Media: Apple's cloud data processing also protects privacy, with virtual black boxes, confidential computing, and self-developed high-end chips all in play

Author: Du Yu

Source: Hard AI

On Wednesday, May 29th, media reports cited four former Apple employees who had been involved in related projects, revealing that Apple plans to process data from artificial intelligence applications in a virtual black box on its servers. Even company employees will not be able to access this data, aiming to process some AI functions in the cloud while reducing the AI workload on personal devices at the processing end, while continuing to protect data security and privacy.

These heavyweight AI-related news may be unveiled on the first day of Apple's annual Worldwide Developers Conference on June 10th. Apple's stock rose 0.9% at midday on Wednesday, approaching its yearly high.

Wall Street News previously mentioned that Apple is reportedly developing its own chips for data center servers, with the internal project codenamed ACDC. This chip may focus on running AI models (inference), rather than training AI models, with the latter still likely dominated by Nvidia's products.

The latest reports indicate that the ACDC project is officially named "Apple Chips in Data Centers," which will ensure that the aforementioned virtual black box processes AI-related personal data. The concept of the virtual black box is similar to the "confidential computing" in the computer industry, ensuring that data is kept confidential during processing.

Sources say that Apple's confidential computing technology will utilize high-end self-developed chips originally designed for Mac computers, providing better data security than chips from competitors Intel and AMD. Through self-developed chips, Apple can also control the hardware and software on its servers, giving it more security advantages than competitors like Google and OpenAI who process AI data in the cloud but rely on non-self-developed chips.

It is reported that Apple intends to promote the processing of personal data on its servers through the virtual black box method, ensuring privacy similar to processing personal data locally on the iPhone side.

Some netizens commented that perhaps the Apple logo will change to a small lock to emphasize the security of the "AI era."

This allows Apple to integrate more AI features into hardware products such as smartphones, smartwatches, and mixed reality headsets in the future, and run some AI functions and data processing on its self-developed cloud servers, no longer limited by the processing power, computing memory, and battery life of the hardware products in users' hands, while also not threatening the long-term commitment to user data confidentiality.

For example, the AI data processing capabilities of Apple's wearable devices will be transferred to cloud servers, making devices such as headsets and smart glasses lighter without the need to equip large batteries and cooling fans for power-consuming processors on the devices.

Furthermore, machine learning tasks such as photo tagging and file indexing can be transferred from the iPhone side to cloud servers to improve the speed and battery life of the phone, and even the generative AI functions for photos and videos on the phone will be improved.

Wall Street News has also mentioned that these practices signify a shift in Apple's attitude towards AI. For years, Apple has prioritized edge AI, processing user data locally on devices through edge computing and viewing it as a better way to ensure security and privacy.

However, through the ACDC self-developed server chip project, components in Apple's cloud servers will be able to protect user privacy through a "Secure Enclave" physical method, isolating cloud-based AI-related data from security vulnerabilities.

The "Secure Enclave" hardware technology was released as early as 2013, used to store user biometric data captured by the iPhone 5S fingerprint sensor. The method involves designing a physical area separate from the main processor in the chip, serving as a safe deposit box for sensitive data such as passwords and encryption keys. Even if hackers invade the iPhone's software and central processor, they cannot access this secure area.

Insiders say that Apple plans to use this "Secure Enclave" hardware technology to help isolate the data processed on its cloud servers, combined with the confidential computing method mentioned above, Apple can process AI-related data in the cloud while making it difficult for hackers to access this personal data even in the event of server data leaks.

However, some analysts point out that Apple's method of processing AI data in the cloud is not foolproof, as servers still face the risk of physical tampering, and it is currently unclear how Apple will enable a single chip to provide this confidential service for multiple users in cloud servers. Given that Apple's installed base of devices is as high as 2.2 billion, it is a question mark how quickly the company can support large-scale AI services through self-developed server hardware.

Some speculate that the AI functions supported by Apple in cloud servers may initially be limited or only provided to users with newer iPhone models Previously, the media summarized that the highlight of Apple's global developer conference this year is how to introduce generative AI into the iPhone. The new generation iOS 18 mobile operating system may introduce new AI features, and Siri smart assistant is also expected to be upgraded through AI:

Apple may consider introducing a large number of new AI features to the iPhone as the most important update since the release of the first generation iPhone.

While Apple, which has prioritized on-device AI for many years, will use high-end self-developed chips configured for Mac for cloud computing servers, allowing Apple's devices to perform advanced AI tasks, with relatively simple AI functions directly implemented by the internal chips of iPhone, iPad, and Mac; the first server chip is the M2 Ultra