AI is one of the hottest trends in technology and it has a big role to play in mobile networking. The use of AI can lead to autonomous network operations over time, and the use of generative AI can improve customer support and service innovation.
These are the points I shared during an interview I had with Yanitsa Boyadzhieva of Telecom TV as part of Rakuten Symphony’s sponsorship of the AI-Native Telco Summit, an online event examining how to make AI a key technology for network operations and service delivery.
As I told Yanitsa, I am most excited about the prospect of autonomous network operations. The five-level pattern for AI-based autonomous systems has been set by the automotive industry, and it applies to mobile networks as well.
The five levels that are defined by the SAE International (technically there are six levels – level 0 is no automation) are: driver assistance (level 1), partial driving automation (level 2), conditional driving automation (level 3), high driving automation (level 4) and full driving automation (level 5).
This hierarchy can work in mobile networks. What excites me is hitting level 3; once a telco crosses this barrier, then there can be autonomy in the processes of monitoring, analyzing, decision making and action orchestration.
Some networks are already there. We recently deployed a use case that uses an ML model to audit the configuration parameters across the entire network and update an inventory replacing a “golden config” approach. The model looks for outliers and updates the database to ensure compliance.
It’s these kinds of closed-loop applications that are so exciting because they bring autonomy to the network and the ability to learn and adapt.
While closed-loop automation benefits the network, generative AI (Gen AI) brings benefits to other parts of the network. In my interview with Yanitsa, I mentioned three areas where I expect Gen AI to have a big impact:
I am realistic about AI – with the great benefits come some challenges. There are technology and transformation challenges. It’s very important that the data scientists developing AI models can get into production quickly. Right now, it can take months to deploy an AI model – that should be weeks or days.
The other technology challenge is making sure the data set is complete. Many MNOs have pockets of data in their network, but they need to have one place where they can find all the data or have a metadata dictionary, they can use to search through to find out what a particular table means, and which are the common primary keys across that you can write a model on.
Perhaps the bigger challenge for AI success revolves around people and processes. A mindset shift is needed by employees to fully embrace AI and its benefits. What we found is that it is important that all the people on the team are aligned with the goals for AI. A lot of times we see that there is this inherent fear around that AI will take away jobs, or AI is like a nuclear button that will destroy the network or cause an outage. There are even discussions around who is accountable if AI causes a network outage.
So, there are a lot of these kinds of apprehensions and fears that are leading to some slower progress than is possible. What is needed – and is emerging - are ways to mitigate these challenges in order to develop a pragmatic AI strategy.
To conclude our interview, Yanitsa asked what success looks like. For me, success looks like having an organization that is ready for AI to succeed, that is the organization must have the right people, processes, a single comprehensive, digitized dataset and an open network that allows for innovation.
This enables AI to be embedded into the network processes and to build a backlog of AI projects that grows as people see the success and as management sees successful KPIs.
If you are charged with developing AI for your network, I encourage you to watch the whole interview on the Telecom TV website.