At a keynote event at VivaTech 2024 in Paris, France, on May 22, Honor’s CEO George Zhao unveiled its four-layered hybrid AI architecture comprising cross-device AI, platform-level, app-level, and cloud AI.
Omdia view
Summary
At a keynote event at VivaTech 2024 in Paris, France, on May 22, Honor’s CEO George Zhao unveiled its four-layered hybrid AI architecture comprising cross-device AI, platform-level, app-level, and cloud AI. This clear delineation between on-device, operating system (OS), and cloud points toward a direction the smartphone industry will likely follow. As only the most premium phones can have powerful multi-modal AI models on-device, many other (and older) phones will have to rely on a lighter AI engine for non-generative tasks and instead use cloud computing for more intensive tasks. As this will be lighter for the processor, it will be better for more efficient battery consumption.
Honor’s four-layer AI structure
During the keynote, Honor CEO George Zhao showcased the company’s four-layer AI architecture and how it integrates with Honor’s MagicOS android skin. The lowest on-device base layer is cross-device and cross-OS AI. This is the foundation of the ecosystem development Honor has been working on, allowing for easier photo and file sharing between phones, laptops, and tablets. Cross-device AI also promises the sharing of computing power and services.
The second layer is platform-level AI, meaning at the OS of your personal device. The biggest feature of the second layer is Honor’s MagicPortal, where on-device AI tries to understand the user’s intention when selecting text, images, or files, offering app suggestions with a simple drag-and-drop.
The third layer is app-level AI, where a lot of smartphone AI innovation (including Samsung Galaxy AI) has focused on, such as generative AI (GenAI) photography features.
The top layer, furthest from the on-device base level, is “Interface to Cloud-AI services.” This would provide users with access to Google’s cloud services.
Figure 1: Honor’s four-layer AI architecture
Source: Honor
For the top Cloud-based AI layer, Honor announced that it would partner with Google Services. Zhao said, “We are also delighted to forge ahead with Google Cloud, leveraging our combined expertise to unlock the potential of this hybrid approach and deliver even more seamless AI experiences to our users.”
Multi-modal AI interaction
One of Honor’s key flagship AI features is MagicPortal, which uses on-device AI to understand what users have selected and suggest tasks based on that. This is completed through a simple drag-and-drop interface, where users can drag an address they have received via text (for example) directly into a map app for directions. Honor calls this the “human-centric focus” of its AI. Understanding different contexts and how users interact with their phones is an important part of multi-modal AI.
As part of the keynote, there was a discussion with Dr. Justine Cassell, Dean’s Professor in the School of Computer Science at Carnegie Mellon University (CMU), currently on leave from CMU to hold the founding international chair at the PRAIRIE Paris Institute on interdisciplinary research in AI. On multimodal AI, Cassell commented, “Humans live multimodally. The rise of on-body devices like rings and pins, alongside powerful smartphones, shows growing comfort with AI integrating verbal language, nonverbal cues like gestures and facial expressions, and paraverbal elements like speech intonation.” Cassell also predicted that adopting more multimodal AI on phones would allow more kinds of people to engage with that AI.
Privacy concerns surrounding hybrid AI
The smartphone and wider consumer tech industry is pivoting from simply developing on-device AI to creating a system for how that on-device AI interacts with and shares information with cloud-based AI services. As this happens, privacy and security concerns over which information is shared and how and whether it is used to train AI models are mounting.
By working with Google Cloud, Honor will adopt Google’s strong privacy and security protections. Zhao also commented that the on-device AI Honor has developed “offers personalized recommendations while safeguarding user privacy by keeping data on the device.”
However, users of AI systems will be increasingly questioning what data they are sharing, especially as AI becomes more embedded on devices and understands everything—from the emails users receive and their calendars to their search history and the contents of their photo galleries.
AI coming to older phone models
Shortly after first announcing Galaxy AI in January, Samsung revealed it would bring its AI features to older phone models, including the S23 series. This continued in April with the release of One UI 6.1, which brought Galaxy AI to the S22 series, the Z Fold4 and Z Flip4. This gradual expansion was part of a test-and-monitor approach to see if features such as Chat Assist, Live Translate, and Generative Edit would work effectively and efficiently on older chipsets.
Along with announcing the launch of the new Honor Magic 200 series on June 12, it was revealed that the latest MagicOS 8.0 and its AI features would be coming to Honor Magic V2, which is powered by a Snapdragon 8 Gen 2 chipset and Honor 90 devices. Although it is not clear which 90 series devices will get this update, the Honor 90 Pro uses a Snapdragon 8+ Gen 1 chipset, while the Honor 90 uses a Snapdragon 7 Gen 1 Accelerated Edition, and the Honor 90 Lite and 90 Smart use a lower-end Mediatek Dimensity 6020 7nm chipset.
With this expansion, AI features are becoming more accessible to users of mid-end devices. When Samsung and Google were the first to introduce GenAI features, it was seen as a premium-only feature, but with the Pixel 8a, Samsung S22, and other devices now in the mid-end price tier getting GenAI features, this expansion is happening faster than many anticipated. Hybrid AI will be a key part of this—where older or mid-range chipsets cannot effectively run large language models (LLMs), more phones will rely on lighter AI models, such as Honor’s, for on-device features and cloud AI computing for bigger GenAI tasks.
Appendix
Further reading
Smartphone AI Market Landscape – 2024 (February 2024)
Smartphone Need-To-Know – May 2024 (May 2024)
Author
Aaron West, Senior Smartphone Analyst, Components & Devices
