Apple Intelligence employs a dual strategy involving both local, on-device models and cloud-based models. This hybrid approach allows for efficient processing: tasks are handled on-device whenever possible, ensuring privacy, while more resource-intensive tasks are offloaded to the cloud with minimal data transfer. This method mirrors Google’s strategy with Android but is distinct in its robust privacy measures.
Apple places a strong emphasis on user data security and privacy. On-device processing ensures that most user data remains on the device, reducing the risk of data breaches. When cloud processing is necessary, only the minimum amount of user data required is sent to the cloud, processed, and then immediately deleted. Apple’s cloud infrastructure features stringent access controls and no persistent storage, ensuring data is not saved or accessible remotely.
Apple distinguishes itself by training its models using publicly available and licensed data, allowing companies to opt out of having their data used. This ethical approach contrasts with the less transparent data sourcing practices of competitors like OpenAI and Google, showcasing Apple’s commitment to responsible AI development.
Despite not having a direct competitor to ChatGPT or Google’s Gemini, Apple enables users to access these models through their devices. Users can opt in to use services like ChatGPT, with provisions for both free and premium accounts. This flexibility allows users to leverage the strengths of Apple Intelligence while still accessing other leading AI models as they prefer.
Here comes AI: Apple Intelligence
At a glance
Apple is considered the laggard in the industry when it comes to releasing a product that competes against OpenAI and Microsoft's ChatGPT, Google's Gemini, Amazon and Anthropic's Claude, and Perplexity.ai's Perplexity, among others. The lack of an AI-powered chat product left Apple at risk of missing the AI hype train, but last Monday's Apple Intelliggence announcement signaled their entry into the race. Smart branding for the Cupertino company's take on "AI", right? :) So what differentiates Apple Intelligence from the rest?
Apple Intelligence is a two-pronged approach to providing "AI" to its users: [1] local, on-device, models and [2] cloud-based models. This is similar to how Google is doing it with Android. Everything that the operating system thinks can be done on device, it processes, no internet connection required, otherwise, it pushes it to the cloud for more powerful processing resources. The similarity stops there. Apple Intelligence does this whilst ensuring that user data is secure and protected, and its privacy is preserved.
Local, on-device, processing provides Apple Intelligence with access to the user data (note that if Apple Data Protection for iCloud is enabled, all, except Mail and Calendar data are end-to-end encrypted). The user data remains on the device, nothing is sent to the cloud - which ensures its security and privacy. This is something other cloud-based providers cannot access easily. What this provides is better personalization and contextualization - Apple Intelligence knows its user better than any other "AI" in the market today.
The second approach of Apple Intelligence requires more processing resources, resources that users' devices are not capable of. This is where the private compute cloud comes in. Apple has designed its cloud services with privacy in mind. When the device operating system determines that the task requires cloud resources, the minimum amount of user data required is sent to the cloud, which is then processed, sent back to the device, and then that user data on the cloud is deleted. The cloud infrastructure does not have persistent storage (so there is no way to save the data), nor is it accessible remotely (no human can remotely login and take a peek). To add to this, Apple adds more checks between device and its cloud - to ensure that the device is indeed talking only to Apple Intelligence private compute cloud. Finally, Apple offers third parties to audit their code to ensure that what they are claiming is indeed true, that user data is secure and private whilst being handled by the private compute cloud.
All property of Apple, Inc.
Apple probably is one of the few companies that designs "AI" in an ethical manner. Apple trained its models using publicly available data on the web, with a facility for companies who do not wish to have their data used to opt out. To augment its training data, Apple also used licensed data. Ask OpenAI and Google where they ethically sourced their training data, you might not get a clear answer. For more information, check out https://machinelearning.apple.com/research/introducing-apple-foundation-models.
Whilst Apple Intelligence is core to Apple's push of using "AI" on their platform, Apple still has no competitor to ChatGPT and Gemini. However, Apple recognizes that users may want to access these models, so Apple provides connection with other models, starting with OpenAI's ChatGPT. Apple provides free access to ChatGPT, whilst also allowing users to use their premium accounts to access the premium features. Note, however, and contrary to the FUD that Elon Musk is spreading, ChatGPT is NOT integrated into the operating system. As such, Apple's operating system can also connect with Google's Gemini in the near future. Apple provides the entire gamut of the best of what "AI" can offer its users - Apple Intelligence, and whatever users prefer to add, ChatGPT, or later Gemini, Claude, you name it, soon it can be connected. Oh, I forgot, using ChatGPT is OPT-IN - users need to explicitly choose when to use it.
Apple's Apple Intelligence shows the industry how "AI" innovation can be done, whilst preserving user privacy and protecting copyright. So do you think Apple is still falling behind in "AI"?