MWC: AI to enable network autonomy and the ‘self-driving telco’

A significant theme at the recent Mobile World Congress was artificial intelligence. There were plenty of robots, driverless cars and digital assistants to attract visitors’ attention to the new user experiences AI will enable (and the new mobile network users, in the case of the robots).

But this is, underneath the glitz and the sci-fi, a trade show for mobile operators, so in the more serious debates, the focus was heavily on how MNOs can harness AI to make their networks more efficient and to deliver new services. This was their AI comfort zone – taking concepts of automation and intelligence which started with SON (self-organizing networks) and self-healing and are now developing with new 5G use cases in mind.

At the intersection of 5G and IoT – where the profits are likely to lie – is AI. It will make it feasible to manage networks of hundreds of thousands of cells and gateways, and millions of devices. It will make automation of those networks far more intelligent so that user experiences can be optimized and personalized. And of course, it will also help to make sense of all the data that is being generated by the devices and sensors, in order to support new services and revenue streams in key target markets for 5G/AI/robotics, such as manufacturing or smart transportation.

Then there is the consumer experience. As with mobile broadband services, some MNOs will leave that to the device or web vendors, content to be a bitpipe (hopefully a high value one). Others will want to control the user experience themselves, which in the AI-driven pre-5G world, means taking on Amazon Alexa and its rivals, as Telefonica is trying to do, unveiling its Aura digital assistant in Barcelona.

These, then, are the three critical areas of AI impact for an MNO:
advanced network management;
data analytics;
and end user experience.

Each operator needs to consider how these fit into its strategy for IoT and 5G, in order to develop new revenue streams, and protect current and future business from other service providers, including the AI giants themselves, such as Google, Amazon and Baidu.

Barcelona was certainly full of companies offering tools and services to apply machine learning and advanced analytics to the vast quantities of data which will soon be traversing mobile networks; and of course, engines which also make decisions based on those analytics, to drive applications like driverless cars. These will be the topic of a separate analysis in future, while this article will focus on the network itself.

Here, some operators and vendors have already fired the starting gun. AT&T’s Brian Daly, director of government and regulatory standards, told last fall’s 5G North America conference: “We are experimenting with AI to make our processes more efficient. There’s a lot of emphasis being put on it.” He believes AI is a “key enabler” of next generation wireless including 5G.

In particular, AI will make networks more cost-effective to deploy and run by supporting automation of even more processes than SON, including self-healing and pre-emptive maintenance. Some companies, such as Nokia, have shown off AI-driven ‘telco drones’ which can inspect equipment on cell towers without the need for humans to climb the structure, or install small cells in remote locations.

Daly said AI would work alongside software-defined networking (SDN), NFV, cloud, big data and mobile edge computing (MEC) as the chief enablers of a different way to plan and run networks and deliver mobile services. AI, he said, will enable next generation network design and will provide “efficiencies that may not exist today”, making networks “more reliable and secure”.

Even more importantly, AI will enable dynamic, flexible networks which will allocate resources intelligently and automatically as required by different users or applications. That, in turn, will greatly enhance the practicality and commercial value of network slicing, with AI providing a new level of automation and intelligence to make SDN/NFV platforms more responsive.

In India, Tata Communication is looking at AI to improve network design, maintenance and security, by capturing the knowledge and best practice of the rare breed of top-flight network designers and including it in automated, machine-based tools. That, says the firm’s future technologist David Eden, will allow MNOs to “pour the existing expertise in infrastructure design into AI algorithms, so machines will discover new design processes and techniques which would never have occurred to humans.”

He added: “AI can help us predict how people behave and identify the vulnerabilities that these behaviours are likely to create, improving our ability to predict and find the possible ways in.”

All this interest was formalized early this year by ETSI, which set up a new working group to define a ‘cognitive network management architecture’ – establishing standardized models for using AI to manage a network. The new ISG (industry specification group) is called Experiential Network Intelligence (ENI) and ETSI said it will use the ‘observe-orient-decide-act’ control model to allow operators to create an AI-based system which can adjust to context, modifying services and quality based on changes in user requirements, environmental conditions and commercial KPIs.

The automated system would learn from previous experience and so add intelligence to automated network planning and monitoring, reducing operation cost, improving capacity usage, and boosting user experience, to a greater extent than automation alone (with current generation SON, for instance).

The group will also study network telemetry, big data mechanisms, machine learning, and how to simplify and scale complex device configuration and monitoring.

“The unique added value of the ETSI ISG ENI approach is to define new metrics to quantify the operator’s experience; this enables the optimization and adjustment of the operator’s experience over time, taking advantage of machine learning and reasoning.” said Ray Forbes, convenor of ETSI ISG ENI.

Mobile operators will have to hope this is not too little too late. As telco networks become virtualized, even RAN functions will move to the cloud, at least in the 5G era. That will increase the need for AI to maximize the expected benefits of Cloud-RAN, such as extreme resource flexibility and density. But MNOs will have to remember that an AI-optimized, vast-scale network is nothing new to the webscale players, which introduced machine learning and automation to their cloud platforms years ago – and Google, Amazon and Facebook may well see the intelligent, virtualized mobile network as the next business they can host in their vast data centers.

For most operators, of course, 5G and Cloud-RAN are concerns for the next decade, but they are still interested in harnessing AI to improve the economics of current efforts in 4G upgrades, densification, virtualization and new services. So how were the vendors responding to that need, at MWC?

Amdocs, Ericsson and NEC were among the many suppliers discussing AI-driven network solutions before and during the Barcelona jamboree.

Amdocs unveiled aia, which it said was designed to enable the “self-driving telco”. It processes network data and user data in real time and triggers actions to adapt the network automatically to changing conditions. It is powered by AI and machine learning capabilities from various partners including IBM Watson and Cloudera, and can be integrated into about 50 telco business processes.

“It goes back to the days when we used to talk about how the service providers have lots and lots of data from multiple sources, from their own internal sources and external sources. But what do you actually do with all this data and how do you turn it into actionable information?” asked the company’s director of market insight and strategy, Uri Gurevitz. “With aia, you would get real time recommendations based on the specific application or specific business process that you’re looking at. It could be around customer management, it could be around planning and forecasting, it could be around the networking side.”

Meanwhile, Ericsson gave a look at ADA, an AI tool which can observe and predict network conditions and apply automated remedies where there are pending failures. Diomedes Kastanis, head of innovation, said: “Machine learning and other artificial intelligence techniques will shape the future of networks. Inside our Chief Innovation Office in Silicon Valley we are accelerating the evolution of the network through four distinctive stages, culminating in the truly intelligent network.”

The four stages, according to Kastanis, are:
Mainly reactive (today’s networks)
Leverage deep learning to predict future outcomes
Think beyond correlative programming and suggest outcome-based scenarios
Truly intelligent and self-sufficient – machines can differentiate between correlative and causal, and pursue their own choice of outcomes beyond the scope of human programming, and before any problems are noticed by subscribers

“This is the new best friend of the network operations centre,” Kastanis said in the interview, adding that the solution is already addressing the first two stages and is working on some US networks.

Ericsson plans to launch ADA commercially, later this year, as a white label solution as well as an element in its own managed services and network portfolio.

NEC has also been applying AI to improving network management and optimization. It believes that AI-enhanced automation will be necessary to enable large-scale densification, and this will apply to the backhaul as well as the access. Its Smart Wireless Transport Network (WTN) applies dynamic optimization techniques to guard against backhaul bottlenecks for customers using its iPasolink EX system, which supports small cell backhaul in the E-band spectrum.

NEC said: “In order to accommodate high speed communications and the IoT, it is necessary to accelerate adoption of small cells and expand the capacity of mobile backhaul that connects cell sites. The Smart WTN solution enables flexible network build-out, makes valid traffic and defect predictions, adapts to environmental and behavioral changes accordingly, and dynamically modifies operations in order to improve efficiency and minimize power consumption.”

NEC Laboratories Europe has also developed Net2Vec, which uses deep learning algorithms to analyse user data in real time and make decisions to improve customer experience – such as protecting users from threats, or adapting tariff plans and promotions based on real conditions. Net2Vec can use inputs from many sources, including equipment logs, measurement probes, user data and deep packet inspection.

Telefonica pitches Aura digital assistant against Amazon:

The digital assistant has been the most visible and much-used manifestation of the new breed of consumer services, driven by AI. Apple Siri, Microsoft Cortana and Google Now may support fairly simple functions, such as diary reminders and voice-activated queries, today, but the technology behind them is capable of revolutionizing the mobile and home user experience in future. And Amazon stole a march on its rivals in the home with its Echo device and Alexa AI engine.

From the voice-activated interface to the increasingly deep levels of personalization and context awareness, these apps and devices are blazing the trail to a new way of interacting with the web, which will rely on service providers knowing more about their users than those users do themselves.

Does the MNO have any role in that? It is a tenet of the 3GPP community that mobile operators can still differentiate their services from those of over-the-top providers by harnessing the unique knowledge they have about their subscribers’ tastes, preferences and behavior. Few, however, have proved that point with any major successes in areas dominated by web providers. MNOs’ attempts to outdo Google and Facebook with their own messaging, content and media offerings have not kept the web giants up at night, and in the case of the GSMA’s great hope for the survival of cellular messaging – Rich Communications Services (RCS) – it is only now growing on the back of support from Google itself.

So could combining the MNO’s knowledge of its subscriber, and ability to control the quality and resources of its network, really give it an edge this time, enhanced by AI?

Telefonica seems to think so, showing off its own digital assistant, Aura, in Barcelona. The product of a secretive two-year R&D project, with significant input from Microsoft, Aura is similar to Apple Siri or Amazon Alexa, but is tied specifically into the Telefonica network. Users can check details of their Telefonica services and request help or upgrades, using a voice interface on their smartphone.

The Spanish telco calls Aura a “cognitive intelligence system” rather than a digital assistant and claims three advantages – the control of a natural language interface; full  “transparency” about their Telefonica fees and services; and a personalized way to discover new service options.

CEO José María Álvarez-Pallete described Aura as a “fourth platform” that would sit on top of the other three legs of the digital telco – physical networks, IT systems and digital services.

Peggy Johnson, VP of business development for Microsoft, told the launch event in Barcelona that several recent technology breakthroughs had been incorporated in Aura, such as more accurate speech and image recognition (now comparable to human recognition, she said, with error rates of less than 6% and 3% respectively). She harked back to a favorite theme of Microsoft CEO Satya Nadella – “conversation as a platform” enabled by natural language interfaces.

Aura will be rolled out in the next 12 months in major markets including Argentina, Brazil, Germany, Spain and the UK though there were no details of pricing or the broader business model – and it remains highly questionable that users will adopt a telco-specific assistant in addition to one from Amazon or Google. They might use the Aura app for convenience in dealing with Telefonica, but it will be a big step to turn that interaction into using the technology for daily life.