If you've attended, watched, or read about a technology industry conference recently, you'll know that artificial intelligence (AI), both generative AI (gen AI) and more classic AI, is You may have noticed that it has been characterized as a sort of panacea for the above problems. industrial. Communication is no exception. The idea of communications AI solutions for everything from network management and optimization to customer management and churn reduction is widespread, partnerships are forming, vendor-driven thought leadership is accelerating, and communications AI clearly become part of the landscape. from now on.
For Rakuten Symphony, the hardware and software provider for Japan's Rakuten Mobile and its sister networks, communications AI is just that: AI. This is because these two parts of Rakuten are part of a larger whole that offers a wide range of services including banking, e-commerce, financial services, and more. The company develops a variety of AI engines and applications that are trained using data from across the enterprise, not just Rakuten Mobile, such as Rakuten Mobile.
The first two big questions about Telco AI – Does this problem require AI? If so, do I have the right data?
Rahul Atri, Managing Director and President of OSS at Rakuten Symphony Orchestra, explained the group's approach to AI. “We always believe in building platforms,” he says, as well as driving adoption and fostering a culture of innovation. He presented a three-legged problem regarding AI. First, do you have the data? Second, what is the cost? Third, has this created new efficiencies that cannot be achieved otherwise?
Rakuten Group maintains an integrated data lake. “We knew data was going to be the new oil,” Atli said. When it comes to costs, he says, “People expect it to come, but not many people are even talking about the costs: the cost of training a model, the cost of cloud resources, the cost of investing in understanding the use case.” ” In terms of efficiency, this is most important. “Do we also need AI?Is it a data insight problem? Can a common workflow engine solve it?? ”
Looking more broadly at automation tools, including the use of news agency AI solutions, Atri laid out the process. The first is analyzing the data, the second is finding the root cause of a particular focus, and the third is making a decision or taking an action. The first step, and what he calls the third, is easy, he said, stressing that finding the root cause is “what we need to do.” [AI model] Training and fine-tuning…focus on step two. ”
Atri also discussed the use of large-scale language models (LLMs) in carrier AI applications. He enumerated the different stages of a typical user's journey, starting with using chatbots and refining natural language processing. The next phase will bring in his LLM, which allows him to access unique data and specialize in domains and businesses to combine data and context. He notes that an operator may have access to his AI solution by a customer care agent, an RF engineer, or a senior manager, all of whom will need different information from the tool based on the different contexts of their positions. I pointed out that it would be.
He also addressed the debate within the telecom industry and more broadly among enterprise AI users. Is building an LLM from scratch for a particular industry a better approach than tweaking an already available LLM? “You can build as many communications LLMs as you like,” he said. Ta. “Do you want to ___? I don't think so.” He compared building a cloud-native communications network to building a cloud to support communications networks. It is noteworthy that Rakuten Group, with its shared data lake, uses multiple LLMs.
Data-driven, software-first operations are key to telco AI success
Therefore, all indications are that AI is the right solution.but Do you have the right data for that solution? Jeff Hollingworth, chief marketing officer at Rakuten Symphony, emphasized the importance of “data fidelity” in automating previously manual network processes. He advised operators to “understand the jargon, the hype, the jargon. Start doing your own research by taking a business-driven, ROI-driven use case, user group, and work stream approach. Understand what these new technologies with automation, data, and AI can do for you, and be very disciplined and prescriptive.
Rakuten Mobile is likened to a self-driving car, with Level 1 having minimal (but some) driving support functions, and Level 3 having most tasks automated but with some conditions requiring manual input. It proposes automation, and level 6 means that the vehicle performs all driving tasks. Any situation that does not require humans.
Increasing the level of autonomy, Hollingworth envisioned a “buttonless keyboard for the network operations center.” But to reach this state, AI is not the starting point. “This journey actually starts without AI, because the only way a machine can see what’s happening to it and what it’s experiencing is if we have data that actually makes sense. And having access to that data is the most important level of transformation…and there are always some aspects of data that are important to understand.The first aspect is: What is the fidelity of that data? That’s it.”
Here's another analogy with Google Earth. “When you're outdoors looking at the Earth from the moon, you don't see a lot of detail, but when you zoom in, the moment you go down to another level of detail, the next thing happens. Based on the information added, you can actually You can make different levels of intelligent decisions. People say that the enemy of good decisions is the average. The second aspect of data is timeliness. So, use that data as close to the point in time as possible. Making it possible and making it available to the engine that is monitoring it (the AI model) and making it interpretable is another step towards reaching full automation, on the journey towards full realization. It is clear that autonomy is a factor.
Mr Hollingworth continued: “And what's interesting is that all of this applies if you just want to automate. So automation is about not just receiving data so that you can decide what to do, but actually analyzing it so that you can actually analyze it. You can also generate data at different levels of granularity. And this is one area where Rakuten has really taken a leadership role, as we've digitized all the processes involved in acquiring the network. That's a huge asset at the time. Without that data, you can't see things, and without horizontally unified data, you're down to level 1 automation, running binary automation, accelerating, But something similar to cruise control in a car is a good example of Level 1 automation.”
Hollingworth said another important thing to consider when adopting a carrier AI solution is around the organization: the talent. This enters an interesting area where telco AI requires data, but much of the relevant process data is stored in traditional ways or held in the heads of employees. So how do you transform your organization's memories into data that can be fed into your telco's AI systems?
Hollingworth called the piece “the biggest problem.” And it's always an interesting question to ask. If someone reaches out to you and asks, “I want AI to solve my problem,” it's always interesting to take a step back and ask the first question, “So, how do you intend today?” What data do you use to make decisions and make those decisions? And how do you access that data?'' And in many situations, especially in highly organized organizations, 10 9 out of 10 times, I haven't used too much data. They don't have a lot of data, so if you don't have the data, you can't expect machines to make better decisions than humans. So 90% of the work in AI is data. ”
And the third big question is, what does this mean for the environment?
There is no doubt that AI will impact telecommunications and essentially every other industry that aspires to improve productivity and streamline operations while differentiating products and services. But at what cost? Training AI models requires large amounts of computational power, which today is primarily provided by power-hungry graphics processing units (GPUs). More AI means more GPUs, more power. And regulators are taking notice. In the United States, lawmakers have introduced legislation outlining the standardization of AI environmental impact assessment and long-term impact reporting. European Union AI legislation requires reporting on AI-related resource consumption over the lifecycle of certain AI systems. Discussions about sustainable AI are also increasing. And while AI can help all kinds of industries optimize and reduce resource consumption, it's a bit circular in that it requires more resource consumption from AI.
Now let's go back to Atri's story. “We're concerned about cost and environmental impact. We've all seen it, we've all noticed it, we've all talked about it. But we're going to step on this and say, 'We're not going to make more than five.' I don't think anyone would say that. [large language models] within a year. Everyone is rushing to create more.Everyone wants to be more efficient…so [I’m] I'm afraid of those effects. ”
He effectively argued that while AI can help drive meaningful and far-reaching progress for individuals, businesses, and the world, the environmental impact must be considered. “I’m very excited that this is going to be something good, but I’m very concerned that every ChatGPT query is consuming water somewhere and consuming power somewhere. ..These are two scenarios, and if we can balance them, I think this technology will be great. We can create a lot of innovation.”
