Artificial intelligence (AI) has been one of the most dynamic and competitive technological fields in recent years. Companies like Google, Microsoft, and OpenAI have led the way with innovations that are reshaping how we interact with technology.
However, one of the world’s most influential companies, Apple, has taken a different and more measured approach to AI. In this article by ITD Consulting, we’ll explore how Apple has developed its own AI model, Apple Intelligence, and how this strategy positions itself in the growing battle for supremacy in the field of artificial intelligence.
The Unexpected Collaboration: Apple and OpenAI
When news of a collaboration between Apple and OpenAI leaked, the tech community anticipated a deep integration of OpenAI’s technology into Apple’s products. However, the reality turned out to be quite different.
Although the Apple-OpenAI alliance is official, the two projects remain separate. Apple has developed its own AI model, known as Apple Intelligence, which operates independently from OpenAI’s ChatGPT.
Apple Intelligence: An Independent Product
Apple Intelligence is a set of several AI models designed to perform "everyday tasks for users." These Apple Intelligence models have been fine-tuned to excel in functions such as writing, summarizing, creating images, and simplifying interactions between applications.
With Apple Intelligence, these tasks are intended to be executed by Siri, Apple’s voice assistant, on devices such as the iPhone, iPad, and Mac.
During the WWDC 2024 presentation, Apple announced that Apple Intelligence would be integrated into iOS 18, iPadOS 18, and macOS Sequoia, with both an ‘on-device’ version and a cloud-connected version, called ‘Private Cloud Compute.’
Despite the lack of technical details during the presentation, Apple’s Machine Learning Research team published a report on the technical features of the language model (LLM) that powers Apple Intelligence.
Technical Features of Apple Intelligence
Apple Intelligence is powered by a language model with 3 billion parameters for the on-device version, and a larger model operating on Apple servers, which run on Apple Silicon chips and renewable energy.
This dual approach allows Apple to offer robust and efficient performance both on individual devices and through its cloud infrastructure for Apple Intelligence.
Comparison with Competitors
Compared to its direct rival, Google's Gemini Nano, Apple Intelligence operates at a nearly identical level. While Apple Intelligence boasts 3 billion parameters, Gemini Nano has two versions: one with 1.8 billion parameters for devices with low RAM, and another with 3.25 billion parameters for high-end devices.
Training Process and Data Security
Apple Intelligence has been trained using Apple AXLearn, an open-source project launched in 2023. For the content used in training, Apple has relied on licensed data and publicly available content collected by its web crawler, AppleBot.
Apple emphasizes that it never uses private user data or their interactions to train its AI model, Apple Intelligence, applying filters to remove any information that could identify users.
Additionally, Apple offers websites the option to request that their content not be used for Apple Intelligence, a significant step toward transparency and ethics in data use.
Optimization of the Apple Intelligence Model
Apple has optimized its AI model, Apple Intelligence, to reduce memory usage. The on-device version of Apple Intelligence has a vocabulary of 49K, while the server model has a vocabulary of 100K.
In comparison, Google's Gemini Nano has a token limit of 32K. This optimization allows Apple Intelligence to function efficiently across a wide range of devices, from the iPhone to Apple’s servers.
Apple Intelligence in Practice
Apple has adopted a specialized strategy for its AI, focusing on specific tasks such as brainstorming, classification, closed-question answering, programming, extraction, mathematical reasoning, open-question answering, rewriting, security, summarization, and writing.
This strategy is reflected in the benchmarks presented by Apple, where its on-device Apple Intelligence model compares favorably with Phi-3 mini by Microsoft, Mistral 7-B, and Gemma 7-B by Google, while its server model competes with GPT-4 Turbo and Mistral 8x22B.
Limitations and Future of Apple Intelligence
Despite the advancements, Apple Intelligence has some limitations. The multimodal capabilities of Apple Intelligence, meaning its ability to work with different types of data such as text, voice, and video, are not mentioned. This is an area where Google and OpenAI have already incorporated multimodal features into their models.
However, Apple promises that the AI functions of Apple Intelligence will primarily be performed on-device, and only connect to Apple’s servers when the task is too complex, using Private Cloud Compute to ensure data privacy and security.
The Alliance with OpenAI and Its Impact
Although Apple has developed its own AI model, Apple Intelligence, the alliance with OpenAI has not been completely overlooked. Starting at the end of this year, ChatGPT will be integrated into Siri and writing tools. For certain queries, Siri will defer the response to ChatGPT, providing Siri users with free access to ChatGPT without needing to create a new account.
Privacy and Security Considerations
The integration of ChatGPT raises concerns about data privacy and security. Apple promises that requests will not be stored by OpenAI and that user IP addresses will be hidden, following the guidelines of Private Cloud Compute. However, this aspect of data processing raises questions about Apple’s ability to maintain user privacy when using third-party services.
Openness to New Alliances
The alliance with OpenAI is not exclusive. Apple has made it clear that, in the future, new AI systems could be added to Siri, including Google's Gemini. This allows Apple to maintain strategic flexibility, exploring new collaborations and adapting to different markets, including China, where ChatGPT is unavailable.
Implications for the Future of Siri
Apple Intelligence and the integration of ChatGPT represent a significant attempt to revitalize Siri and its operating system. This strategy could allow Apple to catch up in the AI race, providing users with an easier and more efficient way to navigate their devices.
However, this move also presents challenges. Elon Musk, owner of Tesla and X (formerly Twitter), has expressed concerns about data security when using OpenAI. Ben Wood, chief analyst at CCS Insight, notes that the integration of ChatGPT could reveal the limitations of Siri, suggesting that Apple is acknowledging its shortcomings in the AI space.
Apple Intelligence: A Specialization Strategy
Apple has chosen a specialization strategy for its AI, focusing on specific tasks and optimizing its model for these functions. While this approach may limit the versatility of Apple Intelligence compared to more generalist models like those from Google and OpenAI, it also enables Apple to deliver superior performance in the areas for which its AI has been fine-tuned.
Apple has shown several benchmarks where Apple Intelligence excels in text writing and summarization, though the chosen benchmarks are not generic. It will be necessary to wait for future tests to determine the extent to which Apple Intelligence can compete against leading models in a broader range of tasks.
The current language model of Apple Intelligence uses 3 billion parameters, but Apple has indicated that this figure will increase with future models, potentially reaching up to 7 billion parameters. This will require devices with more RAM, suggesting that future iPhones and other Apple devices will include more RAM to support these advanced AI functions.
The promise of Apple Intelligence is that most AI tasks will be performed on-device, with cloud support for more complex tasks. Apple has built a cloud infrastructure that promises to be the "most secure and advanced" for AI, with end-to-end encryption and strict measures to protect user privacy.
Apple Intelligence represents a significant advancement in Apple's strategy toward artificial intelligence. While it faces challenges and criticisms, especially regarding privacy and security, its focus on specialization and seamless integration with its devices gives Apple a competitive edge. The partnership with OpenAI, although limited, adds an additional layer of functionality that could be pivotal in the race for AI supremacy.
As Apple continues to develop its AI model and expand its capabilities, it will be interesting to see how it positions itself in the highly competitive field of artificial intelligence. With a focus on privacy, security, and specialization, Apple Intelligence could become a key tool in the next generation of Apple devices and services, taking technological innovation to new heights.
If you want to learn more about Apple Intelligence or how to integrate AI into your business operations, write to us at [email protected]. We have a dedicated team specializing in technology and AI to help you stay ahead of the curve.
EN 
ES