What is a real AI phone? There has never been a clear conclusion in the industry on this question. Mobile phone manufacturers have launched so-called "AI mobile phones", but in fact, the paths and concepts taken are different. Some phones empower end-of-device applications with built-in large models, others use cloud-based AI to enhance functionality, and others simply pre-install third-party AI applications.
In this current situation, Apple, a giant in the mobile phone field, is expected to give a new definition and direction guidance for AI mobile phones at the WWDC conference in June this year. It has been revealed that Apple is ready to implant generative AI capabilities on a large scale on the new iOS system, covering smart assistants, search, health and many other fields.
However, before Apple's AI strategy is fully revealed, the domestic mobile phone brand Honor seems to have drawn an ideal blueprint for AI mobile phones for us. This is a different path, but it may represent the real landing form and development direction of AI mobile phones.
Built-in large model ≠ AI mobile phone
At present, the main strategy of mobile phone manufacturers in the field of AI is to introduce large models into mobile terminals and empower various applications. Users can experience emerging AI capabilities such as image generation, voice translation, and intelligent summarization on their mobile phones.
However, these so-called "AI applications" do not necessarily require direct access to mobile phones or built-in large models. Many third-party apps are already able to invoke cloud AI services through the Internet, bringing users the same or even smarter experience.
So, what is the value of directly introducing large models into the mobile phone side? Can it be called a real "AI mobile phone" just by accessing large models and developing AI applications? Apple has given a possible explanation for this in a recent paper.
In the paper "ReALM", Apple proposed a large model of ReALM that can be run on mobile phones. This model can not only accurately identify various elements on the screen, but also generate text input based on the specific position and relationship description of the elements, which can be analyzed and made by the language model.
In other words, ReALM can "observe" the user's various operations on the mobile phone like a human, and gradually learn and summarize the usage patterns. On this basis, combined with artificial intelligence agent technology, it can provide users with automated and personalized intelligent operation services.
Such a vision is undoubtedly imaginative, even reminiscent of the "Jarvis"-like AI assistant in a science fiction movie. In Honor's view, to achieve this real "AI mobile phone", it is far from enough to introduce a large model into the mobile terminal.
The real quality of AI mobile phones lies in the use of AI to fundamentally reconstruct the operating system and interactive experience, so that mobile phones can learn, think and actively serve like humans, rather than passively adapting to users' instructions. This is exactly the path that Honor has been exploring for years.
The "new kernel" of the operating system
As early as 2016, Honor began to embed AI capabilities on smartphones to achieve basic demand perception and decision-making functions, which is the stage of its AI phone 1.0.
In 2022, Honor launched MagicOS 7.0, a platform-level AI-enabled all-scenario operating system. This system can proactively provide scenario suggestions and services based on information such as geographic location and user habits, which initially reflects the concept of "reconstructing the system with AI".
In January 2023, Honor released MagicOS 8.0, in which the "Arbitrary Door" function is based on user intent recognition to achieve an intelligent flow experience across applications and devices, marking the further deepening of this concept.
In traditional mobile phone systems, the operating system kernel is mainly responsible for managing hardware resources. In Honor's vision, platform-level AI plays the role of a "new core", responsible for managing human-related factors, such as personal knowledge base, location status, usage habits, etc., so as to help the system accurately identify user intentions, efficiently schedule services, and bring users a smart experience of "knowing what you need and guessing what you think".
Such a design concept stems from Honor's redefinition of "AI mobile phone". In the view of Zhao Ming, CEO of Honor, AI mobile phones mean using AI to completely reconstruct mobile phone applications, operating systems and user experience, and AI should be ubiquitous and become the foundation of the entire system. This is fundamentally different from the current "mobile phone AI" route where most manufacturers only regard AI as the icing on the cake.
Agent-driven AI capabilities
Based on the innovative design of platform-level AI, Honor has further launched a "new ecological paradigm based on intent recognition".
In this paradigm, with the YOYO intelligent assistant as the carrier, the large model capabilities on both sides of the device and cloud can be efficiently integrated: the platform-level AI large model on the device side assumes the role of "central control", understands the user's intention, disassembles tasks, and securely and efficiently connects to the cloud large model to obtain the required services, and finally forms a closed loop.
In this process, the device-side model filters personal privacy by protecting the network to ensure that sensitive data is not uploaded to the cloud, effectively protecting user privacy and security. At the same time, the YOYO intelligent assistant can not only connect with Honor's own AI capabilities, but also integrate the services provided by third-party cloud models to bring users a "one-stop" intelligent experience.
Furthermore, Honor's goal is to build a unified ecosystem covering general models and vertical industry models on the MagicOS system, providing each user with comprehensive services of intelligent distribution and task closed-loop.
Clearly, this architecture has elevated AI capabilities to the operating system level, which is fundamentally different from simply integrating standalone AI applications. In Honor's vision, AI will fully dominate the mobile phone interaction experience and become a kind of "operation and decision-making center", driving the entire system to provide users with proactive and personalized services.
This vision coincides with the concept that Apple presented in the ReALM paper. However, compared with Apple, Honor has been working in this field for a long time, and has been at the forefront of the industry in terms of technology implementation and product implementation.
A new level for smart assistants
To a certain extent, the concept of "AI mobile phone" pursued by Honor represents a new height in the development of artificial intelligence in the mobile field - the arrival of the era of intelligent agents.
As we all know, the current development of AI is dominated by large model technology. However, a large model is only a powerful infrastructure, and how to efficiently apply its capabilities to real-world scenarios and bring real value to users is the key.
On this point, the recent views of well-known AI scholar Andrew Ng are thought-provoking. He believes that the workflow of AI agents will soon make a major breakthrough, and the degree of intelligence may exceed that of the next generation of basic large models. There may be an "agent moment" in the user experience – just like the wow of the first experience with ChatGPT.
In fact, this agent-based AI interaction mode is the core of Honor's "AI phone" concept. What Honor pursues is to let AI agents drive large model capabilities, dominate the interaction between mobile phones and users, and provide users with end-to-end intelligent services.
If we can send various commands on the phone to the AI assistant driven by the agent, it can understand the user's intent around the entire scene like a human, and use the full capacity of the system to complete more complex tasks by breaking down and orchestrating tasks, such as arranging trips, planning trips, organizing parties, etc.
At the same time, thanks to operating system-level resources, agents can also sense changes in scenarios, learn user habits, and actively think about decisions and invoke appropriate services. As a result, large models can truly undertake complex tasks and bring users a new interactive experience.
This interactive experience will go far beyond the current "human-to-machine" model based on standalone applications. Instead, there is a new human-machine relationship in which "machines understand people", and the large model will be able to achieve seamless dialogue and natural interaction with users, and actively provide users with personalized suggestions and services.
It can be said that AI interaction driven by agents is the highest level of Honor's "AI mobile phone" concept. Only when the large model capability is empowered by the agent and deeply integrated into the entire operating system can human-computer interaction be fundamentally innovated, and AI mobile phones can be truly implemented as a new terminal form.
Infinite possibilities for ecological integration
The vision of AI mobile phones depicted by Honor also contains more possibilities, that is, the deep integration of the future smartphone operating system and the AI ecosystem.
As Zhao Ming said, Honor hopes to connect the general model in the cloud and professional models in various industries on MagicOS to create a unified AI ecosystem. This means that a single smartphone operating system is expected to become a medium to connect various AI capabilities, so that users no longer need to download and install multiple independent apps to obtain comprehensive AI empowerment services.
In fact, this trend of integrating the operating system with the AI ecosystem is not unique to Honor's imagination. As early as last year, Google presented its "AI phone" concept at the Google I/O conference - the future Android system will be deeply integrated with Google's large model ecosystem to provide users with a multi-scenario AI-enhanced experience.
Compared to Google, Honor's exploration path may be more advanced. It not only elevates AI capabilities to the core level of the system, but also is expected to create an open platform that transcends the ecology of a single vendor, bringing together various large models, AI applications, and cloud services to provide users with a "one-stop" intelligent experience.
In this sense, Honor is injecting new development vitality into the mobile ecosystem. The operating system will no longer be a closed hardware management system, but a mainstay that integrates various AI capabilities and becomes the intelligent brain of digital life. Mobile phones will also evolve from simple mobile terminals to full-scene intelligent entrances connecting the real and virtual worlds.
Even if mobile phones as a single hardware form may be replaced by new device types in the future, their development concept of integrating operating systems and AI ecosystems will also run through the evolution of the entire intelligent era. We have reason to believe that based on this development path, the connection between human beings and the intelligent world will become closer and more seamless.
Challenges lurk in the future
Of course, although the path of Honor's "AI mobile phone" is forward-looking, it will take time to test whether it can finally succeed and win market recognition. There are still significant challenges lurking on the road to achieving this vision.
First of all, the limitation of computing power on the device side will make it difficult for large models to be fully deployed and applied on mobile phones. At present, even a leader like Honor can only introduce large model capabilities through device-cloud collaboration, and the real "large model on the device" is still a distant goal.
Second, the current working ability and stability of AI agents in complex scenarios need to be further improved. Under the current technology, it is difficult for AI assistants to complete the whole process of tasks from intent recognition to multi-service orchestration. There are still uncertainties in the accurate understanding, task decomposition and service scheduling of agents.
In addition, the privacy and ethical risks of large models also need to be considered. Although Honor has adopted the practice of isolating private data from uploading to the cloud, how to find a balance between protecting user privacy and obtaining sufficient training data is still an urgent problem to be solved.
In addition, the hardware, software, and ecological aspects around the development of AI mobile phones all need to be invested on a large scale and more original innovations. At the hardware level, it is necessary to pave the way for the computing power of large models, at the software level, it is necessary to pursue better architectural innovation, and at the ecological level, it is necessary to build an open and win-win new order of artificial intelligence.
Overall, while the vision of an AI phone is tempting, its path is also destined to be tortuous and complicated. However, anything great needs to start with a solid foundation. Through years of deep cultivation, Honor has laid a solid foundation for achieving this goal.
It is believed that in the near future, with the breakthrough of terminal computing power, the acceleration of model innovation, and the increasing maturity of the ecosystem, AI mobile phones will no longer be just the unique concept of Honor, but the proper form of the entire intelligent era. At that time, human interaction with the digital world will be completely new, and intelligence will be truly ubiquitous.