The Roadmap for Autonomous Networks and Dark Network Operation Centers

What do autonomous networks really mean for telecommunications? What could the future of digital networks look like? Step by step, we are turning this vision into reality - building a network that not only serves as the foundation for business but also transforms how society is organized. While we haven’t fully realized this ambitious dream yet, we have made significant advancements in both technology and understanding, bringing us closer than ever to making autonomous networks a reality.

What is NOC in digital networks?

The network operations center (NOC) has always been a critical part of managing telecommunications and IT networks. Even as these networks merge into unified digital techco systems, the importance of NOCs remains unchanged. Despite the decentralization of management and the creation of control elements at many levels and in many places, the NOC remains essential. This is because a central element is needed to implement policies of conduct in the management of the digital network, depending on customers' dynamically changing requirements.

As networks and services become increasingly complex, the demand for precise and efficient operations continues to grow without  any loss to availability and quality. The ability to respond swiftly to customer needs, adjust the scale and location of data streams, and implement quick yet secure network configurations has never been more critical. Interestingly, not all "customers" of digital networks have good intentions. Hackers, in their own way, influence network strategies just as much as legitimate users. Their activities drive the need for specialized centers solely focused on safeguarding digital systems. And with so much of our lives dependent on secure, reliable networks, these efforts are not just justified - they’re essential.

The role of NOC

As the role of NOCs is evolving, their importance steadily grows. Once primarily focused on repairs, NOCs now play a pivotal role in modifying and improving networks and services, reflecting several simultaneous trends. The rapid advancement of technology has accelerated the adoption of successive generations of equipment, with discussions about 6G arising even before the widespread implementation of 5G. At the same time, networks are becoming increasingly complex, integrating diverse technologies, systems, and multi-vendor equipment on a massive scale.

Additionally, while equipment prices have generally decreased, their performance and reliability have improved significantly, leading to the creation of highly redundant systems at every level, from individual components to entire networks. In today’s world, where efficient digital communication underpins both business operations and daily life, the demand for faster responses to outages, quicker implementation of changes, and the delivery of new services is higher than ever.

Dark NOC ensures consistency of actions, the possibility of dynamic change of the way of acting in new conditions, the possibility of an evolutionary transition from the classic NOC to the dark NOC. Instead of recreating real employees' expertise with AI, the aim is to create AI instances to solve recurring incidents and problems in the network. Virtual specialists are efficient and precise, learning quickly on historical data shared with suppliers and other operators and quickly implement company policies in a uniform and simultaneous manner in all areas. The idea is that assembling a team of many experts to work on repetitive tasks is difficult and, above all, expensive, while AI can handle them immediately.  The use of AI also opens up several possibilities that have been thought about for some time, i.e., new parameters for decision-making, tracking the results of implementing decisions through a larger number of observed parameters - which, of course, can also be done by employees, but more and more of them are needed.

Artificial intelligence in NOC and SOC

The changing demands for service availability, quality, and NOC operation speed aren’t new, but the growing complexity of digital security threats highlights the need for skilled professionals in NOCs and security operations centers (SOCs). At the same time, concerns about rising costs have pushed many organizations to adopt technologies such as artificial intelligence (AI) to streamline workflows and control expenses. Initially a tool for automating tasks and aiding decision-making, AI has advanced enough to operate autonomously under human oversight. This shift is now one of the fastest-growing trends in tech, with new innovations emerging almost quarterly.

Several factors have fueled this rapid development. The increasing performance of hardware has enabled AI solutions to run efficiently, while generative AI models, such as those based on large language models (LLMs), have expanded AI's capabilities and normalized its presence in both industry and everyday life, helping to reduce fears and misconceptions about this technology. Hardware and software providers are also collaborating to standardize solutions, ensuring compatibility and interchangeability, while virtualization has made it possible to move and install logical components seamlessly on physical hardware. Additionally, the adoption of cloud environments has given users greater flexibility to select, modify, and deploy optimal solutions according to their needs.

This convergence of advancing technologies and growing operational demands has created a powerful impetus for innovation. For many in the industry, these developments represent the realization of long-anticipated goals and open the door to unprecedented opportunities in the tech sector.

The challenges of implementing a dark NOC

Non-uniform network architecture

Developing a universal, independent dark NOC is difficult due to the diversity in network architecture. Each network consists of multiple parameters, such as differing equipment or security protocols, making it difficult to create fully universal solutions. This is why common features and historical data serve as a strong foundation for building effective localized AI models. To address this issue, Comarch specialists developed tailored solutions managed by multiple AI elements, creating a collaborative environment that quickly collects data to refine operating models. Early on, this approach accelerates AI systems' learning, which are ready for self-installation from the cloud, largely equipped with a package of basic machine knowledge and models.

The need for clear guidelines

Another issue is the potential for NOC to operate based on changing rules. Establishing effective operational policies is critical. Without clear guidelines, many benefits of AI cannot be realized. Traditional NOCs depend on engineers following documented policies to prioritize tasks based on customer needs, which is a time-consuming process. Updating these operational rules requires significant training and can jeopardize contractual obligations if not communicated effectively.

Further steps in dark NOC development

We have now arrived at an era where artificial intelligence agents sourced from various suppliers create a federation to autonomously govern networks. Competing and user-replaceable, they are mutually managed by artificial intelligence. In a transitional model, AIs can support decisions still made by people.

However, history suggests that the transition to the AI federation is inevitable. Today, we need a flexible model such as the one developed by Comarch to build trust in AI and show that the fears surrounding autonomous networks are unfounded. This approach does not contradict the concept of a dark NOC, where operational activities can already be entirely managed by multiple AIs.

Strategic decision-making, however, doesn’t have to occur within the NOC or follow a 24/7 model. Instead, human oversight remains crucial at a higher level, focusing on shaping strategies, defining rules of engagement, and establishing best practices. These efforts are supported through intent management interfaces designed to bridge the gap between human intentions and AI execution - but that’s a story for another time.