AI Image Tools Are Creating New Content-Moderation Problems

[00:00:18] [SPEAKER_02]: Welcome to Tech News Briefing. It's Friday, September 6. I'm Zoe Thomas for The Wall Street

[00:00:24] [SPEAKER_02]: Journal.

[00:00:25] [SPEAKER_02]: iPhone sales make up a big chunk of Apple's business, which means the launch of its latest

[00:00:31] [SPEAKER_02]: one is pretty important for the company. We'll take a look at how Apple's dependency on

[00:00:37] [SPEAKER_02]: smartphone sales is impacting Wall Street's expectations ahead of the company's event

[00:00:42] [SPEAKER_02]: on Monday.

[00:00:43] [SPEAKER_02]: And then, AI image generators are driving new questions about content moderation and

[00:00:50] [SPEAKER_02]: how the companies behind these tools should build safeguards. Our reporter Megan Boborowski

[00:00:55] [SPEAKER_02]: will tell us about some of the controversial images created by Google's and XAI's new

[00:01:01] [SPEAKER_02]: tools.

[00:01:03] [SPEAKER_02]: But first, on Monday, Apple is expected to unveil the iPhone 16. There are high hopes

[00:01:10] [SPEAKER_02]: for the phone to help spark new demand. The iPhone has been in a notable slump. iPhone

[00:01:16] [SPEAKER_02]: revenue fell 2% in Apple's fiscal year that ended last September. Here to tell us about

[00:01:21] [SPEAKER_02]: what investors will be looking at is our Heard on the Street columnist Dan Gallagher.

[00:01:26] [SPEAKER_02]: So Dan, put into perspective, how significant are iPhone sales to Apple's business?

[00:01:32] [SPEAKER_04]: Well, very significant. They're more than half of the company's annual revenue. And

[00:01:36] [SPEAKER_04]: that's really been the case for a long time. Apple's whittled that away over the years

[00:01:40] [SPEAKER_04]: a little bit because they built up other big products like Apple Watch, AirPods have done

[00:01:45] [SPEAKER_04]: well, services have done well. So you've seen over the years iPhone dependents has

[00:01:49] [SPEAKER_04]: come from like over 60% of revenue down to like the low 50s. Apple's one of the largest

[00:01:54] [SPEAKER_04]: companies in the world and it's by far the largest that relies on a singular product

[00:01:58] [SPEAKER_04]: line for that percentage.

[00:02:00] [SPEAKER_02]: And how is Apple's stock done this year?

[00:02:03] [SPEAKER_04]: It's done well. Over the last six months, it's been the strongest of the mega cap. It's

[00:02:07] [SPEAKER_04]: even outperformed in video over that particular time slot. Apple's kind of been this late

[00:02:11] [SPEAKER_04]: entry into AI, but there's a lot of excitement about what they might be able to do because

[00:02:15] [SPEAKER_04]: they do have a pretty high degree of success when they enter a new market and do new things

[00:02:19] [SPEAKER_04]: with it. So investors do have some high hopes for what they can do here.

[00:02:23] [SPEAKER_02]: So let's talk a bit about the iPhone 16, which is going to have on-device AI capabilities.

[00:02:30] [SPEAKER_02]: How is that shaping Wall Street's expectations for this latest phone?

[00:02:33] [SPEAKER_04]: Wall Street's actually keeping its expectations somewhat in check for a few reasons. On-device

[00:02:38] [SPEAKER_04]: AI has not yet proven to be a huge reason to sell hardware. Samsung put that into its

[00:02:45] [SPEAKER_04]: latest Galaxy phones. Based on some third-party data, they sold okay. It hasn't been like a

[00:02:51] [SPEAKER_04]: huge seller. But part of what complicates it here is because when they launch these

[00:02:55] [SPEAKER_04]: new phones and they typically go on sale in later September, Apple Intelligence, which

[00:03:00] [SPEAKER_04]: is their AI tool, isn't going to be available yet. That doesn't launch until October. And

[00:03:06] [SPEAKER_04]: not all the capabilities they've been showing off are launching with it. So you're going

[00:03:11] [SPEAKER_04]: to have kind of a limited AI capability early in this particular iPhone cycle. So I think

[00:03:16] [SPEAKER_04]: it's actually going to take time. And Wall Street's estimates indicate that, where there's

[00:03:21] [SPEAKER_04]: more of an expectation that this coming year that we're going to see with the iPhone 16

[00:03:25] [SPEAKER_04]: is going to sell okay, possibly pick up though for iPhone 17, which will presumably have

[00:03:31] [SPEAKER_04]: like a fuller AI capability and story to it.

[00:03:34] [SPEAKER_02]: What are some of the other factors that might impact iPhone sales performance?

[00:03:38] [SPEAKER_04]: The biggest one is that you have a lot of pent up demand. Remember, iPhone is mostly

[00:03:42] [SPEAKER_04]: an upgrade game at this point. Each generation doesn't draw in a huge base of new buyers

[00:03:48] [SPEAKER_04]: from Android. It's usually people that are already on the iPhone and people are holding

[00:03:52] [SPEAKER_04]: onto their devices longer. Apple has just had two to three relatively weak cycles for

[00:03:57] [SPEAKER_04]: the iPhone. So there's a fair amount of the iPhone base out there right now that's using

[00:04:02] [SPEAKER_04]: devices that are three to four years old. And that's about when a lot of customers are starting

[00:04:06] [SPEAKER_04]: to look to upgrade.

[00:04:08] [SPEAKER_02]: And so if we're watching Apple, you know, on Monday after the launch, what should we

[00:04:12] [SPEAKER_02]: expect to see?

[00:04:13] [SPEAKER_04]: This is a unique launch because the big selling point for this iPhone they showed three months

[00:04:17] [SPEAKER_04]: ago at their big developer conference when they debuted Apple Intelligence. Are they

[00:04:21] [SPEAKER_04]: going to show some new stuff about Apple Intelligence that they didn't show three months ago or some

[00:04:26] [SPEAKER_04]: other types of enhancements that maybe add to the appeal of the iPhone 16? I think we'll

[00:04:30] [SPEAKER_04]: be watching for any kind of surprise factors.

[00:04:33] [SPEAKER_02]: That was Dan Gallagher, a columnist for the WSJ's Heard on the Street. Coming up, what

[00:04:38] [SPEAKER_02]: a picture of Mickey Mouse drinking a beer could signal about misinformation risks in

[00:04:44] [SPEAKER_02]: the 2024 U.S. election? That's after the break.

[00:04:50] [SPEAKER_01]: AI may be the most important new computer technology ever, but AI needs a lot of processing

[00:04:55] [SPEAKER_01]: speed and that gets expensive fast. Upgrade to the next generation of the cloud, Oracle

[00:05:00] [SPEAKER_01]: Cloud Infrastructure or OCI. OCI is the single platform for your infrastructure, database,

[00:05:06] [SPEAKER_01]: application development and AI needs. Do more and spend less like Uber, 8x8 and Databricks

[00:05:12] [SPEAKER_01]: Mosaic. Take a free test drive of OCI at oracle.com slash wallstreet, oracle.com slash wallstreet.

[00:05:28] [SPEAKER_02]: Artificial intelligence image generators are testing the boundaries of platforms' policies

[00:05:33] [SPEAKER_02]: and the ability of companies to put effective guardrails around the use of this new powerful

[00:05:39] [SPEAKER_02]: technology. Case in point, Google's Gemini image generator created racially diverse images

[00:05:46] [SPEAKER_02]: of Nazi soldiers. The company suspended the chatbot's ability to produce pictures of people

[00:05:52] [SPEAKER_02]: for six months after that. But last week, Google said it will allow Gemini to generate

[00:05:57] [SPEAKER_02]: images of people again. So does that mean the risks are gone? Here to tell us more is

[00:06:03] [SPEAKER_02]: our reporter Megan Bobrowski. Megan, what is Google's approach to creating safeguards

[00:06:08] [SPEAKER_02]: and dealing with the potential pitfalls of AI image generation tools?

[00:06:13] [SPEAKER_00]: Google has been trying to really build in guardrails and safety mechanisms to make this

[00:06:18] [SPEAKER_00]: work and to not make it show bias. And so you have the issue of its image generator

[00:06:24] [SPEAKER_00]: making historically inaccurate images because it's trying to show diversity and not be biased.

[00:06:31] [SPEAKER_00]: But at the same time, you have it also generating images that are offensive to people. In one

[00:06:38] [SPEAKER_00]: case recently, one of its image generators did generate an image of SpongeBob wearing

[00:06:44] [SPEAKER_00]: a swastika and that offended people. And so you have it on both ends of it. It's trying

[00:06:50] [SPEAKER_00]: to not be too loose, but at the same time, some of its restrictions and some of its attempts

[00:06:55] [SPEAKER_00]: to make its chatbot not show bias are actually leading to what some people perceive as issues

[00:07:00] [SPEAKER_00]: on that side as well. What did Google say about that?

[00:07:04] [SPEAKER_00]: They said that this isn't OK for them to be showing images that are offensive or that people

[00:07:10] [SPEAKER_00]: don't feel comfortable with and that they're going to try to fix these things.

[00:07:15] [SPEAKER_00]: Google is taking a sort of slow and steady approach to try to get this right.

[00:07:19] [SPEAKER_02]: You know, we should say that several companies, including Google and OpenAI, they won't allow

[00:07:24] [SPEAKER_02]: their tools to generate images of specific recognizable people. But the newly released

[00:07:30] [SPEAKER_02]: AI tool from Elon Musk's ex-AI Grok2, it does. What is the position of that company?

[00:07:37] [SPEAKER_00]: So we don't know too much about the decisions that they made on what to restrict because

[00:07:41] [SPEAKER_00]: they didn't respond to our request for comment. But through my own personal experience of

[00:07:46] [SPEAKER_00]: using this and through other people's experience, you can create people, recognizable people.

[00:07:50] [SPEAKER_00]: You can create Donald Trump. You can create Kamala Harris. So it's a bit different from

[00:07:54] [SPEAKER_00]: some of the other companies.

[00:07:56] [SPEAKER_02]: So what are some of the more controversial images people have created with Grok?

[00:08:00] [SPEAKER_00]: There's one that's been circulating around of Mickey Mouse smoking a cigarette, one of

[00:08:05] [SPEAKER_00]: Mickey Mouse drinking a beer. There's ones of Donald Trump and Kamala Harris holding

[00:08:10] [SPEAKER_00]: hands, looking like they're in a relationship. So things that are just clearly not true and

[00:08:16] [SPEAKER_00]: in some cases might be offensive to the people that they're about.

[00:08:20] [SPEAKER_02]: Have ex-AI Musk or Black Forest Labs, which is the company that powers the AI image generation

[00:08:26] [SPEAKER_02]: tool, have they commented about these images?

[00:08:29] [SPEAKER_00]: No. Ex-AI, ex-Elon Musk, Black Forest Labs, none of them responded to requests for comment.

[00:08:36] [SPEAKER_02]: AI-powered image generation tools have been a hot button topic since they came out a

[00:08:41] [SPEAKER_02]: couple years ago. Where do things stand with concerns about deepfakes, especially in the

[00:08:46] [SPEAKER_02]: lead up to the election?

[00:08:47] [SPEAKER_00]: Experts in this field who follow these things are concerned about this. In one case, Donald

[00:08:52] [SPEAKER_00]: Trump, a head of the DNC, posted what appeared to be an AI-generated image of Kamala Harris.

[00:08:59] [SPEAKER_00]: In this image she's speaking to a crowd and there's a red flag with a hammer and sickle

[00:09:04] [SPEAKER_00]: on it in the background and it's implying that she is a communist.

[00:09:09] [SPEAKER_00]: So experts I spoke to said that these image generators might actually be more harmful than

[00:09:15] [SPEAKER_00]: misinformation that's spreading because photos and videos can be more convincing.

[00:09:20] [SPEAKER_00]: From the experts I spoke to, one of the main problems that they were talking about was

[00:09:24] [SPEAKER_00]: this idea of pictures being used as a way to know what's real historically. You might

[00:09:30] [SPEAKER_00]: be familiar with the phrase, pics or it didn't happen, which is a common phrase among Gen

[00:09:35] [SPEAKER_00]: Z and it basically means show me pictures of this or I don't believe you that it happened.

[00:09:40] [SPEAKER_00]: And so now if you can show a picture that's not real, then that makes trying to figure

[00:09:47] [SPEAKER_00]: out the truth all the more difficult.

[00:09:50] [SPEAKER_02]: Did the Trump campaign comment at all about the post depicting Vice President Kamala Harris?

[00:09:55] [SPEAKER_00]: No, his campaign didn't respond to a request for comment about the image that he posted.

[00:10:00] [SPEAKER_02]: What are X's policies on posting fake images?

[00:10:04] [SPEAKER_00]: So X has a slew of policies including one on manipulated media, which says that you

[00:10:09] [SPEAKER_00]: cannot post content that is manipulated or tweaked in some way that is intended to deceive

[00:10:17] [SPEAKER_00]: people or harm people.

[00:10:19] [SPEAKER_02]: Megan, one of the other big issues with image generation has been the question of copyright.

[00:10:24] [SPEAKER_02]: Where does that issue stand now?

[00:10:26] [SPEAKER_00]: So it's unclear whether these companies are allowed to train their image generators on

[00:10:36] [SPEAKER_00]: content that is copyrighted or content that's just taken across the web. And so there's

[00:10:41] [SPEAKER_00]: a few court cases that are pending currently and a law expert that I spoke to said that

[00:10:48] [SPEAKER_00]: these cases will set the precedent for what's allowed and what's okay. But as of right now,

[00:10:55] [SPEAKER_00]: the law doesn't make it clear what these image generators are allowed to do and what they're

[00:10:59] [SPEAKER_00]: not since there's such new technology.

[00:11:02] [SPEAKER_02]: That was our reporter Megan Boborowski.

[00:11:05] [SPEAKER_02]: And that's it for Tech News Briefing. Today's show was produced by Zoe Kolkin. I'm your host

[00:11:11] [SPEAKER_02]: Zoe Thomas. Jessica Fenton and Michael LaValle wrote our theme music. Our supervising producer

[00:11:16] [SPEAKER_02]: is Catherine Millsop. Our development producer is Aisha Al-Muslim. Scott Salloway and Chris

[00:11:21] [SPEAKER_02]: Sinsley are the deputy editors. And Philana Patterson is The Wall Street Journal's head

[00:11:25] [SPEAKER_02]: of news audio. We'll be back this afternoon with TNB Tech Minute. Thanks for listening.