The AI Phone War Has Begun: Can Intelligent Agents Replace Apps? Chinese Smartphone Manufacturers Accelerate Their End-Side AI Layout

The AI Phone War Has Begun: Can Intelligent Agents Replace Apps? Chinese Smartphone Manufacturers Accelerate Their End-Side AI LayoutWith Apple Intelligence set to launch overseas and plans to integrate ChatGPT services into Siri and introduce powerful image generation tools in December, the AI phone war has officially begun. Domestic Android manufacturers have followed suit, releasing upgrades to their end-side AI and operating systems, with concepts like AI agents and AIOS emerging left and right

The AI Phone War Has Begun: Can Intelligent Agents Replace Apps? Chinese Smartphone Manufacturers Accelerate Their End-Side AI Layout

With Apple Intelligence set to launch overseas and plans to integrate ChatGPT services into Siri and introduce powerful image generation tools in December, the AI phone war has officially begun. Domestic Android manufacturers have followed suit, releasing upgrades to their end-side AI and operating systems, with concepts like AI agents and AIOS emerging left and right. Although it's still unclear when Apple Intelligence will be available in the Chinese market, the massive impact it has brought has forced domestic smartphone manufacturers to further leverage AI as a selling point in the competition for flagship models.

Every smartphone manufacturer has made it clear that they want to create system-level AI, AIOS, and AI agents. Guo Tianxiang, research manager at IDC China, said that Android manufacturers and Apple share a similar approach to AI, both focusing on end-side models and intelligent agents. "Domestically, we are not far behind in AI."

Can Intelligent Agents Kill Apps?

As iPhone's voice assistant, Siri can be used for simple tasks via voice commands. However, because its answers were primarily based on search engines in the past, its intelligence was very limited and failed to replace the app interaction model on phones. With the development of large models, phone assistants like Siri are poised to become more intelligent, developing from voice assistants to AI agents. For example, to book a hotel trip, users won't need to open an app anymore, but can instead directly interact with an AI agent to get it done.

When asked if intelligent agents will replace apps, Zhao Ming, CEO of Honor, believes that the development will likely head in that direction but that apps and intelligent agents will coexist for a long time. "This involves user habits and various unexpected experience barriers, so coexistence for a long time, or even permanent coexistence, is inevitable."

AI Screen Recognition: The First Step Towards Intelligent Agents

As the first step towards intelligent agent interaction models, AI screen recognition has already begun to take root on domestic Android phones. The recently released OPPO Find X8 features a one-click screen query function. This function can intelligently analyze screen information and interact with users based on the content, providing answers and operations accordingly.

"For example, if you take a picture of a scenic spot, you can simply click to have AI identify it and answer where it is and what stories it holds. It might sound simple, but it involves millions of data points from over 16,000 3A-rated scenic spots nationwide for specialized training," said Zhang Jun, product director at OPPO AI Center.

Honor has also launched MagicOS 9.0, its AI operating system built with intelligent agents. Zhao Ming introduced that intelligent agents can now simulate humans to click on screens, understand text on screens, think slowly, find key information, and then perform relevant operations. Currently, there are two main categories: "autonomous driving" agents and agents that interact with applications.

"Autonomous agents do not require third-party intervention, starting with analyzing and understanding user intent. For example, if you say 'Help me order a drink,' the agent can understand the information and logic behind the intent, break down the intended scenario into executable instructions, and ultimately complete the operation of ordering coffee. The other kind requires collaboration from the application side. For example, Honor and China Mobile's Lingxi large model. When checking your phone bill balance or adding 50 yuan, Lingxi's model is called in to take over. In the future, both types of agents will coexist. There will be aspects that require ecological intervention, and there will be operations that can be performed automatically."

Intuition and Efficiency: The Mainstream of Future AI Interaction

Regarding the future development of AI interaction on phones, several industry insiders believe that intuitive and direct methods will ultimately prevail. Guo Tianxiang said screen recognition interaction is a new type of interaction for AI phones, making it easier for users to use and reducing the learning curve. Currently, future AI interaction will likely be based on the most direct and simple methods, starting from human instincts.

Liu Zuohu, OPPO's chief product officer, also believes that intuitiveness is the most basic AI principle. "Every week I hold AI-specific meetings and I always stress one thing: no matter what it is, it must be intuitive. We see a lot of things that may be showythat look simple, but require very advanced technology behind them. Take one-click screen query as an examplehow to recognize user intent, recognize the screen... a lot of routing technology is involved. But ultimately, technology must return to users to become products. For example, when using navigation, the address is open when you open it, and you can reach your destination with one click. The AI era is about more intuitive efficiency, which is the most basic AI principle."

End-Side Models: Balancing Experience and Performance

While the potential of large models being embedded into phones is immense, challenges also exist. The limited computing power on phones means that end-side models cannot be too large, but models with small parameters are also limited in their capabilities.

Guo Tianxiang stated that the focus of end-side models is no longer on the size of the model parameters but on balancing user experience, memory usage, and power consumption.

Liu Zuohu acknowledged that end-side models have very high performance requirements, requiring both high performance and memory. Therefore, continuously optimizing the architecture and achieving high-energy efficiency to unleash the potential of chips is still a long road ahead. "There's still a lot we can do. For example, platform cooling might seem simple, but it's actually very difficult. Also, how to manage the bottom layer memory and so on. Honestly, AI is just starting out in the phone industry. We'll see a lot of changes in AI going forward."

AILoRA: Reducing Memory Footprint

Zhang Jun revealed that OPPO will soon release a new end-side architecture called AILoRA to reduce memory and other resource consumption. "The biggest bottleneck for end-side AI is the use of computing resources on phones. For example, if you have three end-side functions running on your phone at the same time, they typically take up three corresponding resources. Imagine models as train engines. If you have three models, you need three engines plus carriages. The LoRA architecture uses a base model + application model approachyou only need one base model, meaning only one engine. The subsequent application models are like three carriages, like a six-shooterthey can be swapped out. When a model is needed, you just add the corresponding carriage. This can reduce peak memory usage by 75%."

The Post-AI Phone Era: Intelligent Agents Will Replace More Manual Operations

Regarding the development of large models, the general consensus is cautious in the short term and optimistic in the long term. This is also true for their implementation on the end-side. Liu Zuohu analyzed that the changes in the AI era are extremely fast. "In the past, we planned our phone operating systems on a six-month or yearly basis. But in the AI era, this won't be the case. Who knows what AI will look like in a year? AI products may not even be planned every three months, but every month. Models change too fast, and technology is exceeding our imaginations. Honestly, I myself feel a great sense of urgency."

Liu Zuohu emphasized that in the AI era, products need to be fast. "You have to run, and you have to run fast, otherwise you'll fall behind. You need to keep up with the changing technology." Recently, the China Academy of Information and Communications Technology released the world's first "Terminal Intelligence Level Research Report," dividing terminal intelligence levels into five levelsL1 to L5. The higher the intelligence level, the higher the degree of autonomous participation of the terminal, and the lower the human participation. L1 and L2 have a certain degree of intelligence and can complete single types of tasks. L3 and L4 gradually evolve from understanding complex intent to recognizing potential intent. L5 has comprehensive intelligence and can autonomously plan and complete all types of tasks.

Zhao Ming said that the current level of terminal intelligence is at L3, and the time it will take to reach the next stage of L4 and L5 will be longer, requiring more accumulation. "Today, we can achieve 950 user intent categories. In the future, we will be able to cover all aspects of phone operation, gradually eliminating the need for more human intervention in traditional phones. We can now make phone calls with a single voice command, video chat on WeChat is also possible, and ordering coffee is also doable. The next step is to achieve more, more vague instructions, and an understanding of more complex relationships."

Conclusion: The AI Phone Battle Has Just Begun

With the launch of Apple Intelligence and the accelerated layout of domestic manufacturers, the AI phone battle has just begun. Optimization and application of end-side models, the continuous evolution of intelligent agents, and the improvement of user experience will all become focal points of future AI phone competition. In this race of technology and experience, who will ultimately win? Let's wait and see.