Carriers Will See Continuing Pressure to Adopt AI

“Any operator that wants to … manage a network in a cost-effective way and wants to remain relevant will have to adopt AI,” said Paul Patras, CEO of Net AI and associate professor at the University of Edinburgh. Hyperscalers like Amazon Web Services are starting to move into the telecom space, he said, which “means that the incumbents might be threatened by more agile companies that are quicker to innovate,” Patras said: “It’s all about adoption and being open-minded and willing to try new paradigms to remain relevant.”

AI can help carriers with the transition to a digital world, said Azita Arvani, former CEO at Rakuten Symphony. “AI is not just an enabler, but a force multiplier,” she said. “You’re getting this very intelligent 'workforce' for a … relatively low cost.”

But deploying AI won't happen without decisions, Arvani said. For example, what data should AI use? How will a company ensure data is clean? And what AI model will work best for your company? “One thing we don’t talk about enough is the cost versus performance,” she said. “We all want great performance, but it all comes at a cost.” Generative AI can be very expensive, but carriers can find ways to reduce those costs, she said.

AI can make carriers' networks more secure, but at the same time, AI increases the potential attack surface. Accordingly, “you need to make sure that you’re dealing with that as well,” Arvani said.

Given the unique characteristics of each network, building an AI model “from the ground up” is necessary, Patras said. The human element is critical when training AI models and making sure they are updated, he said. In addition, carriers must decide on the use case they’re trying to address with AI “because there are so many facets to a network,” he said. AI won’t eliminate jobs but can make work more efficient, freeing technical staff to focus on complex problems, he said.

AI is thought of as something that can be purchased “off the shelf,” said Monica Paolini, principal at Senza Fili. “You need to make sure that you have the right data to start -- you train the model, you customize the model the right way, and then you test it to make sure it fits your goals,” she said. “You can throw a lot of money at any problem” but at some point companies see declining returns, she said.

Carriers can build an AI model from scratch, as some are, or use an existing one and customize it to meet network needs, Paolini said. The cost, complexity and technical requirements, and what can be achieved are different depending on the model, she said. The amount carriers can achieve in cutting operating costs is difficult to predict, she added. You can use AI models for things other than automation and carriers can have automation without AI, she said.