AI, in its different forms: predictive, descriptive, prescriptive, or generative, as well as computer vision, natural language processing, or speech recognition is, or will quickly become part of our daily life. Probably the most known type of AI is Generative AI, popularized by models such as GPT, Gemini, Claude, Llama, or Mistral.
AI in Telco: Standardization and Beyond
As in most industries, the major players in the mobile communication space have been looking at ways in which various types of AI can be used in the telco world. 3GPP started to examine AI/ML already with Rel. 15, but the first formal study was done in Rel. 17 and continued throughout Rel. 18. The initial studies primarily focused on Network Data Analytics Function (NWDAF) enhancements with predictive, prescriptive, and descriptive AI as well as AI for anomaly detection. Rel. 19 set the foundation for AI in 6G, researching how AI can be applied to classical, but critical procedures such as mobility management, beam management, or channel estimation.
6G, for which the first study items will be completed in Rel. 20 is set up from the start to be AI-native, meaning that AI will not be only an add-on to the network, but will be embedded in the architecture, protocols, and decision-making processes. One key area for AI/ML in 6G will be the Radio Access Network (RAN), expanding the work on Channel State Information (CSI) and Mobility Management. Additional study items look at energy efficiency, AI/ML model lifecycle management, and AI explainability.
Apart from the standardization efforts, AI is being applied by vendors in the telco industry as a tool to enhance their products, improve experience, and create efficiencies for their users and the businesses they serve. One feature, that users have grown accustomed to, thanks to ChatGPT, Gemini, Claude, or Mistral, is having a natural language, conversational, interface, with which day-to-day tasks can be solved faster, without the need to learn complex tools or workflows.
AI Assistants for All
AI Assistants, typically powered by a combination of generative and deterministic AI, together with natural language input, have the potential to radically change the way users interact with the OSS/BSS tools used by modern CSPs. In the pre-AI world, engineers and staff who had to work with such tools required specialized training to get familiar with complex user interfaces, extensive parametrization, and often repetitive workflows. Even with Graphical User Interfaces (GUIs) becoming friendlier and more automated, in many cases, users of such tools, especially novice ones, cannot immediately benefit from the full spectrum of capabilities that the tools offer.
AI Assistants can radically change this. For example, working with systems for performance management and service assurance requires in-depth knowledge of the service being evaluated, the capabilities of the test system, the KPIs, and their optimal values. Furthermore, it requires knowledge of how to configure, report, and interpret the results using a particular tool.
What if we could ask a tool’s AI assistant to do all that? Users could simply chat with the AI Assistant and ask “Which services are showing bad performance?” or “Can you summarize the main voice and data KPIs for the last week?” or maybe “Can you set up tests for VoLTE with my top 20 roaming partners?”. This capability doesn’t only make the work of experienced engineers more efficient, by removing the repetitive and tedious overhead of interfacing, with the often complex UIs of classical telco tools. At the same time, AI Assistants lower the barrier to entry for beginners, allowing more stakeholders to leverage existing investments in tools and processes.
Building Trust Through Hybrid AI Models
Obviously, as with any technology that changes drastically the way we work, but particularly with AI, trust is a key factor. An AI Assistant must deliver consistent and correct results, particularly when used by beginners who might not be able to distinguish between a good answer and a problematic one coming from an AI Assistant.
One method to address this, being used in Mobileum’s AI Assistants, is to combine generative and deterministic AI. This method benefits from the creative problem-solving capabilities of generative AI while having the consistency and speed of deterministic AI. Deterministic AI is particularly suitable for tasks that require predictability, traceability, or address regulated topics, which are often encountered in telco workflows.
As the adoption of AI Assistants grows, users will be able to leverage the tools at their disposal to their maximum capabilities, without the steep learning curve typically required until now, maximizing the investment made in such systems. Furthermore, the elimination of configuration and administration tasks will allow staff to focus on business outcomes rather than tool management and administration.
Conclusion
The potential of AI Assistants is high, and it’s critical for telcos to embed them across their operations. Doing so can enhance user experience, unlock greater value for customers, and most importantly, enable better business outcomes with reduced operational effort.