doesn't get to act all shocked

00:15:48.900 that people are then naturally skeptical of their intentions.

00:15:53.180 And there just is a better way.

00:15:55.600 On the podcast I make with the Mozilla Foundation called IRL

00:15:58.880 that explores ethics in AI,

00:16:01.360 we spoke to artists Matt Dryhurst and Holly Herndon

00:16:04.200 who are building AI consent systems at haveibeentrained.com

00:16:08.840 that allows people to search popular AI training datasets

00:16:11.820 to see if their work shows up in them

00:16:13.660 and makes it easy for people to opt out

00:16:16.040 of having their work used to train AI models.

00:16:18.900 Holly's a musician, and she built this publicly available version of her own voice that anybody can play with.

00:16:24.880 So they're both big advocates of what can be created using AI when consent is at the forefront.

00:16:31.200 In a piece for Art Review called AI Art and the Problem of Consent,

00:16:35.260 they underscore the importance of the intersection of AI and creativity being grounded in consent, writing,

00:16:41.100 My hope is that a new era of abundant media will affirm this social value of art and artists.

00:16:46.360 And this is why establishing new protocols of consent is so vital.

00:16:50.580 I am a deep believer in AI augmented expression and have no desire to limit experimentation

00:16:55.640 or enshrine outdated IP laws.

00:16:58.160 I would simply like to avoid what a world looks like in the absence of consent.

00:17:02.480 When consent is absent, beautiful relationships and connections are stymied that could instead

00:17:07.580 have been nurtured.

00:17:08.700 I hope the dress rehearsal of an AI art future we are witnessing highlights what an opportunity 0.96

00:17:13.260 we have to finally fix those damned broken links. 0.74

00:17:16.360 And I think that's exactly what we have here. 0.95

00:17:18.820 Trust has been eroded, consent has been violated, and those links have been broken.

00:17:23.760 So what's next?

00:17:25.140 Well, ScarJo is not backing down, and Homegirl is notoriously not afraid of a lawsuit.

00:17:31.240 This is the woman who mounted a legal contract dispute against Disney, a notoriously litigious company, and walked away with a settlement.

00:17:40.240 So I don't see Scarlett Johansson's team backing off without a fight.

00:17:44.100 OpenAI did pause the use of Sky and apologize directly to Johansson, saying,

00:17:49.840 we are sorry to miss Johansson that we didn't communicate better.

00:17:52.860 And I want to get into all of this more in our episode, Breaking Down Her,

00:17:56.560 but even the conversation about them wanting to use Scarlett Johansson's voice

00:18:00.620 tells me a lot about how OpenAI is thinking about technology in this moment.

00:18:05.480 It's like they want to give us the illusion that they're finding thoughtful ways to link creativity,

00:18:10.580 connection, and technology in our futures, which is why they're trying to reference Johansson's

00:18:14.940 voice from her, a movie that is at its core about sad, lonely, creative people surrounded by

00:18:21.000 technology and searching for connection. But it's just that, like a shallow, empty movie reference.

00:18:27.160 Open AI cannot actually give us connection or community. The best they can do is a movie

00:18:32.620 reference that uses a smooth, stolen voice to kind of mimic those things if you don't look at it too

00:18:37.980 closely. Sam Altman told Scarlett Johansson that using her voice would give people comfort and ease

00:18:44.880 around this technology. People who maybe were a little bit anxious about how quickly AI technology

00:18:50.680 has proliferated. And it's like they can't actually sort out a way to make technology that gives

00:18:55.900 people a genuine reason to feel comfort about it. They can only offer a quick workaround. And worse,

00:19:02.860 in trying to reference this, tech leaders are showing that they think it's okay to do all of

00:19:07.720 this without a strong foundation of consent. They think they have rights to whatever woman's body

00:19:13.200 or whatever woman's voice they want without permission. And none of this bodes well for

00:19:18.560 the kinds of futures they are going to use technology to try to create. And if these

00:19:23.020 are the people who have appointed themselves to build all of our futures, then for everyone's

00:19:27.360 sake, they have got to be better. Got a story about an interesting thing in tech or just want

00:19:37.040 to say hi? You can reach us at hello at tangody.com. You can also find transcripts for today's episode

00:19:42.060 at tangody.com. There Are No Girls on the Internet was created by me, Bridget Todd.

00:19:46.420 It's a production of iHeartRadio and Unbossed Creative. Jonathan Strickland is our executive

00:19:50.280 producer. Tari Harrison is our producer and sound engineer. Michael Amato is our contributing

00:19:54.960 producer. I'm your host, Bridget Todd. If you want to help us grow, rate and review us on Apple

00:20:00.000 Podcasts. For more podcasts from iHeartRadio, check out the iHeartRadio app, Apple Podcasts,

00:20:05.080 or wherever you get your podcasts.

00:20:19.040 If your bookshelf and your For You page

00:20:21.340 are equally important to your personality,

00:20:23.560 welcome home.

00:20:25.140 Pro Society is a weekly podcast

00:20:27.020 that's part book club,

00:20:28.280 part group chat for anyone

00:20:29.360 who thinks Pride and Prejudice

00:20:30.620 and Love Island

00:20:31.320 deserve the same level of discourse.

00:20:33.440 Each week, we're connecting the dots

00:20:34.800 between books, the internet, and pop culture

00:20:36.920 with your favorite writers, book talk

00:20:38.980 creators, and plenty of overthought

00:20:40.920 opinions. Yeah, I'm obsessed, I'm obsessed.

00:20:43.460 Listen to Pro Society on

00:20:44.920 the iHeartRadio app, Apple Podcasts,

00:20:47.180 or wherever you get your podcasts.

00:20:49.500 I survived

00:20:50.820 nine months in captivity,

00:20:52.740 and I've spent my life exploring how

00:20:54.940 other people survive what should have

00:20:56.800 destroyed them. I'm Elizabeth

00:20:58.820 Smart, and these are the Survivor

00:21:00.920 Files. Every week, I'm

00:21:02.880 with survivors who lived through the unthinkable,

00:21:05.540 abducted, stalked, controlled, and nearly silenced.

00:21:09.360 These are stories about what it takes to make it out alive.

00:21:13.200 Listen to The Survivor Files with Elizabeth Smart

00:21:15.980 on the iHeartRadio app, Apple Podcasts,

00:21:19.000 or wherever you get your podcasts.

00:21:22.900 Hey, this is Hayes Davenport.

00:21:24.840 And Sean Clemens.

00:21:26.040 We host the podcast Hollywood Handbook.

00:21:28.100 Every episode, we're trying to help our guests

00:21:29.820 improve their careers by mainly focusing

00:21:31.600 on how they can help us improve our careers.

00:21:34.120 The show's famously super accessible,

00:21:35.820 so you can easily jump into any of our 650 episodes

00:21:38.680 and understand what's going on.

00:21:40.260 We recommend some of our recent episodes

00:21:41.760 with Ben Stiller, Danny McBride, Riz Ahmed, and Mary Steenburgen.

00:21:45.200 Listen to Hollywood Handbook on the iHeartRadio app,

00:21:48.000 Apple Podcasts, or wherever you get your podcasts.

00:21:50.980 This is an iHeart Podcast.

00:21:53.860 Guaranteed human.

There Are No Girls on the Internet - May 22, 2024

Scarlett Johansson’s Open AI voice fight shows the need for consent in tech

Episode Stats

Transcript