Apple Intelligence

Machine learning led to more powerful technologies people ultimately dubbed "artificial intelligence." While it is mostly a misnomer based on science fiction, Apple eventually began to embrace the term until it moved past it with Apple Intelligence.

Apple CEO Tim Cook will tell you that Apple Intelligence isn't named because of the "AI" that came before it. Instead, he'll insist that it was given the name that best described the technology.

Regardless, Apple Intelligence is Apple's take on generative technologies that have taken the world by storm. OpenAI, Google, and others have flooded the market with their imperfect implementations in the hope of wrestling away some early market share.

Despite pundit declarations that Apple was behind and would never catch up, the company has had a hand in machine intelligence for over a decade. Now, finally, Apple has chosen to appease these complainers by introducing a more nuanced version of the technology.

Concerns regarding Apple Intelligence

So-called artificial intelligence hit the market running and has been iterated at a full sprint since. The explosive growth of the AI industry relied on having access to mountains of data gathered from the internet and its users.

Some companies, like Google, had a strong foundation to train their models thanks to the literal mountains of data collected from users. However, there wasn't any way to predict exactly how models would react to learning from the trove of human-produced data.

The result, so far, has been chatbots and search engines that hallucinate because of poorly sourced data from non-factual sources. Replies to searches and chats can sometimes result in funny suggestions like glue on pizza, but other times result in something potentially dangerous.

Apple Intelligence is a group of foundational models built to run either at a smaller scale for iPhone, iPad, and Mac or at twice the size or bigger on a server. It is a testament to Apple's vertical integration of hardware and software.

Since the models are local to the user's device, they can access sensitive data like contacts, photos, and location without the user needing to worry about compromising that data. Everything gathered by the Appel AI never leaves the device and isn't used to train the model.

In the event a request can't be run on the device, it can be sent to a server owned by Apple, running Apple Silicon with a Secure Enclave. All data sent off the device is encrypted and treated with the same protections it would have if it stayed local. This server-side technology is called Apple Cloud Compute.

Even with privacy and security at its foundation, Apple Intelligence still had to be trained on something. Apple purchased licensed content for some of the training, but other data was sourced from Applebot, a web scraper that has existed since 2015 for surfacing Spotlight search data.

There is an ethical conundrum surrounding Apple's model training practices, but the company has assured that it has taken steps to ensure all public data is freely available on the web, lacks copyrighted or personal content, and is filtered for low-quality content. It's a small reassurance, given other companies had no regard for such precautions.

There has also been some concern about Apple Intelligence and the environment. On-device models don't have an impact on Apple's green energy initiatives, and server-side computation is performed on Apple Silicon, which is powered by green energy sources.

Apple Intelligence features

Apple Intelligence is made up of several models that have specific training to accomplish different tasks with minimal risk of error or hallucinations. There are several planned features at launch, with more coming over time to iOS 18, iPadOS 18, and macOS Sequoia.

Writing Tools will work anywhere text entry is supported across the device's operating system. It can analyze text to proofread for errors, change the tone of the text to be more professional, or summarize the text.

Priority Notifications analyze incoming notifications to ensure only the most important are at the top, and they are summarized too. It ensures messages, deliveries, and other important information is available at a glance.

Data summarization is everywhere in the supported operating systems. Summarize web pages, emails, text threads, and more.

Image Playground is an all-new app from Apple that generates images like emoji or cartoon-styled art. There is a standalone app, but the function appears in different locations, such as in iMessage.

Siri gets a huge boost from Apple Intelligence by having access to data presented by the various on-device models. The smart assistant is more knowledgeable about the user, their plans, who they're talking to, and what data is within apps.

Users will be able to carry out more natural conversations with Siri, even if they make a mistake and correct themselves mid sentence. And as long as the user asks questions within the same relative timeframe, Siri will remember context from previous queries.

The voice interactions with Siri will primarily serve as a way to activate Apple Intelligence features from anywhere. However, Siri will have additional abilities thanks to app intents providing context for what is visible on the display.

Apps with Apple Intelligence Siri support:

• Books
• Calendar
• Camera
• Contacts
• Files
• Freeform
• Keynote
• Mail
• Magnifier
• News
• Notes
• Photos
• Reminders
• Safari
• Stocks
• Settings
• Voice Memos

The upgrade to Siri could take some time. Rumors suggest Apple will not release the new Siri until iOS 18.4 sometime in spring 2025.

Apple has more AI features separate from its proprietary systems thanks to a partnership with OpenAI. They are off by default, but if enabled, users can send some requests to ChatGPT. Every request must be screened and approved by the user, and all data from the interaction is discarded.

Apple Intelligence requirements

Apple claims that Apple Intelligence couldn't exist until it could run locally on a device, and that wasn't possible until the A17 Pro. So, the available devices for the technology are very limited.

The base requirements appear to be tied to the size of the Neural Engine and the amount of RAM. The iPhone 15 Pro and iPhone 15 Pro Max have 8GB of RAM and run the A17 Pro with a sizeable Neural Engine.

Apple is expected to release the iPhone 16 lineup with A18 processors capable of running Apple Intelligence on every model, including non-pro ones. It could spur incredible demand for the lineup.

Macs and iPads can run Apple Intelligence as long as they have an M-series processor. The desktop-class processor has always had a large Neural Engine and enough RAM at 8GB minimum for the chipset.

Apple Intelligence release date

Apple Intelligence will release as a public beta in the fall of 2024. It is expected to begin a slow rollout to users once iOS 18 is released in September.

Instead of just rolling out like a software release, there will be a waitlist. There haven't been any details about how the waitlist will work or how users will be added to it.

Since it is a beta, Apple is expected to alert users when the features are available over time. It isn't known how Apple will perform the rollout or who will be let in and when.

Everything below this point is based on pre-WWDC 2024 rumors and leaks. It has been preserved for reference.

Pre-announcement AI rumors

Apple is known to adopt its own marketing terms to describe existing phenomenons, like calling VR or MR by the in-house term "spatial computing." It seemed like Apple was going to lean into Artificial Intelligence, or AI, as a term, but it is expected to add a twist by calling it "Apple Intelligence."

Whatever Apple calls it, the result is the same. There will be system-wide tools that run using generative technologies that are much more advanced than the machine learning algorithms they'll replace or augment.

Apple may reveal plans for a server-side LLM during WWDC, but rumors suggest it isn't ready for prime time. Instead, the focus will be on local models made by Apple that can call out to third-party server-side LLMs like ChatGPT.

Local Apple Intelligence

OpenAI's ChatGPT and Google's Gemini Large Language Models (LLMs) have one thing in common — they are gigantic models trained on every ounce of available data on the internet that live in servers.

The "Large" in LLM is literal, as these models can't live on something like an iPhone or Mac. Instead, users call the models and their various versions through the cloud via APIs. That requires data to leave the user's device and head to a server they don't control.

Server-side models are a must for such enormous data sets, but they come at the cost of privacy and security. It may be harmless to ask an LLM to generate a photo or write an essay about historical events, but asking it to perform a budget may cross a line.

Apple hopes to address these types of concerns by providing more limited local models across the iPhone, iPad, Mac, and Apple Vision Pro. The pre-trained models will serve specific purposes and will contain very limited but specialized data sets.

The on-device models will learn based on user data. The data will never leave the device without explicit permission.

Upgrading Siri

A part of local Apple Intelligence will be providing significant upgrades to Siri. Apple's assistant is expected to become much more conversational and understand contexts better.

A specific model, likely based on the "Apple Ask" pre-trained model that's been tested internally, will be the driver behind Siri. Its limited data set is said to be a protection against hallucinations, which plague LLMs like Google Gemini and ChatGPT.

Users can ask Siri to perform many tasks across Apple-made apps, from simple to complex. It isn't yet known if an API will allow developers to target this new Siri, but it is likely.

Apps with Apple Intelligence Siri support:

• Books
• Calendar
• Camera
• Contacts
• Files
• Freeform
• Keynote
• Mail
• Magnifier
• News
• Notes
• Photos
• Reminders
• Safari
• Stocks
• Settings
• Voice Memos

Many Siri functions in these apps involve general navigation and search. Like opening a specific book or creating a new slide.

It seems many of the functions found in the new Siri commands are derived from previously available accessibility options that could control the device via voice. This melding of features will provide more control for all users, not just ones relying on accessibility options.

Apple Intelligence at WWDC

All will be revealed during WWDC 2024. Apple is expected to reveal Apple Intelligence during its keynote, where iOS 18, iPadOS 18, macOS 15, tvOS 18, watchOS 11, and visionOS 2 will also be announced.