Explore the impact of the new operating system scenario under the Honor AI trend on the future

On January 10, the Honor 2023 Developer Conference was held as scheduled. At this conference, Honor launched a new operating system, MagicOS 8.0, which introduces human-computer interaction functions based on intent recognition. At the same time, Honor also released a self-developed AI large model, which is called the "Magic Large Model" and has 70 billion parameters. This large model uses end-to-end AI technology to provide users with innovative functions such as conversational filmmaking, gallery semantic search, and multi-modal schedule management. The introduction of these functions will bring users a more intelligent and convenient experience.

On January 11, Academician Lu Jian of Nanjing University, Vice President of AutoNavi Dong Zhenning, Professor Huang Xuanjing of Fudan University, Associate Professor Yu Chun of the Department of Computer Science of Tsinghua University, Dean of the New Species Research Institute Wu Bofan, and President of Product Line of Honor Terminal Co., Ltd. Fang Fei and Deng Bin, President of R&D Management of Honor Terminal Co., Ltd. gathered together to discuss "Insight Trends: A New Generation of Operating Systems Driven by AI Large Models" and demonstrated the achievements of MagicOS in exploring and thinking about future operating systems. They deeply discussed the development trend of operating systems driven by artificial intelligence large models, and explored how to better use AI technology to improve the performance and functions of operating systems. This discussion provided valuable ideas and inspiration for the future development of operating systems.

Scenarioization and intelligence will represent the future of operating systems.

Academician Lu Jian of Nanjing University looked forward to the development of next-generation operating systems at the opening of the forum. He believed that these operating systems will support a new world of three-dimensional integration of humans, machines, and things. Later, Wu Bofan, director of the Commercial New Species Research Institute, concluded that this point of view expressed the core point of this exchange. IT House believes that this sentence can best summarize the development trend of the next generation operating system.

New generation operating system: Scenario is the goal, intent recognition is the road, and AI large model is the driving force

How to understand "scenario-based and intelligent operating system"? What is the relationship between large models and operating systems? Academician Lu Jian put forward his own opinions from a macro level on the spot.

The big models that are popular today are the products of the rapid development of the new generation of information technology. The characteristics of the new generation of information technology can be summarized as "six everything", that is, computing is integrated into everything, the network connects everything, data represents everything, intelligence spiritualizes everything, software defines everything, and trust is ubiquitous in everything. These characteristics demonstrate the broad application potential of large models.

The linked development of these six things has introduced us into a new world of three-dimensional integration of humans, machines, and things. This is a new era of superposition and integrated development of human society and nature, the world of physical creation and the world of information creation. In this new era, what the human-machine-thing integration application scenario represents can be regarded as a broad computer or computing platform. In this sense, the future operating system can be understood as a scenario-based and intelligent operating system built on a platform where "the application scenario is the computer", supporting this new world of ternary integration of humans, machines, and things.

In this new world, humans and machines need to work closely together, and the physical world and machines also need to be closely integrated. Humans, machines, and objects break down the boundaries and collaborate with each other to create rich and diverse application scenarios. In these scenarios, software plays a decisive role, they define everything. Therefore, the core goal of the next generation operating system is to provide the best services for these integrated application scenarios.

Fang Fei, President of Honor Product Line, said during the conversation:

When we first started designing the MagicOS architecture, we also had ideas about people, machines, and things. We said at that time that we would break the system, break the application, break the device, and build a human-centered and intelligent system.

Person-centered, in fact, is centered on the scene in which people are located.

So, from the perspective of the operating system, how to serve these application scenarios? The answer is: intent recognition.

As we all know, the development of human-computer interaction in operating systems has mainly experienced two interaction paradigms. One is the earliest batch processing paradigm, and the other is the command-based interaction design that appeared around 1964. Command-based interaction has also gone through different stages of development, including the command line interactive interface (CLI) represented by DOS, and the graphical interactive interface (GUI) represented by Windows, which uses keyboard and mouse input. Although the interaction between the two The methods are different, but they all start from the equipment-centered logic, and there is limited room for efficiency improvement.

The arrival of the AI ​​era has brought the first new paradigm and change to human-computer interaction in the past 60 years: intent-based result specification.

In other words, AI needs to understand human intentions. Users no longer need to tell the computer what to do. On the contrary, they only need to tell the computer what results they want. This will undoubtedly become a new direction for future OSs in human-computer interaction.

So, the question comes again, how to achieve accurate intent recognition? The emergence of AI large model technology makes all this possible.

The core concept of intention recognition human-computer interaction is to use deep learning and artificial intelligence technology to enable the system to more accurately understand and respond to the user's intentions, achieving more intelligent, multi-modal and natural human-computer interaction. The large AI model has very powerful intelligent information processing capabilities, allowing AI to truly understand people's intentions more accurately and interact naturally with people in a multi-modal manner.

Fang Fei also said in the roundtable exchange: "We need to truly understand the user on the client side. The initial understanding of this technology was relatively primitive. Today, the big model is here, for the identification of intentions, and the understanding of personal portraits and behaviors. , it has faster intelligence capabilities."

There is still a question here, how to enable large models to achieve accurate personalized intention recognition? This talks about the inevitable trend or direction of the development of large AI models: device-cloud collaboration.

The advantage of a large cloud model is its large computing power and huge number of parameters. Inference cost overhead and network delay must be considered. Security cannot be guaranteed in the process of migrating private data to the cloud. What is important is that without key personal data, it is impossible to understand personal needs and speculate on personal intentions based on long-term use experience.

If you want to combine personal data during the use of large models, you can only do it locally, and it is guaranteed not to go to the cloud. Therefore, the terminal-side large model is the inevitable way to realize intention recognition. But at the same time, it also requires the powerful computing power and rich application ecosystem of the cloud, so device-cloud collaboration is a trend that must be followed.

Professor Huang Xuanjing of Fudan University also used a vivid example to illustrate the scenario of a large model of device-cloud collaboration:

In the new year, I want to write a personal summary for 2023, including how much I spent last year. These things are highly private. We can do some privacy protection locally and replace some numbers and names that should be replaced. , to the large model in the cloud, it helps me pull up a framework that is almost written, and I can also make some local substitutions.

It can be seen that the next generation operating system is aimed at service scenarios and uses intent recognition as its path, and the AI ​​​​large model is the driving force behind it.

This is also the core concept and basic logic of Honor MagicOS 8.0.

The industrial value of Honor MagicOS 8.0, using AI large model experience and ecological reconstruction of business models

Starting from the above discussion direction, how is Honor MagicOS 8.0 doing? I believe that everyone already has the answer to this question based on the wonderful functions and scenes demonstrated at the press conference.

For example, the Honor Anywhere Gate that interests me the most is based on the rapid understanding of the user's true intentions, allowing everyone to travel through different applications with one click, achieving a fast closed-loop multi-tasking, and bringing a new human-computer interaction experience. .

On the Honor Magic6 series, you only need to long-press the text, pictures, or files on the interface and drag them to the edges of the left and right screens to trigger the Honor any door, and then quickly transfer to other application services, reducing the need for traditional cross-application Steps and times of service operations.

For example, when you need to take a taxi, you only need to press and hold the text message containing the address and drag it to enter the taxi app and start taking a taxi. In the past, you had to first find the text message, then select the text containing the address, then copy and switch to the desktop to find and open the taxi app, paste the address bar, a series of tedious operations, but now it only takes one step to complete.

The "soul" of this function is that it is based on Honor platform-level AI. The moment you trigger any Honor door, Magic OS completes the recognition of user intentions. It can accurately understand and predict Determine what you might want to do next and put atomic application services in front of you.

This is the essential difference between Honor Any Door and all previous "one-stop" functions.

Currently, this function already supports scenarios and services such as taxi hailing, navigation, search, shopping, password resolving, printing, sharing, and global collection.

Similar functions include the dialogue films, smart capsules, etc. demonstrated at the press conference. These functions all reflect the leadership of human-computer interaction based on intent recognition. When other mobile phones have built-in large models, they basically just stay there. When generative AI dialogue is an early adopter, MagicOS 8.0 has been able to go deep into the bottom of the system and give platform-level AI capabilities to each module to effectively help users complete tasks. Moreover, you only need to directly express what you want, and the rest is left to AI.

Fang Fei believes that this interactive logic is a "two-way rush", which can make the mobile phone become the user's "twin". "It can let you find services when you want to find them, and recommend services when you really need them." for you."

Behind these is actually a logical change from "people looking for services" in the past to "services looking for people" now. In the past, in the early days of the development of smart phones, a phone had small memory and few installed applications, so the services it could provide were natural. There are only a few, and it is easy to "find services". However, today's smartphones often carry hundreds of GB or even 1 TB of data, and there are hundreds of applications that provide massive services. It is increasingly difficult to quickly find the services you want. . On the other hand, letting the service find people is an inevitable choice.

In addition, if you want to do "AI large model-driven intention recognition" well, it is not enough to rely on the operating system alone. It requires the cooperation of the operating system and application developers.

For example, in the taxi-hailing scene above, Deng Bin, President of Honor R&D Management, shared the details of the cooperation with Amap. AutoNavi recently released a safe travel model based on artificial intelligence and big data technology to provide the public with a new model of smarter and safer travel services. And when it is combined with the Honor Magic large model, there will be more accurate intention recognition, providing the industry ecology with environmental perception, planning, and decision-making of time and space locations, and generating more and better applications.

As Deng Bin said:

When cooperating with the Gaode model, will our efficiency in completing a specific task be improved? Will the speed of quickly searching for the last travel location be faster? Will we be able to complete a specific item at a location? Whether purchasing, visiting specific attractions, or purchasing specific tickets will be more effective? This is a solution we create together with various professional platforms, professional tools, and professional models around the complete behavior of users. Everything can be calculated, everything can be software-enabled.

In this process, we can undoubtedly glimpse a greater value behind Honor Magic6: through the combination of the magic AI large model and the operating system, it fundamentally changes the relationship between the operating system and developers. of transfer to jointly face user scenarios and co-create production models and development models.

The changes in this relationship will further reconstruct the business model of the Internet. A "model-as-a-service" technology and industrial ecology is gradually taking shape, bringing huge room for imagination.

In the roundtable forum exchange, Fang Fei, President of Honor Product Line, also further elaborated on this value:

We are a scenario-based operating system that better understands consumers’ real-time intentions and thoughts.

Combined with large models and localization capabilities, we can break down barriers and gradually build a new service model. This is a huge opportunity for the future development of terminals and will simultaneously bring new developments in the application market. change.

In fact, Honor has also seized the opportunity and started to build a service ecosystem based on large models. At this press conference, it announced the arrival of the "Honor 100-Mode Ecosystem Plan" with the purpose of achieving win-win results with millions of developers. Create, work with Internet developers to create a production model for users, and lead a new chapter in the development of OS.

It also includes the industry's first "large model central control open architecture", which realizes safe and efficient links of multiple large models based on user intentions, providing each user with better service distribution and closed-loop experience.

On this basis, the Honor Magic Model can not only work with Baidu Intelligent Cloud Qianfan Model Platform, but also the device and cloud to collaboratively generate professional content. At the same time, it can also be linked with large vertical models of traditional applications such as Amap and Hanglv Zongheng to bring new experiences and personalized services.

During the exchange, Deng Bin defined the MagicOS experience of "the more you use it, the better it becomes, and the more you use it, the better you understand" under the joint enablement of this large device-cloud model, as, "You don't know that you can live a better life. Don’t know you could do better”.

In the future, general large models on the cloud side and large models in various industry verticals can be integrated into MagicOS through "shelf-leveling" to jointly build a new ecosystem in the AI ​​era.

Conclusion

The popularity of AI large models has lasted for more than a year, during which we have witnessed the scene of the "Thousands of Models War". However, after experiencing the madness of "reinventing the wheel", the entire technology industry has gradually realized that only by actually implementing large models into application scenarios and onto users' terminals can its value be truly released. Smartphones are the best terminal carrier at this stage.

China has the world’s largest smartphone market and industry cluster. In the Chinese mobile phone market, as the pioneer and market share leader in mobile AI, Honor is unrivaled. According to IDC's China smartphone shipment rankings for the third quarter, Honor once again topped the Chinese mobile phone market with a market share of 19.3%.

Honor has launched a unique strategy at the beginning of 2024. It not only brings the innovation-led Honor Magic 6, but also explores intention recognition human-computer interaction with MagicOS 8.0, supports platform-level AI capabilities with magic large models, and convenes AI with the Hundred Model Ecological Plan. China's innovation power has demonstrated its actions and determination to build an AI large model ecosystem and lead the next innovation cycle. Glory, which has entered its fourth year of independent development, is rooted in China and faces the world, officially triggering the starting gun for a new journey of "World Glory".

The above is the detailed content of Explore the impact of the new operating system scenario under the Honor AI trend on the future. For more information, please follow other related articles on the PHP Chinese website!