Orange's CTO examines pros and cons of GenAI

This is the final part of a three-part interview with Laurent Leboucher, Group CTO and Senior Vice President of Orange Innovation Networks. Here we discuss the potential for generative AI in telecoms and why more collaboration among standards-development organizations (SDOs) and open-source groups is needed to advance automation. Leboucher sees the potential for groups to work together on what he calls a “horizontal operating model” for the network.

Part one of our interview focused on Orange’s goal to reach Level 4 Autonomous Network for some processes by 2025, and part two focused on why intent-based automation is so critical as Orange transforms its network to be cloud native and software defined. Our conversation has been edited for clarity.

DB: What are your thoughts about generative AI and its usefulness in telecoms?

LL: Obviously, generative AI is the hot topic today. Within Orange we have a positive view on the opportunities. We are just at the beginning, so we will be in a testing mode to start with, to understand.

We know that there are some risks that we will need to mitigate, especially what we call ‘hallucinations’, meaning that in some cases the behavior of generative AI could lead to wrong answers. We need to understand in depth when that happens and how we can mitigate this kind of risk if we want to apply it to network automation.

The way we want to start is by using and testing generative AI in our own processes, across the business and networks in Orange. We are very careful about the compliance, confidentiality and ethical aspects, and we also have some guidelines on how we allow people to use generative AI.

For instance, with some solutions, the lack of description about how data is managed according to laws and regulations (for example, GDPR) can be a problem. So, we need to work with only selected partners. We are also very careful about transparency. Whenever we use generative AI for a specific problem, even if it’s protected for security reasons in a safe environment, we need to review outputs and be aware internally.

So, yes, the answer is we are very positive on generative AI. At the same time, we need to embrace it in a safe way. We need to go step by step, and we are eager to learn from [other CSPs].

DB: What are some potential use cases that interest you?

LL: There are a bunch of things that we could dream of with generative AI, starting by facilitating automation scripting or such as using it as an assistance for all the interactions between our network operating center (NOC) and field technicians, who need to interact with NOC in a very efficient way. We also uses AI in a more traditional way by leveraging all the capacity of machine learning and also graph vector representations of the network, it is possible to drill down into problem resolution and then feed that to technicians in a way which is easy to understand by people on the ground so that they can be as efficient as possible in the interaction.

There is also a general problem, which is interesting as well. When you want to create shared capacity in a telco that can be used by different affiliates, people are often using different languages (e.g. Spanish, Slovak, Romanian, Polish, French, English, …) for instance for ticketing. At some point, if we’re able to translate languages in real time and very efficiently, we can solve the Babel problem.

DB: I also wanted to ask you about standardization and the need for collaboration among standards bodies and open-source groups as communications service providers move toward network automation.

LL: In order to move to the next level in terms of automation, we need to introduce something I have not described in detail, but we call it a ‘horizontal operating model’, which is necessary for automation across domains (a condition for level 4, as automation must be done end to end). The model needs to leverage a common telco cloud stack, on top of which we implement GitOps at scale and pipelines for network functions and network services chains aggregating different network functions. If we want to do that, we need to be very clear about the operating model (who does what, who is responsible for what – for example, for onboarding new partners, to- solve a problem and solve the root cause, etc.), so that everyone understands it in detail and so that we can also map roles onto this operating model different from today vertical operating model by technologies. I believe that this is something where NGMN could play a quite interesting role with TM Forum, and we’d like to see if we could do that together.

Also, when we speak about intent-driven management, we want to apply it to specific use cases on the transport, IP, fixed, and radio access network, and also to the radio network of the future in the open RAN context. For all these use cases, defining intent-driven management is necessary. A little bit like we in TM Forum started the Open API movement eight years ago, I think we could do exactly the same for intent-driven management. It’s very important to be open, with clear northbound APIs, so this is an area where TM Forum could play a role.

Another area could be the methodology, the assessment and maybe also benchmarking so that we can look at this transformation over a period of time between now and maybe the next five or six years, because it will take time to reach the goal that we all want to achieve. TM Forum could also provide services to help educate operators and vendors. The Forum has some very interesting assets because it traditionally brings together IT for telco. Now, we need to bring those IT practices to the network with network vendors.