Communication Service Providers (CSPs) are revamping their networks in preparation for the era of Artificial Intelligence (AI): making substantial investments in AI solutions to tackle current challenges, fostering collaboration across organizations (e.g., Artificial Intelligence Operations (AIOps)), and restructuring internally (e.g., centralizing AI teams). Despite the thoroughness of these adjustments for AI integration, CSPs have yet to streamline network architectures around AI. To address this gap, a fresh approach to network automation has surfaced, focusing on enhancing the synergy between intelligence and Information and Communications Technology (ICT). The proposal advocates for embedding the latest intelligence capabilities within an End-to-End (E2E) network and service management layer, incorporating a variety of AI models while consolidating AI within the intermediate layer throughout the telco stack to facilitate ongoing feedback between ICT and intelligence.
Envisioning the AI Future
1. Industry Challenges
Presently, generative AI holds the promise of revolutionizing network operations as a universal solution, although its application remains limited to specific use cases. These models are tailored for targeted applications such as:
- Network Planning: Providing routing recommendations, conducting controlled digital-twin experiments
- Network Analytics: Offering configuration suggestions, detecting incidents
- Network Operations: Proposing or documenting code, generating synthetic data
- Resource Efficiency: Improving compression through encoding/decoding techniques
While customizing models for specific purposes is crucial and will persist, the fragmented distribution of AI resources across network layers remains a hindrance. Despite AI’s broad applicability across network domains, the segregation of AI resources across layers impedes the pooling of AI capabilities and resources essential for cost efficiency and synergy with ICT.
Furthermore, the proliferation of generative AI contributes to increased data traffic across networks, a trend that will escalate as AI evolves to encompass new content formats, interactive modes (e.g., Virtual Reality (VR)), and connectivity paradigms (native-full connection). To leverage these advancements, CSPs require intelligent and adaptive ICT architectures informed by AI practices. The absence of AI intelligence shaping ICT decisions poses a critical question: Why isn’t ICT guided by AI intelligence and methodologies? Establishing AI as a consistent intermediary in ICT upgrading decisions is imperative.
These pressing issues surfaced prominently at the Mobile World Congress (MWC) in Barcelona 2024, challenging the telecoms sector and scrutinizing its embrace of AI.
2. Innovative Approach
A novel perspective on AI’s role within networks is emerging to tackle these challenges effectively.
Primarily, it entails enriching the network and service management layer with cutting-edge AI technology to establish an AI-driven, cost-effective network core capable of disseminating intelligence across the network seamlessly. This approach enables the deployment of diverse AI applications and models (both traditional and generative) across the network while consolidating resources for improved visibility, operational efficiency, and interaction with ICT. The intelligent network core focuses on fundamental AI model development tasks (e.g., training, tuning, inferencing), subsequently optimizing data management and scheduling within the network. It further extends to serving users as they engage with AI in increasingly sophisticated manners.
To support this evolution, ICT must adapt to the evolving AI landscape distributed across the network. This entails reinforcing cloud infrastructure for model training, optimizing network capabilities to facilitate the transfer of large model parameters for local processing, and establishing a scalable edge to support model inferencing. Such adaptations equip ICT to support the intelligence layer amidst the anticipated surge in user-generated data.
This architectural framework must also adhere to industry standards. For the requisite agility and scalability, CSPs must prioritize End-to-End (E2E) cloud-native architecture, incorporating automated container orchestration and microservices.
The Potential of AI Transformation
1. Current Market Landscape
Generative AI is still in its infancy concerning network management, necessitating continuous third-party innovation for sustainable progress. Encouragingly, solution providers are learning to channel generative AI through deterministic and rule-based environments to mitigate risks effectively. CSPs are also exploring lower-risk applications, such as generating synthetic data or integrating with traditional models for anomaly detection. While advancements are evident in the current market landscape, generative AI predominantly impacts customer care and the Operating Support System (OSS)/Business Support System (BSS) domains.
2. Strategic Evaluation
At this juncture, the application of generative AI in network contexts is predominantly focused on specific areas due to the limited number of use cases available. This strategic focus on narrow applications serves the dual purpose of minimizing risks and addressing immediate network challenges.
Nevertheless, indications suggest that the current market phase is progressing, with network hurdles being surmounted and user-generated data steering ICT towards new horizons. CSPs are compelled to keep pace with these advancements to avoid falling behind.
The advancements in network AI align with the strides made in network automation. During the Level 2 and Level 3 autonomy phase (2022 to 2028), CSPs face the challenge of formulating a comprehensive AI strategy and operations framework: integrating generative and traditional models, establishing guardrails, surmounting ICT obstacles for model training/tuning, and setting norms for AI explainability and content trust. Progressing to Level 4 and beyond (post-2028) demands the distribution, scaling, and automation of hybrid AI solutions through 5G cloud-native infrastructure, saturating operations with generative AI content. Transforming the network and services layer into a universal intelligence hub, powered by cutting-edge AI technology and facilitating continuous ICT feedback, aligns with both short-term and long-term objectives: holistic operations, widespread network AI integration, and intelligent ICT expansion. This vision is not only forward-thinking but also timely.
The advantages of this vision over conventional approaches are evident when considering the following trends:
- Anomaly detection and natural language-based service orchestration are popular test cases for generative AI within network environments. By consolidating these use cases within the same intelligence layer, efficient model training and cohesive network orchestration can be achieved compared to their segregation by network domain. As new network use cases emerge for generative AI, amalgamating them into a unified intelligence fabric enhances operational efficiency. Notably, both use cases can be distributed across the network.
- The present cloud transformation era has rendered ICT more scalable, elastic, and responsive to network demands. However, creating truly intelligent ICT necessitates more than cloud capabilities. Centralizing the intelligence layer empowers CSPs to gain insights into AI-driven network requirements, resource allocations, and data assets, facilitating smarter responses network-wide. While aligning with cloud transformation trends is crucial, optimizing networks for generative AI necessitates additional measures.
This visionary approach is poised to deliver solutions that surpass conventional methods by enhancing network efficiency and potentially elevating service quality. ABI Research underscores that the most time-intensive stages of generative AI network implementation involve enhancing network transparency, developing ontologies and policies, and formalizing these for model tuning, as well as integrating generative AI seamlessly into broader network processes while establishing guardrails. The proposed synergy between AI and ICT addresses both aspects, maintaining data within an intelligence loop for enhanced visibility and processing, and centralizing intelligence within the network and services layer.
3. Concluding Remarks
Consumers play a pivotal role in this vision, being the primary recipients of intelligent AI services and the driving force behind data generation funneled through the network. Given the centrality of consumer AI services, the application of this AI and ICT vision to alternative end users warrants consideration. Particularly, the anticipated trends in enterprise AI adoption and their capacity to support an active intelligence layer and adaptive ICT merit exploration, especially for CSPs offering enterprise AI solutions. The exponential growth in consumer AI utilization and the ensuing surge in network traffic are likely to compel most CSPs to embrace AI reform.
Harnessing these consumer-driven AI trends necessitates ICT investments that transcend standard cloud transformation initiatives. In addition to cloud capabilities, CSPs must ensure that their network architecture facilitates the creation, distribution, and optimization of AI through the cloud. Centralizing the latest AI resources within the network and services layer enables seamless integration of new generative AI applications as they emerge, dynamically reallocating existing AI resources as necessary. This approach also mandates robust ICT infrastructure capable of handling resource-intensive tasks like model training at the network core and inferencing at the edge.
