China Mobile's Bold AI Plans
China Mobile is making grand AI plans. As the service provider approaches the 1 billion mobile subscriber threshold, it's looking to embed artificial intelligence in every level of its business, including network operations, marketing, customer service and management.
China Mobile boasts more paying customers than any communications service provider in the world -- it has 925 million mobile subscribers and 156 million fixed-line broadband subscribers. Facebook and Google have bigger user bases, but the vast majority of those users are non-paying.
With that scale, running the network efficiently is a top priority, and AI is essential to that goal. "We rely on AI in networking to get more out of the technology we have and deliver more value to our customers," Junlan Feng, China Mobile chief scientist for artificial intelligence and intelligent operation research, tells Light Reading.
Network intelligence is China Mobile's "top concern," but AI will also play a role throughout the business. China Mobile is implementing AI in management, including office automation, document management, enterprise resource planning, human resources, and finance and accounting. In human resources and accounting, for example, AI can respond to frequently asked employee questions.
Feng laid out several challenges and questions operators need to resolve to get the full benefit of AI during a keynote at Open Networking Summit in April and during a follow-up interview with Light Reading.
Feng shared her definition of artificial intelligence, a term now commonly -- and mistakenly -- used as a catch-all for any policy-based operation. An essential element of AI, Feng says, is that the system learns and improves. "An AI-based application has to be a living thing; it needs to grow on its own," she says. Static software will become less useful over time as conditions change, but AI improves and adapts itself. "If it doesn't improve on its own, it decays. Nothing is stable."
To achieve maximum AI benefit, operators need to implement AI across the network in a "systematic" way: Rather than implementing individual projects, operators need to think about the entire network, and how AI in individual parts of the network can work together. Wireless, video, transport and the core network all need to work together to deliver value to the customer, Feng said.
Network operators need to consider how network intelligence is different from the AI algorithms we have today, Feng said. Today's AI is designed for speech, video, natural language processing and data analysis. The industry has plenty of data sets to help artificial intelligence identify cars, crosswalks and storefronts -- if you've ever tried to log into a website using Google CAPTCHA you've helped with that -- but it lacks data sets useful for studying networking needs.
To get the most from AI on the network, the industry needs to get comfortable sharing data on network behavior, while "massaging" the data to preserve customer privacy, Feng says. "The problem is pretty serious," she says. Networking data used in research is ten years old, and doesn't reflect the state of the industry today. To close the gap, China Mobile is hosting a nationwide network intelligence competition, using real data from its network made available to researchers.
The volumes of data to be considered are enormous: AT&T has 235 petabytes of network data, but China Mobile has more, due to its bigger customer base.
- Big Data Is Slowly Paying Off for Axiata
- Intel Reinventing Xeon for AI – but Is It Too Late?
- AT&T VP: Using AI to Hyper-Automate Processes
- Verizon: Vendor AI Not Ready for Prime Time
- Magyar Telekom's AI Trial Success Baffles Engineers
— Mitch Wagner Executive Editor, Light Reading