Large Language Models as the Power Behind Autonomous Networks

Autonomous networks are gaining increasing attention from telecommunications operators, to the extent that the subject now dominates at industry events organized by TMF. There is some debate about whether “autonomous” is just a new way of describing “self-organizing”, but setting that discussion aside we can say that one thing has certainly changed: ChatGPT has come of age.

ChatGPT – the right tool at the right time

Why is this important? Because it appears that, in ChatGPT, we finally have the right tool to build true network automation and autonomy. And there is a difference – an autonomous network is one that can be run by high-level defined intents (originating from business or customers), which dictates what is expected from the network (rather than how something should be done), and then decides how to realize that intent while continuously learning how to do that job better.

The journey towards fulfilling the dream of truly autonomous networks has been a long one, with many more steps ahead. Effectively, it’s been about how to apply AI/ML in real-life situations and benefit from that, which is an endeavor not confined only to the telecommunications industry.

Such endeavors have been boosted by the ChatGPT revolution and the advent of large language models (LLMs) which underpin ChatGPT. LLMs are expected to be transformative in efforts to a problem about data; that problem isn’t that telcos have too much or too little data, but about using such information effectively. LLMs make it possible for data in the form of texts consisting of real human languages (sometimes with informalities) to be utilized by AI/ML.

LLMs in action

One example of where this could be particularly effective is the assurance domain. Large operators which have been on the market for a long time could have a great many trouble tickets, often written by customers in their own language. Thanks to LLMs, such information can be used by machines to learn the patterns of symptoms (from customer and network perspectives) and how to resolve them based on instructions originally meant for humans to follow.  In fact, some operators have already begun work on this, and we may expect advances in the coming year that also lead to the realization of dark NOC.

The autonomous networks of tomorrow

The benefits of autonomous networks are clear. They reduce costs, but also increase revenue – for example in the field of automating service delivery. This can be particularly useful in B2B, especially services involving network slicing and private 5G. Monetizing both of these has so far been hindered by an inability to automate on the basis of customer intent (meaning, information provided by vertical-specific non-telco experts), but in the coming years we should see ambitious attempts to use LLMs in a way that allows data delivered in human language to be correctly interpreted for the automated delivery of the right services.