Two European incumbents, France’s Orange and Telecom Italia, are the latest to provide details of how they aim to use artificial intelligence to make their 5G networks more efficient (and reduce staff).
Orange has been engaged in a research project with IBM and Nokia to use AI to predict future demand patterns in mobile networks so that it can plan them accurately, rather than always having to provision for peak usage.
The CogNet project is being conducted under the auspices of Europe’s 5G -PPP (5G Public Private Partnership) program, and it also includes Spain’s Telefónica, another frontrunner in the world of telco network AI, plus cloud services provider Interoute and Germany’s research foundation Fraunhofer Institute.
CogNet got started back in July 2015 and has just announced the first details of its work, at the AI Net conference in Paris earlier this month. Orange’s Imen Grida Ben Yahia, a senior R&D engineer in Orange Labs, said the project had used real network data from Manhattan, New York, to show how AI – in particular, machine learning – could help predict demand patterns and anomalies in dense urban areas; and how a large-scale event could “drastically affect traffic demand”.
In another CogNet study, the team used AI to address the issue of a “noisy neighbour” – when two or more virtual network functions (VNFs) on the same cloud infrastructure cause interference.
Yahia told the conference audience that the technology could also help with the thorny issue of mobile network SLAs (service level agreements), which are notoriously hard to create because mobile network performance is tough to guarantee. She said: “We could map this to low level metrics we are collecting to predict if a given SLO [service level objective] is going to be violated.”
Telecom Italia has been more upfront than many of its peers in acknowledging the human cost of using AI to support what Deutsche Telekom calls “brutal automation” in network management and other areas of the business, such as customer service. The Italian operator outlined how it would replace a large number of customer service staff with chatbots as a result of an AI partnership with Microsoft, though it did not give numbers. It has been engaged in a broader headcount reduction over the past year.
It said its Microsoft deal would be an important component of its DigiTIM program of digital transformation and would help simplify and enhance its relationship with its users.
Some operators have downplayed the threat to jobs from automated networks and customer interactions, but Telenor expects to cut 20% of jobs as a result of operations automation, while MTS in Russia plans to cut almost 10% of its 11,000 customer service positions this year, replacing them with chatbots.
Telefónica is also using Microsoft technology to underpin its Aura digital assistant, which will answer customer queries, while Vodafone and Orange are working on chatbots and assistants with IBM and its Watson AI platform. IBM is also pushing Watson as an enabler or automated network management and operations.
Last month, at the Zero Touch & Carrier Automation Congress, Telefónica said it was planning to increase its investment in automation tools, including a “cognitive self-optimizing network (SON)” technology based on AI.
This would see Telefónica making an early move towards what Rethink Technology Research, in its recent report about AI-driven SON, regards as a third generation of SON tools. The first focused on self-organizing more than self-optimizing and was largely concerned with housekeeping activities like automatic neighbour relations.
The second added a wider range of capabilities including many aspects of optimization, and extended SON systems beyond the RAN, into end-to-end networks and taking account of virtualization and cloud platforms. The third brings AI into play to enhance the performance, flexibility and value of SON still further.
Juan Carlos Garcia, Telefónica’s director of technology and architecture, said AI SON would be essential to cope with the complexity, and the huge numbers of moving parts, in a 5G network.
“The next step for 5G will be moving to a cognitive SON where AI and deep learning algorithms will be necessary to improve the efficiency of SON activities in our networks,” he told the conference. Most of the work will take place between 2021 and 2025, in tandem with 5G roll-out in many of the operator’s territories. That will include automation and programmability of core and transport networks, not just the RAN, while all these elements will be virtualized over time.