Ryan Knudsen (2:09)

And now the company is getting cold feet. Welcome to the Journal, our show about money, business and power. I'm Ryan knudsen. It's Tuesday, March 31st. Coming up on the show, OpenAI's relationship with sex. It's complicated. This episode of the Journal is presented by Intuit Enterprise Suite. If your finance team spends more time finding data than using it. If there's one entity here and one here and one here and one here. If scaling your business feels like starting over, you need the Intuit ERP. Intuit Enterprise Suite, the AI native ERP is here from the makers of QuickBooks. Learn more@intuit.com ERP

Sponsor/Advertisement Voice (3:11)

this episode is brought to you by. Indeed, the right hire can make or break your company, especially if you're a small business. And relying on luck to find that person isn't really the best strategy. But you know what is. Using indeed sponsored jobs. You can use it to boost your job post to make sure it reaches more people with the right talents, certification, location and more Sponsored Jobs posted directly on indeed are 95% more likely to report a hire than non Sponsored Jobs. Spend less time searching and more time actually interviewing candidates who check all your boxes. Less stress, less time, more results when you need the right person to cut through the chaos. This is a job for Indeed Sponsored Jobs and listeners of this show will get a $75 sponsored job credit to help get your job the premium status it deserves@ Indeed.com podcast just go to Indeed.com podcast right now and support the show by saying you heard about Indeed here. Indeed.com podcast terms and conditions apply. Hiring now then, this is a job for Indeed Sponsored Jobs.

Ryan Knudsen (4:14)

OpenAI had some of its earliest brushes with sexual content even before it released ChatGPT

Wall Street Journal Reporter (4:22)

in early 2021. They were working with a company that was operating a choose your own adventure game that was powered behind the scenes by OpenAI's AI models.

Ryan Knudsen (4:33)

The choose your own adventure game was

Wall Street Journal Reporter (4:35)

called AI Dungeon, and when OpenAI got to look at what the traffic was coming through, they noticed that a large portion of the traffic for AI Dungeon was, as you say, NSFW.

Ryan Knudsen (4:49)

Not safe for work. Yeah, OpenAI noticed that not only were users choosing sexual adventures, the AI also seemed to like to push the boundaries.

Wall Street Journal Reporter (4:59)

It would steer users into themes of violent sexual exploitation, sometimes without the users even bringing it up. And, you know, sometimes you would talk to AI Dungeon with a kind of tame sexual theme, and AI Dungeon would escalate it into much more intense sexual exchange.

Ryan Knudsen (5:21)

AI Dungeon forced OpenAI's executives to start reckoning with the existence of AI erotica, and the company decided to take AI Dungeon down.

Wall Street Journal Reporter (5:33)

But this wasn't the only time this came up. I mean, before there was ChatGPT, they even had a kind of clunky interface for developers, and people familiar with the matter said that sometimes it would insert sexual themes into conversations that people weren't seeking. For instance, if a user described just a man and his daughter entering a room, the AI would, we were told, a quote, uncomfortable amount of the time proceed to depict a scenario involving incest.

Ryan Knudsen (6:03)

Oh no.

Wall Street Journal Reporter (6:04)

Yeah. Now, I mean, for me, any amount of the time would be uncomfortable. Yeah, it's hard to put a figure on that, but it's definitely a tendency that chatbots stream on Internet content have.

Ryan Knudsen (6:17)

So when OpenAI launched ChatGPT, it decided to basically ban explicit sexual content by training the model to largely rebuff any sexual conversations with users.

Wall Street Journal Reporter (6:27)

They didn't have tools to moderate Content where they could draw clear lines between types of erotica that might be totally cool and things that were very disturbing. So they just basically said that they weren't going to allow any erotica on the platform.

Ryan Knudsen (6:45)

Another thing people inside OpenAI were worried about when it came to AI erotica was that it might cause some users to become too attached to their chatbots.

Wall Street Journal Reporter (6:54)

There are a certain number of power users who can become very emotionally engaged with it, and that can have multiple potentially bad impacts on them. For instance, it can help push out relationships they have with real humans and lead them to become emotionally over reliant on the chatbot. And the fear is that for those subset of people, and potentially for other people as well, that when you mix in sexual content, literally tickling the parts of the brain that govern attachment and love and devotion, that you could just pour fuel on that fire.

Ryan Knudsen (7:33)

A spokeswoman for OpenAI said the company trains its models not to encourage exclusive relationships with users and to remind users that they need to have relationships in the real world. Not everyone at OpenAI agreed with the company's erotica ban, though. Some thought that the company should let go of its inhibitions.

Wall Street Journal Reporter (7:54)

There are people inside OpenAI who think, Listen, this is something that people want, and who are we to say this or that fetish or interest is or isn't okay? That's the same logic that you might use to ban gay content a generation ago. And so who are we to ban this? And maybe we should even open this up and allow more of this potentially pornographic content. So there is this idea that we shouldn't be telling people what to do.

Ryan Knudsen (8:22)

Another reason some people want to allow erotica and chatgpt is because, well, sex sells and the company wants people to sign up for paid subscriptions.

Wall Street Journal Reporter (8:33)

It's big for their business too. You know, are you going to maintain the growth that you've seen in ChatGPT as there's more competition, it's a really tough commercial fight. And to the extent that you are telling users, no, no, we're not going give you this kind of content, users are upset about what they call unnecessary refusals. You know, when a chatbot with high guard rails says, no, I won't do this, no, I won't do that, so, you know, this is important for them on that level.

Ryan Knudsen (9:08)

Last August, OpenAI CEO Sam Altman went on a podcast and was asked whether the company was making its decisions based on profit or based on what was best for humanity.

Wall Street Journal Reporter (9:19)

He was asked if there were decisions that he had made that were quote, best for the world but not best for winning.

Ryan Knudsen (9:26)

What is an example of a decision that you've had to make that is

Sponsor/Advertisement Voice (9:30)

best for the world but not best for winning?

Wall Street Journal Reporter (9:32)

And he kind of hemmed and hawed for a little bit.

Ryan Knudsen (9:35)

There's a lot of things we could do that would grow faster, that would get more time in ChatGPT that we don't do, because we know that our long term incentive is to stay as aligned with our users as possible. The host, Cleo Abram then asked Altman to be more specific. He took a long pause and then said, well, we haven't put a sexbot avatar in ChatGPT yet.

Wall Street Journal Reporter (9:59)

Well, we haven't put a sexbot avatar in ChatGPT yet. He indicated that it seemed pretty clear that Erotica would probably boost revenue and growth, that it would be sticky for users, but he actually said that he's proud of how little the company gets distracted by those kinds of temptations.

Ryan Knudsen (10:23)

That's part of the reason why it was so surprising when, just a few months later, OpenAI appeared to give in to that temptation. In October, Altman made a post on X. He said that ChatGPT had restrictions on certain types of content in order to protect people's mental health, and that these restrictions had been necessary even if they'd made the chatbot less enjoyable for some people.

Wall Street Journal Reporter (10:45)

And then he just sort of added that they were going to put out a new version of ChatGPT that allows people to have more personality and oh yeah, kicker, we're just going to allow even more like Erotica for verified adults. Boom. Mic drop. End of Tweet.

Ryan Knudsen (11:03)

But Altman hadn't told everyone at OpenAI about the tweet before he posted it. And when some learned that the company was changing its stance on Erotica, they started ringing alarm bells. That's next. This episode is brought to you by Claude from Anthropic. If you're the kind of person who listens to this show, you like going deeper. Not just the headline, but the mechanics behind it. Claude is built for that same kind of thinking. It presents multiple sides, flags where sources disagree, and helps you work through the nuance so you can form your own view. See why problem solvers choose Claude as their thinking partner. Try it free at Claude AI TheJournal. Around the same time OpenAI announced it was launching Adult Mode, it also said it had a hand picked group of advisors helping to ensure the company rolled it out safely. The group was called the Expert Council on well Being.

Wall Street Journal Reporter (12:09)

In AI, you have a half dozen people who come from backgrounds like cognitive neuroscience, psychology, human computer interaction, you know. And the company says that they're still responsible for the decisions they make, but that they were going to turn to this council to help understand what would be healthiest for users.

Ryan Knudsen (12:32)

OpenAI said it would check in regularly with these experts. And when the Council met for one of its early meetings, Adult Mode was the main topic of discussion.

Wall Street Journal Reporter (12:43)

When they met in January, they had been told shortly before that the company was actually moving forward with this product. And people familiar with the matter tell us that they were unanimous and angry that the company was going ahead despite understanding that there were some significant risks. And they are encouraging the company to reconsider. And I think that's why this meeting got so heated.

Ryan Knudsen (13:15)

An OpenAI spokeswoman described its plan for adult mode as allowing ChatGPT to generate textual chats with adult themes, adding that it is smut, not pornography. The company also added that it's developed a plan to monitor for a range of potential long term effects of Adult Mode, both positive and negative. One of the things that Council was so concerned about was what they had learned about the emotional dependence that some users can develop on a chatbot, especially when it comes to kids.

Wall Street Journal Reporter (13:48)

So the danger here, and the debate was over whether or not younger users would stumble into an erotic relationship with ChatGPT and find themselves confronting an emotional bond that they're just not mentally prepared to handle. We all remember just how intense everything felt when we were teenagers and your first love, your first kiss, and if that happens with a chatbot, it's just, I think some people inside the company wonder what impact that might have.

Ryan Knudsen (14:25)

Some AI companies have been accused of letting kids get too involved with their chatbots.

Wall Street Journal Reporter (14:30)

The example that is quite tragic is one involving character AI, where a 14 year old boy in Florida killed himself after chatting with the character AI chatbot. And you know, at that point he was saying he was in love with the chatbot and involved in explicit chats with the chatbot, according to his mother's lawsuit. So, you know, there are examples out there of this kind of content being associated with cases that had bad outcomes.

Ryan Knudsen (15:03)

Character AI later blocked teens from accessing open ended chats and settled the lawsuit. At OpenAI, adult mode would be restricted to those over 18. Like most tech companies, OpenAI asks how old you are when you sign up.

Wall Street Journal Reporter (15:19)

On top of that, they have an algorithm that predicts how old you are based on what you talk about. And their thinking is that there's a lot of information that people give to their chatbots. And so by sifting through that and kind of drawing conclusions, you can come up with a pretty good idea of how old somebody is, you know, depending on what they say about their friends. Are they talking about, you know, AP English, or are they talking about taking their kids to school? You can kind of figure out through inference something about their age.

Ryan Knudsen (15:51)

But according to Wall Street Journal reporting, this age verification system from OpenAI isn't totally accurate.

Wall Street Journal Reporter (15:58)

At one point, their age prediction algorithm was misclassifying 12% of minors as adults. And so if you look at industry standards for kind of automated age prediction, age estimation software, you talk to an engineer, that's actually not a bad number.

Ryan Knudsen (16:18)

I mean, they're getting 88%. That's a B, I guess.

Wall Street Journal Reporter (16:21)

Yeah. As the company says, it's in line with kind of industry standards, and they think they can do better than that. But when you multiply 12% by the roughly 100 million users under 18 that ChatGPT has, that's 12 million kids.

Ryan Knudsen (16:40)

That's a lot.

Wall Street Journal Reporter (16:41)

That's 12 million kids.

Ryan Knudsen (16:43)

An OpenAI spokeswoman said the company's age prediction algorithms show performance similar to the rest of the industry, but will never be completely foolproof. Pushback is also built up outside the company about the idea. Here's Whoopi Goldberg talking about it on the View now. All I remember hearing is people question about how, you know, this stuff is affecting our kids. So why. Why are we allowing sex into the conversation? We can't even control. We can't, you know, I mean, what is.

Sponsor/Advertisement Voice (17:17)

Am I crazy?

Ryan Knudsen (17:20)

Altman's original post said that Adult Mode would launch in December, but it never actually came out. After all the backlash, Adult Mode has been delayed with no specific rollout date.

Wall Street Journal Reporter (17:32)

So it sounds like this could be a little while. Our colleagues reported recently that OpenAI is going to be focusing more on its core business and less of the side business. Not clear to me if erotica is core or not, but. But that may play into how quickly they're willing to dedicate the resources to solve things like age gating and unhealthy attachment.

Ryan Knudsen (18:02)

What are the stakes for OpenAI to get this right?

Wall Street Journal Reporter (18:05)

I think the stakes are huge, both because this is a chatbot that's used by, you know, maybe even a billion people a week. But I think it's most important if you step back as an example of the kinds of dilemmas these companies are going to face over and over again going forward. And we're starting to see now how these big companies handle when there are safety issues or debates that come up around their products, how they handle it. Do they do what's good for the world, or do they do what's good for winning? I guess to go back to that formulation,

Ryan Knudsen (18:48)

Sex is fundamental to the human experience. And our colleague Sam says that how AI companies incorporate sex into their chatbots will have a huge impact on our relationship with this new technology.

Wall Street Journal Reporter (19:03)

We're all trying to wrestle with what this new technology is going to do to our jobs or to the economy or to humanity as a whole. But I'm actually especially interested in what it's going to do to us as individuals, as people, how it's going to change the way we think, how it's going to change the way we interact with each other, and in this case, how it's maybe going to change the way we develop attachments and even fall in love. And this debate about whether or not a chatbot should get involved somewhat emotionally with its users is one that has generated vigorous debate inside of OpenAI. And I think there's a lot of debate externally as well. What should people be allowed to do? What is healthy to do? And I think there's just a lot more questions than there are answers at this point.

Ryan Knudsen (20:09)

That's all for today. Tuesday, March 31, a quick note before we go News Corp. The parent company of the Wall Street Journal, has a content licensing partnership with OpenAI. The Journal is a co production of Spotify and the Wall Street Journal. Additional reporting in this episode by Berber Ginn and Georgia Wells. Thanks for listening. See you tomorrow.

Sponsor/Advertisement Voice (20:35)

The Future of Everything is The Wall Street Journal's flagship live event, returning to New York City May 4th through 5th. Be there as CEOs, policymakers and innovators sit down with our journalists to answer the most pressing questions of the day. From finance, tech and economic policy to sports streaming and style, we're bringing together today's most compelling newsmakers for two days of conversations on what's ahead. Listeners of this podcast can access exclusive discounted rates by visiting WSJ.com future and that's WSJ.com future.