Generative AI is shaping the longer term of telecom network operations. Potential applications to enhance network operations include predicting the values ​​of key performance indicators (KPIs), forecasting traffic congestion, enabling the transition to prescriptive analytics, providing design consulting services, and acting as an assistant to network operations centers (NOCs).
In addition, generative AI can revolutionize driving tests, optimize resource allocation within the network, automate fault detection, optimize truck trips, and improve the shopper experience through personalized services. Operators and suppliers are already recognizing and capitalizing on these opportunities.
However, challenges remain by way of speed of implementing generative AI-powered use cases and avoiding siloed implementations that hinder comprehensive scaling and optimizing return on investment.
In a previous blog, we introduced the three-layer model for efficient network operations. The primary challenges related to applying generative AI on these layers are:
- Data layer: Generative AI initiatives are data projects at their core, with inadequate data understanding being one in all the primary difficulties. In the telecom space, network data is commonly vendor-specific, making it obscure and use efficiently. It can also be scattered across multiple operational support system (OSS) tools, complicating efforts to achieve a unified view of the network.
- Analysis level: Base models have different capabilities and applications for various use cases. The perfect base model doesn’t exist because a single model cannot consistently address similar use cases across different operators. This complexity arises from different requirements and unique challenges that every network presents, including differences in network architecture, operational priorities, and data landscapes. This layer hosts quite a lot of analytics, including traditional AI and machine learning models, large language models, and highly customized base models tailored to the operator.
- Automation level: Foundation models are great for tasks similar to summarization, regression, and classification, but they should not standalone solutions for optimization. While foundation models can suggest different strategies to proactively address predicted problems, they can not discover the very best strategy. To To evaluate the correctness and impact of every strategy and recommend the optimal strategy, we’d like advanced simulation frameworks. Foundation models can support this process, but cannot replace it.
Key considerations for generative AI in any respect three levels
Rather than providing an exhaustive list of use cases or detailed framework specifications, we’ll highlight key principles and techniques that concentrate on effectively integrating generative AI into telco network operations in any respect three levels, as shown in Figure 1.
We would really like to emphasise the importance of sturdy data management, tailored analytics and advanced automation techniques that overall improve network operations, performance and reliability.
1. Data layer: Optimizing telco network data using generative AI
Understanding network data is the start line for any generative AI solution in telecom. However, each vendor within the telecom space has unique counters with specific names and value ranges, making the information obscure. In addition, there are sometimes multiple vendors within the telecom landscape, adding to the complexity even further. Learning these vendor-specific details requires specialized knowledge that will not be all the time available. Without a transparent understanding of the information they’ve, telcos cannot effectively develop and deploy generative AI use cases.
We have found that architectures based on retrieval-augmented generation (RAG) may be extremely effective in addressing this challenge. Based on our experience from proof-of-concept (PoC) projects with customers, these are one of the best ways to leverage generative AI in the information layer:
- Information on supplier data: Generative AI can process extensive supplier documentation to extract vital details about individual parameters. Engineers can interact with the AI ​​using natural language queries and receive accurate answers immediately. This eliminates the necessity to manually search through complex and extensive supplier documentation, saving a whole lot of effort and time.
- Create knowledge graphs: Generative AI can robotically create comprehensive knowledge graphs by understanding the complex data models of various vendors. These knowledge graphs represent data entities and their relationships and supply a structured and connected view of the seller ecosystem. This contributes to higher data integration and utilization within the upper layers.
- Data model translation: With a comprehensive understanding of various vendors' data models, generative AI can move data from one vendor model to a different. This capability is critical for telecom firms that must harmonize data across different systems and vendors while ensuring consistency and compatibility.
Automating the understanding of vendor-specific data, generating metadata, constructing detailed knowledge graphs and enabling seamless data model translation are key processes. Together, these processes, supported by an information layer with RAG-based architecture, enable telcos to unlock the total potential of their data.
2nd level of study: using different models for network insights
At a high level, we will divide the use cases of network evaluation into two categories: use cases that revolve around and use cases that
For the primary category, which incorporates advanced data correlation and creating insights into the past and current network state, operators can leverage large language models (LLMs) similar to Granite™, Llama, GPT, Mistral and others. Although the training of those LLMs didn’t necessarily involve structured operator data, we will effectively use them together with multi-shot prompting. This approach helps bring additional knowledge and context into the interpretation of the operator data.
In the second category, which focuses on predicting future network state, similar to predicting network outages and predicting traffic load, operators cannot depend on generic LLMs. This is because these models lack the obligatory training to work with network-specific structured and semi-structured data. Instead, operators need base models tailored specifically to their unique data and operational characteristics. To accurately predict future network behavior, we’d like to coach these models on the particular patterns and trends unique to the operator, similar to historical performance data, incident reports, and configuration changes.
To implement specialized base models, network operators should work closely with AI technology providers. Establishing a continuous feedback loop is crucial, where you frequently monitor model performance and use the information to iteratively improve the model. In addition, hybrid approaches that mix multiple models, each specialized in numerous elements of network evaluation, can improve overall performance and reliability. Finally, bringing in human expertise to validate and fine-tune model results can further improve accuracy and construct trust within the system.
3. Automation level: Integrating generative AI and network simulations for optimal solutions
This layer is liable for determining and enforcing optimal actions using insights from the analytics layer, similar to predictions of future network state and network operational directives or operations team intent.
There is a standard misconception that generative AI can take over optimization tasks and determine the optimal response to predicted network conditions. However, for this to occur, the automation layer must integrate network simulation tools. This integration enables detailed simulations of all potential optimization actions using a digital network twin (a virtual replica of the network). These simulations create a controlled environment for testing different scenarios without affecting the live network.
By using these simulations, operators can compare and analyze results to find out the actions that best meet optimization goals. It is vital to notice that simulations often use specialized base models from the evaluation layer, similar to masked language models. These models allow manipulation of parameters and evaluation of their impact on specific masked parameters within the network context.
The automation layer leverages one other set of use cases for generative AI, namely those actions triggered by network insights or human-provided intent require bespoke scripts to update network elements accordingly. Traditionally, this process has been done manually in telcos, but with advances in generative AI, there’s potential for automatic script generation. Architectures using generic LLMs augmented with retrieval-augmented generation (RAG) perform well on this context, provided operators ensure access to manufacturer documentation and appropriate methods of procedure (MOP).
Generative AI will play a major role in future telecom operations, from predicting KPIs to acting on network insights and user intent, however it is critical to deal with challenges similar to efficient data understanding, specialized predictive analytics, and automatic network optimization. IBM has practical experience in each of those areas and offers solutions for efficient data integration, specialized base models, and automatic network optimization tools.
Do you ought to implement generative AI use cases in your network? Bring us your use case and allow us to unlock its full potential. Contact us at maja.curic@ibm.com and chris.van.maastricht@nl.ibm.com.
Shape the longer term of your enterprise with AI
