Rakuten Symphony’s OSS President Rahul Atri shares his vision for how artificial intelligence can drastically advance the way mobile networks are designed, deployed, operated and monetized - all through a system’s learned ability to understand a user’s intentions. He believes successful AI use is not only a technological challenge, but also one of ROI and sustainability.
Networks of the future will be fully automated and enabled by AI - they won’t need people to figure out things that are manually integrated. In the very near future, network operations teams will collaborate with AI tools to efficiently share their intent and deliver positive business outcomes. This is the next step in the evolution of data – from insight-driven actions to AI-driven actions.
Let’s think about how we solve problems as humans. We define a problem with a view to achieving an outcome - this is our intent. The experience we have acquired over the years in a specific domain is our context, and the outcome is the results we deliver. Let’s translate this to an MNO’s network operations - your AI companion should be able to offer intent and context-aware suggestions that can help to generate successful outcomes.
When Rakuten started investing in AI, it was always a Group-wide initiative, not just within Rakuten Mobile and Rakuten Symphony. All of Rakuten’s 70+ companies are thinking of AI as a service delivery platform. That approach comes from the company’s leadership - that in itself helps to create a DNA of AI use and avoids the bureaucracy it could otherwise bring. To echo the words of Rakuten Group Chief Data Officer Ting Cai, AI needs to be able to take our services to the “next level”.
An AI system needs to understand each individual user and their specific intent with regard to their domain or workstreams - it needs to understand that the user is asking a certain question and the logic behind why they asked it. Everyone, from sales teams, to RAN engineers and even the CEO, has their own specific objectives in terms of what is needed for them to be successful and effective. How can we find a way to make sure all of their varied - and not to mention deeply complex - goals are met?
Here’s how Rakuten Symphony plans to reach that level of sophistication.
We believe that the user’s intent must be understood by an AI system – as an organization, we are now investing a lot of energy and resources on the classification of intent and contextualization in order to find the most efficient solution to the problem. That’s across a number of scenarios - whether it is an NLP to SQL query, semantic search, triggering a workflow or generating a domain-specific prompt for large language models (LLM).
Every time an individual user speaks to the tool, that tool needs to become more intelligent and grow its understanding about the user and their domain and context. We want the AI to do those things for you so that you can do more meaningful work and do it more efficiently. Rakuten Symphony has already showcased working versions of this at MWC Barcelona 2024 - using live Rakuten Mobile data - and we are progressing rapidly.
Almost all types of dashboards and reports are traditionally very static. Whichever tools operators currently use, static dashboards can provide them with a certain level of insight, but that does not necessarily address the specific initial intent of the user that needed the answers.
What needs to happen is for LLMs to speak to your database and platforms in order to become more domain-specific - to drill down into the specifics of what the user needs. It means understanding context-related things such as who is requesting answers - a customer care agent, a radio frequency engineer or an executive. That requires the training of the data - so it knows that a senior executive is asking how well the 5G network is performing compared to the 4G network. The executive doesn’t need data on specific KPIs, just the ARPU on the 5G network compared to the 4G network.
In the near-term, I believe that we will all become AI trainers. Radio engineers will become ‘radio AI engineers’ - they will train AI models with their intent, their context, and they will teach the AI to solve the problem in their own specific way. In the long-term, we will all become domain-independent. We will even be able to exchange neural graphs – our own individual intents, contexts, knowledge and experience - so that we can share whatever knowledge we have acquired over the years.
Let’s look at one example. Today, fine tuning or customizing a radio resource scheduler is a real headache for radio frequency (RF) engineers and requires complex coding and expertise. But what happens if the network scheduler can be tuned via an AI tool and someone can convey their intent to a radio site in a football stadium - where people are constantly uploading Instagram videos - to tune itself for heavy uplink traffic? The scheduler can pre-emptively adapt to this request. You don't have to be an RF engineer to do this.
A simpler example is when a cloud engineer needs to define which 10 things to monitor every day across the network’s operations to know that their cloud is healthy. The AI should be able to decide what those things are and provide relevant and accurate updates on their status. A RAN engineer should be able to say that they can scan their network and see which KPIs are underperforming - and if that’s the case, which action they should take; the tool needs to be intent-aware.
Let’s move on to someone at C-level - the most important day-to-day decision makers in an MNO. They might want to know how much they should invest in 5G infrastructure so that they can achieve a certain level of profit at a certain point in the future. That begs a lot of questions: in which city should they invest in building sites? What benefits and consequences would that decision bring? The AI should be able to tell the CXO that you should only launch, let's say, 1,000 sites and focus on one particular city because 4G sites are 80% loaded there, 80% of users already have a 5G handset, and your competitors already have a distinct advantage with 5G coverage.
The AI might tell the CXO that if they only want to spend a certain amount on infrastructure, and with a specific timeframe in mind, this area probably isn’t the best place to launch 5G. That’s just one aspect. Next, we can ask how the AI thinks the CXO or sales team should price the network’s plans. A tool might tell you that you should launch a monthly 4G plan at $10, for example, with 5G at $12, but for the first six months you should offer it for free so that people can get used to the service.
These are just some examples of how we design the AI platform its capabilities, so that we can not only build technology but solve use cases and measure ROI of the AI.
Rakuten Symphony is committed to end results, delivering the use cases in production and solving the real customer problems. We specialize in turning intent into actions and then into assurance - we always believe that every use case gap is a product or a feature in the making.
