AI Ushers in a New Era of Intelligent Communication

The rapid development of AI technology has become a core driver for a new wave of technological revolution and industrial transformation. From network O&M to new service and capability innovation, AI is reshaping every aspect of the communications field, and is set to have a profound impact on the development of the core network.

Large Models Drive a Leap in AI Capabilities

Large models are driving a qualitative leap in AI capabilities. From natural language understanding and text generation to multimodal interaction, time-series prediction, and complex reasoning, AI has achieved remarkable progress.

Scaling from 100 billion to 10 trillion parameters, large language models (LLMs) demonstrate strong performance in semantic understanding, emotional analysis, text generation, and emergent capabilities, subverting how industries operate. The Transformer architecture has expanded from language processing to more domains such as computer vision, speech recognition, structured data, time-series prediction, and multimodal processing, enabling a huge leap in the capabilities of related domain models. Multimodal LLMs (MLLMs) excel in extending modality and generalizing application scenarios. Time-series models show great potential in power load prediction, energy scheduling optimization, weather prediction, and disaster prediction. Since 2024, mixture of experts (MOE) has developed rapidly, significantly reducing inference costs and making large models more accessible.

At the application framework level, AI agent technology is entering its inaugural year of development. It integrates memory and planning capabilities with LLMs and rapidly builds agents for a variety of application scenarios by linking different tools. Retrieval-augmented generation (RAG) greatly accelerates knowledge injection into large models, driving the deployment of AI applications such as enterprise knowledge management and intelligent Q&A. Emerging architectures and  technologies, such as multi-agent coordination, large-small model coordination, and cloud-edge-end collaborative inference, are promoting the integration of large models into complex scenarios.

AI applications are becoming a key business driver for cloud vendors, and an important engine for enterprises to reduce costs and improve efficiency. Large model-based applications are expanding from consumer to business domains, driving digital transformation across industries.

AI Igniting a New Wave of Network Intelligence

In 2024, major operators and equipment vendors worldwide released AI strategies to advance network intelligence, conducting extensive research and practices in domain-specific models, application scenarios, and product solutions.

At MWC 2024, the Global Telco AI Alliance (GTAA) was officially launched by Deutsche Telekom, e& Group, Singtel, SoftBank and SK Telecom. The GTAA aims to develop LLMs for the telecom sector, offering more trustworthy, lower-cost, and more efficient Telco LLMs for members, and enabling AI applications such as virtual agents, fraud filtering, and personal AI assistants.

Deutsche Telekom’s app-free AI smartphone unveiled at MWC 2024 is visionary. By applying LLMs, it provides users with a unified portal for trusted access to internet services and the AI world.

Telefonica has formulated a comprehensive AI strategy spanning from strategy formulation to implementation and deployment. The strategy is being implemented at multiple levels, including customer service, service processing, user experience, and content platforms.

The GSMA proposes that AI has brought unlimited potential to the telecom industry but emphasizes the need for responsible AI, advocating for the establishment of industry specifications through principles, strategies, and standards.

In China, the three major operators have released self-developed LLMs for the network realm. China Mobile has launched a series of network LLMs, including a network natural language model, a network structured data model, and a network vision model, to enhance the value of autonomous networks. China Unicom introduced its Yuanjing 2.0 LLM, an updated version to Yuanjing 1.0 designed to empower various sectors. China Telecom unveiled its Xingchen LLM, which includes the Xingchen semantic model, Xingchen voice model, and Xingchen multimodal model. In terms of AI applications, China Telecom was the first to propose using its self-developed Xingchen network model within its enterprise for tasks such as monitoring, troubleshooting, maintenance and optimization processes, improving troubleshooting efficiency by more than 30%.

Standards related to AI in telecom are advancing rapidly. 3GPP defines the network data analytics function (NWDAF), which enables network performance and efficiency improvement on the intelligent plane. AI endogeneity has become a key trend and important research direction in 6G standard development. The TM Forum (TMF) is introducing generative AI into autonomous networks, exploring application scenarios, frameworks, and implementation solutions. CCSA in China is leading the formulation of protocols and specifications to enable network intelligence through multiple large models.

Deep Integration of Core Network and AI, Shaping the Future of Intelligent Communication

In the 5G and 5G-A phases, network capabilities have been greatly improved, yet challenges such as high construction costs, complicated O&M, and insufficient value realization remain. AI technology provides strong support for network intelligence. By incorporating AI and edge computing, the 5G-A core network enhances network intelligence, reduces O&M costs, improves computing network resource efficiency, optimizes network service quality, and expands value-driven service scenarios.

While networks are moving toward multi-factor integrated services, O&M is advancing to higher-order intelligence. Intelligent technologies are required to automate O&M in complex scenarios and tasks. Starting with assistants for interaction, analysis, and generation, the evolution will progress towards decision-making and control scenarios, based on model accuracy and task risk assessments. For example, large models for the O&M field are introduced to link existing AIOps tools to enable the creation of agents for alarm analysis, fault diagnosis, and repair. Multiple agents can collaborate to orchestrate a cascaded flow for complex tasks, achieving automated network fault diagnosis and repair. Time-series models can analyze historical fault data and real-time network operation data to facilitate proactive fault prediction, and automatically take corresponding repair measures or provide optimization suggestions.

As 5G networks rapidly advance and user requirements diversify, network operations need to shift from scale-based to performance-based—focusing on  user value mining, expanding network and service innovations, and providing differentiated and refined services for users. At the product design level, AI and large models are leveraged to support operation analysis, product design, and user retention. In terms of service innovation, large voice models are used to recognize user intent via voice interaction, addressing the issue of poor interaction in the new-calling video mode. LLMs and MLLMs enable real-time translation and interesting calls. Meanwhile, large anti-fraud models perform emotional analysis and intent recognition, solving the problem of low accuracy in traditional rule-based fraud call detection. At the service level, telecom intelligent customer services powered by large language models evolve from menu-based interactions to natural language interaction, from semantic understanding to emotional perception, and from domain experts to encyclopedia-level experts, offering 24/7 high-quality online services.

The vision of 6G is to create a more intelligent network environment. AI for Network will evolve from a plug-in feature to an endogenous capability, with AI integrated into all layers of the network architecture. At the same time, AI applications will become ubiquitous, and Network for AI will become the main theme. The core network should prioritize AI applications as its primary target scenario, enhancing its architecture, capabilities, and performance to support more extensive and complex scenarios.

Innovating architecture for network and AI integration: The 6G core network needs to innovate its network architecture to deeply integrate computing, network, and AI, achieving endogenous AI. For example, by deploying a collaborative computing and network platform at the network edge and in the cloud, the AI model can be trained and inferred in the most appropriate locations. Edge nodes will process AI tasks with high real-time requirements, while the cloud will be responsible for large-scale and complicated AI computing. The network will intelligently allocate tasks to different nodes.

Meeting complex AI application requirements: Based on integrated communication-sensing and communication-intelligence, the network can automatically perceive the computing and network resource requirements of AI tasks—such as large-scale deep learning model training or inference—and dynamically schedules network topology, bandwidth, and computing nodes to better support complex new AI services. With an integrated space-air-ground-sea network architecture, the system can provide seamless connectivity for widely distributed AI sensors and computing devices,  providing real-time massive data transmission for immersive services such as holographic communication, immersive communication, and glasses-free 3D, ensuring high-quality information transfer and presentation. It also guarantees sufficient bandwidth, low-latency performance, and data security for large-scale AI model training.

The deep integration and mutual reinforcement between the core network and AI is an inevitable trend. The development and application of AI and large models drive the evolution of core network intelligence. In turn, the core network provides powerful connectivity and computing support for AI, meeting users' requirements for high-quality communication and intelligent services, while providing the foundation for the digital transformation of industries.