AI front and center at MWC24
Rarely has a single theme dominated Mobile World Congress as much as AI has this year and I’m sure a good percentage of the hundred thousand delegates at the Barcelona event this year were feeling a little AI fatigue before the show. 15 months on from the launch of ChapGPT which triggered a worldwide obsession with GenAI and AI/ML more broadly, has much changed? Well, yes. Once we dig under the hype a little there are some genuinely interesting and impactful uses of AI that are ready for consumption by the telecoms industry today. I will write a more comprehensive round up of our time in Barcelona, but for now, here are three things that struck me about Gen AI this week:
Small language models (SLMs) are still large (possibly several billion parameters) but they are nowhere near the size of the large language models (LLMs) that telcos are already using, and as such they are compact enough to run on a mobile device. They can be trained on very focused data sets from specific domains within CSPs operations, so SLMs can be used for specific GenAI use cases like engineering co-pilots. I have heard several times this week that CSPs are concerned about the cost of LLMs and getting a good return on investment as they scale their GenAI capabilities, so SLMs may be a good way to speculatively build new applications before scaling.
Any data which is used by GenAI is key to its effectiveness in complex telecoms industry settings, but the current state of data models and architectures across network and service operations isn’t ideal. Several times this week CSPs have mentioned that addressing the way their data is stored will be a gateway to leveraging the maximum impact of AI. Defining an AI-ready standardised data model for CSPs is collaboration work currently going on in the TM Forum Modern Data Architecture Group, so we expect to see progress on this in the coming months.
One of the first use cases for Gen AI in telecoms that we predicted last year was help for CSRs when dealing with customer enquiries, complaints or sales. Many major vendors have now released this module into their BSS portfolios and CSPs deploying them in large numbers. Several have natural language interfaces so as the CSR is talking to the customer, the co-pilot is listening and pulling up the relevant information on screen. We’re also seeing co-pilots being used in a engineering settings and there are many other uses to add to the list. This is a prime example of Gen AI being used as a tool to augment human roles and make job functions more effective.