When Apple Intelligence is released later this year, it will be unlike any other product Apple has ever put out. Because Apple has to do something. But they never will. They have to give up control. It's generative AI, and Apple has control over every aspect of the user's experience and the words and images the machine spits out. It might produce great results, but it might not be on point or exactly what the user wanted. And Apple knows that. The company said so in a study and released it. I don't think I've ever been more excited to try a new Apple product because there are so many unknowns. What can I say? Chaos is fun, but it's not all mystery. This week, we got a sneak peek at Apple Intelligence in action and learned some things that weren't told in the keynote. Another story this week is all the little things we didn't know before and why it might be cool for Apple to give up some of the control. I'm Bridget Carey. Another thing, Apple Intelligence isn't something you can try just yet. The developer beta of iOS 18 is out, but it doesn't have AI. Apple Intelligence will be released as a beta in the fall, and not all features will be released right away. It's a beta, so of course things could change from what I saw. But what I learned at least gives me some understanding and how it will shape up. I think Apple's approach gives us some freedom. But at the same time, it also has some protections. Let's start with the feature that users jump on first: the ability to create crazy cartoon images. With Gen Moji, you enter a description of the emoji you want to see, and it becomes a reality. You'll be presented with several options. You can pick the one you like the most and set it in the body of your message, send it as a sticker, or send it as a reaction to tap to reply. There's also Image Playground, and this tool is in apps like Notes, but it also has a dedicated app. This tool suggests what you want the image to be, based on the context of the conversation. But in the example Apple gives, you can give the machine some input to tell it what kind of image you want to create. Three descriptive words are often used to create an image, but sometimes a bit more than three. You can choose from general categories, or be a bit more specific about what you want in an image. For example, a landmark like the Golden Gate Bridge, or even the Statue of Liberty. So what about people? You can model an image of a real person on it, as long as it's an established person in your library. If you don't have that person in your photo library, you can generate a generic character. Just choose the gender, skin color, and hair, like a video game avatar. So what does all this tell us? Well, maybe there are intentional obstacles in the design to make it easier to troll people who aren't famous in your own photo library. Also, take a look at all the example images Apple lists. They're all cartoony. They're not that serious, and they're not trying to fake reality or trick people. All the examples Apple gives are happy, but you can express other emotions. You can make sad or angry ones. I'm sure a lot of you are immature and will try to make awful images, because that's what humans do. But Apple thought that there would be filters to protect against certain images being made that could be harmful. For example, nudity is a borderline. Surely Apple has to strike a balance between giving people creative freedom and allowing people to do horrible things. Yes, there is a control issue here too. Now, Apple doesn't think that even if they protect people with filters, they can stop people from using it in weird ways. Because the eggplant emoji became the eggplant emoji because people had no other option to communicate what they wanted to communicate, and people make anything into a conference. For example, if you say the word “gun” on TikTok, your video will be censored. So what will people say instead? They call it the “mouse co tool.” That way, people will try to create their own meaning from certain images. And we'll have to wait and see where this new AI takes us and what it can create. There are some image edits that are not very cartoony and can be deceptive. Apple's AI can be used to edit people out of photos. However, metadata can indicate that a cleanup tool was used to make corrections to the photo. It will be an interesting time to see how people use these tools to change reality. And as we navigate this new territory, different groups will need to share best practices. Apple has published the results of a study on generative AI models. We have already heard Apple talk about the importance of privacy, but the document also places a strong emphasis on responsible AI development. It is worth noting that Apple is working to ensure that the words and images the AI creates do not reinforce stereotypes or systemic biases, and to be mindful of how the tools can be misused or used to cause potential harm. Again, this is all in beta. So Apple will be seeking feedback on Apple's research. The company scored the AI's results with a human satisfaction score, saying that it excelled at being useful and safer than other models. It also said that its AI handled adversarial prompts better than other models, but it did not receive a perfect score in the quote Apple wrote. We understand the limitations of safety benchmarks, given the broad capabilities of large language models. We actively conduct both manual and automated red teams with internal and external teams to continually evaluate the safety of our models. While Apple may not have full control over what the AI produces every time, this shows how important it is for it to continue to receive feedback and make adjustments. Apple appears to give users a fair amount of control over the type of interaction they have with the AI if they want to, i.e. if they want Apple to help them improve their text. Writing tools help you structure your writing better. They not only point out spelling and grammar errors, but also suggest ways to make your tone and message stronger. These AI writing tools can be invoked anywhere you would typically write text. You can tell if an AI tool is being used by seeing rainbow lights. Siri will also change with the AI's smarts. The whole area around the phone will be filled with colorful lights. Now, as we discussed before, Apple Intelligence does not pull information from the web, but instead gathers information from your personal accounts, photos, texts, calendars, and emails to make suggestions. But if you have a question that requires outside knowledge, Apple gives you the option to ask the question to Chat GP T 40. Not everyone wants to do that, so you don't have to during setup. You can decide whether you want Siri to offer Chat GP T 40 as an option when she doesn't know the answer. Conversely, if you want to use an open AI system, you can connect your Chat GP T account to enrich your Chat GP T results history. Apple is Apple because they are so protective of their experience. Apple has to approve apps in the App Store when you watch Apple TV. Plus, they have certain high standards for shows that are allowed on the service. Look at the iPhone. It took Apple 18 years to control where icons are placed on the screen. Yes, this is a feature coming in iOS 18 later this year. So it's surprising to see Apple being a little more accepting of things that we might not be able to control with AI results. But overall, I feel positive about the way Apple is approaching the big questions that many of us had about generative AI. I'm happy to see the beta slowly rolling out. Because if we're going to put this kind of power in the hands of iPhone users, it's probably best to go slower and keep it more controlled. Let us know in the comments. If you have any more questions about Apple intelligence, let us know what you'd send to someone on Image Playground first. Actually, don't bother. You might not want to know. See you next Friday for another story about the world of Apple. Thanks for watching.
