Robot Wars: Telecom's Looming AI Tussle
In the not-too-distant future, when the autonomous car has become a regular feature of the cityscape, governments are likely to ban driving by humans as a reckless or unnecessarily risky activity.
It may take longer to rid the telecom organization of people, and there is no reason to expect an outright ban on the network engineer or customer services assistant. But telecom jobs are disappearing at a breakneck rate. Expressions like "zero touch" and even "self-driving network" are now dangled before the industry to describe a utopia where networks powered by artificial intelligence (AI) can run themselves, safe from our grubby, interfering hands. (See Efficiency Drive by Major Telcos Has Claimed 74K Jobs Since 2015.)
The evolution of telcos into utilities performing an escalating number of routine chores makes them ripe for automation, even if it takes many years. Any efforts by operators to avoid becoming "dumb pipes" have been largely futile. More than ever before, they are conduits for the services other companies provide -- TV from Netflix, voice from Microsoft-owned Skype, messaging from Facebook's WhatsApp. "They are heading in the utility direction and nothing in the world will stop that," says Bengt Nordström, the CEO of the Northstream consulting business.
Indeed, far from countering this trend, investments in software and virtualization will probably make the telco a more efficient dumb pipe. Operators talk about using virtualization in 5G, for example, to provide different service characteristics over the same physical network. Essentially, this "slicing" is an alternative to costlier investment in several physical networks, each designed for a specific need. By making a service launch less expensive, time-consuming and risky, a virtual network might create the conditions in which true innovation can thrive. But it will not provide the spark that innovation needs.
Software and virtualization will also enable companies to compete more vigorously in communications markets where they do not own infrastructure. This development can already be seen in the enterprise sector. Added to regulatory and economic pressure, it could ultimately drive more countries toward the "neutral-host" approach -- a single physical network supporting a multitude of service providers. These neutral-host networks will be the apogee of the dumb pipe. (See Trump's 5G Plan Is Not as Bonkers as They Say.)
Fully automating such a network should be more feasible than taking manual processes out of a creative business. What's more, efficiency will be of paramount importance for that network and its tenants if the commercial arrangements are to work, giving investors more incentive to use automation in the most extreme way possible. "There is a huge opportunity for cost cutting and efficiency gains. In an ideal world, you should not need so many staff to manage operator networks," says Nordström. A lightly staffed dumb pipe powered by an economical but sophisticated AI will look more desirable than ever.
That belongs to the realm of science fiction, of course. AI is not poised to assume control over telecom networks any more than cyborgs are to replace the army. But a long-term future that grants humans a significant role in the operation of a utility is almost inconceivable. And just as the military gets ready for the robot wars to come, so operators are taking baby steps in the brave new world of AI.
Next page: Three-way fight
In this environment, a different type of robot war looms as companies jostle to provide the AI systems that will control the networks of the future. A three-way fight is shaping up, with traditional network vendors on one side, technology players on another and operators caught in the middle -- taking products from third parties while developing their own capabilities at the same time.
Viewing AI as a potentially lucrative new sales opportunity, network equipment vendors such as Sweden's Ericsson AB (Nasdaq: ERIC) and China's Huawei Technologies Co. Ltd. are starting to market their nascent AI products as an essential ingredient of 5G, a forthcoming next-generation mobile technology. Huawei reckons 5G wireless networks will have 50 times as many configurable parameters as their 4G predecessors -- too many for humans to handle efficiently. It is developing a "wireless intelligence" system that uses big data analytics and self-learning capabilities to automate some of these processes. (See Huawei's $800M 5G Budget Piles Pressure on Ericsson, Nokia.)
Ericsson boasts something similar with its "machine intelligence" line-up. Its "load-balancing" tool, which can make rapid adjustments to network settings in response to usage levels, is already in trials with Vodafone España S.A. The operator plans to use the AI-based technology in commercial networks this year. (See Humans Beware: Ericsson Readies Machines to Run the Network.)
But many service providers evidently fear being led blindly by vendors into an uncertain AI future. Deep-seated anxiety about the interoperability of products and services from different suppliers appears to have sharpened this fear. The prospect of "vendor lock-in" may be especially troubling when it comes to AI. If a third party's self-learning tool comes up with solutions that technicians and rival technologies cannot fathom, the operator may worry about losing control. (See DT Demands Automation, Cloud Tech From Pan-Net Suppliers.)
While this is partly an ethical dilemma, several of the world's biggest operators hope to overcome some of the interoperability concern through the Facebook-led Telecom Infra Project (TIP). In late 2017, Germany's Deutsche Telekom AG (NYSE: DT) and Spain's Telefónica formed a TIP working group focused on artificial intelligence and applied machine learning. Like Ericsson and Huawei, it aims to build the tools that will manage the networks of the future, helping operators to cope with traffic growth and service complexity. One of its three work streams is all about developing multivendor data exchange formats. The working group has drawn support from South Korea's SK Telecom (Nasdaq: SKM) on the telco side as well as Cisco Systems Inc. (Nasdaq: CSCO) from the vendor community. Facebook and Avanseus, an Indian software company, are also listed as project champions. (See Facebook's TIP to Launch AI Working Group.)
Outside TIP, nervousness about the growing might of vendors and technology giants is spurring some operators to hire data scientists and invest in their own AI tools. Norway's Telenor, which plans to cut about a fifth of its workforce in the next three years through automation, is one of the most prominent. "When it comes to the solutions and models, we don't believe in buying them," said Bjorn Taale Sandberg, the operator's head of research, in a recent conversation with Light Reading. Telenor Group (Nasdaq: TELN) is currently testing machine learning for resource optimization and planning. In future, says Sandberg, this might help it to cut software spending: Telenor could buy fewer licenses and move them between basestations as traffic flows around the network, he explains. (See Downsizing Telenor Pins Margin Hopes on Automation.)
In some cases, it is a lack of vendor offerings that has driven operators to seize the initiative. Using AI algorithms, Telefónica has built its own data analytics platform to improve the efficiency of its network operations. "Three years ago, when it looked into this, there weren't any commercial solutions," says James Crawshaw, a senior analyst with the Heavy Reading market research business. The in-house platform might today be used to automate some field force management, he says, and even to ensure capital expenditure is not wasteful.
SK Telecom has gone even further, becoming an AI vendor in its own right. Much like Telefónica, it has been using internally developed AI for network management and operations. Tango, as the system is called, allows it to gather and analyze data in real time, says Park Jin-hyo, the operator's chief technology officer. "With this we can have almost zero-touch operations," he says. "The operations guys hate me." Having seen the benefits at its own business, SKT began providing Tango to India's Bharti Airtel Ltd. (Mumbai: BHARTIARTL) last October. (See India's Airtel Calls on SKT for Tech Help.)
Rather like Huawei, SKT expects the importance of AI to grow with the rollout of 5G. "It is impossible to manually operate a 5G system because there will be so many alarms and messages," says Park Jin-hyo. Tango was accordingly designed with 5G networks in mind. When those finally appear, it should be able to accommodate 5G features such as network slicing, dynamically apportioning resources to specific services and customers as needs dictate.
Seeking a new growth opportunity, Finland's Elisa, a much smaller telco on the world stage, is similarly morphing into a supplier. Facing resource constraints and surging traffic levels on its mobile network, it began automating network operations and radio systems years ago, training network engineers to use popular coding languages such as Python. More recently, Elisa Corp. has been marketing its automation tools and expertise to other telcos. Romania's RCS & RDS was unveiled as its first customer in February. (See Finland's Elisa Is Selling Its Automation Smarts to Other Telcos.)
While AI does not currently figure in Elisa's offer, that could soon change. "We have to be sure that when we use new technologies we have tested them properly, but we are very excited about the possibilities," says Kirsi Valtari, a senior vice president at Elisa's telco efficiency business.
Next page: Control freaks
If frustration with vendors, and a desire for control, are responsible for some of these in-house efforts, few telcos have the resources to challenge the main equipment vendors for technological supremacy. China's Huawei plans to invest between $10 billion and $20 billion each year in R&D. Sweden's Ericsson last year pumped $4.1 billion into its own R&D activities. Overall capital expenditure at SKT in 2017 scraped $1.9 billion. At tiny Elisa, it was just $303 million. (See Huawei Commits Up to $20B for Annual R&D, Fleshes Out AI Pitch.)
A sense of telco kinship may have driven Airtel and RCS & RDS to use products from SKT and Elisa, instead of relying on their traditional suppliers. The Romanian operator has certainly hinted that Elisa better understood its needs than a typical vendor would have done, including its requirement for tools that work in a multivendor environment. Yet other companies might prefer to build their own technology, in the manner of Telefónica, SKT and Elisa, than add an operator to the list of suppliers.
An in-house project might not look so daunting if, as in the case of Telefónica, it is a relatively small-scale initiative. Built by a handful of employees, Telefónica's data analytics platform did not entail a "massive investment," says Heavy Reading's Crawshaw. For more specialized requirements, such as load balancing, he doubts many operators would try developing their own tools.
Ulrika Jägare, Ericsson's director of analytics and machine intelligence, also sees limited overlap between Ericsson's AI investments and those of its customers. "They are spending on churn analysis and marketing campaigns and that sort of thing," she previously told Light Reading. "When it comes to AI from a network perspective, they are asking us what to do."
But if many operators still feel comfortable in the passenger seat, an urge to grab the steering wheel seems bound to grow unless vendors tackle some of their core concerns. Interoperability is a perennial bugbear. Huawei, which plans to launch an AI-enabled computing platform called Atlas this year, has acknowledged that it will be "very challenging to use AI in a multivendor environment." Jägare concurs. "I think very little is being addressed at the moment in this space," she recently said. "We need to see more standardization happening."
With its new AI working group, TIP might provide some answers. The initiative, which is pushing open source technology into the networks business, today lists hundreds of members, including service providers, vendors, startups and other organizations. It is already shaking up parts of the network equipment industry. Yet it is missing some of the biggest names in AI. Google (Nasdaq: GOOG) -- unsurprisingly, as a Facebook rival -- is absent. So too is IBM Corp. (NYSE: IBM), whose Watson technology is seen as one of the frontrunners in the AI race. And on the networks side, neither Huawei nor Ericsson has joined.
That does not mean Ericsson and Huawei have no time for open approaches. Jägare thinks open source has given a spur to AI development. Ericsson, moreover, has been using open source code to develop a "chatbot" aimed at field technicians. Yet companies that have prospered by selling proprietary technologies and intellectual property are bound to be wary of a group trying to overhaul the existing business models. (See TIP Players Voice Open Source Misgivings.)
Next page: Technopocalpyse
Perhaps the biggest challenge to the network equipment vendors comes not from the operators and TIP but from the likes of Google and IBM, which often make headlines for their AI activities outside the world of telecom. Their rapid advance into the communications sector makes the Silicon Valley giants a worry for operators, too. Arguing that in the next 20 years AI will underpin every industry and product, Telenor's Sandberg has voiced concern that Amazon.com Inc. (Nasdaq: AMZN), Facebook and Google may become all-powerful unless others "take the necessary steps."
Despite that concern, technology companies have started to infiltrate the telco business. Right now, their presence is most apparent in the digital assistants that operators provide to customers. Telefónica, for example, has partnered with Microsoft to build a voice-activated assistant called Aura, which allows customers to check on monthly spending or switch TV channels at home. Vodafone UK used IBM Watson to develop a "chatbot" for customer services. Watson also powers the digital assistant for Orange Bank, the French operator's new mobile banking service. Interactions currently happen through text messaging, but there are plans to incorporate voice functionality in future, says Jean-Philippe Desbiolles, vice president for cognitive solutions at IBM's global business services unit in France. (See Telefónica Takes Aura AI Tool Into 6 Markets and Orange Plans Bank Raid With AI, Digital Weapons.)
Mobile TeleSystems OJSC (MTS) (NYSE: MBT) is treading more cautiously, however. Plotting a widespread chatbot rollout later this year, the Russian operator is now piloting two chatbots for customer service interactions, one developed internally and one with a partner (whose identity it has not disclosed). It also recently hired a Russian AI expert called Arkady Sandler to take charge of AI development. "His role is to help us implement AI into chatbots and reduce calls and make service operations more digital," said Kirill Dmitriev, the operator's head of sales and customer services, during a conversation with Light Reading at this year's Mobile World Congress. (See Russia's MTS to Cut 1,000 Jobs as AI, Chatbots Arrive.)
As yet, there is little sign that Silicon Valley's AI systems have burrowed their way extensively into network operations. That is not surprising, says Crawshaw. "Many activities in AI are around image recognition and natural language processing, which are very different problems from network management," he explains. "There might be scope for AI to help, but it is not just a case of sucking up random data and hoping some magical AI super-being will pluck out the answer."
Indeed, using a "general AI" like Watson to solve network management problems might be "overkill," in Crawshaw's opinion. "There are not just the license fees you pay IBM, but the costs of data processing and clean-up," he says. For smaller operators, in particular, the AI bill could be a potential showstopper. One former telco executive reckons the processing costs of AI-based network slicing could make it uneconomical for all but the largest players.
Specialists developing so-called "narrow AI" technologies may have more success in the short term. Targeting very specific problems, such companies might analyze a data set and then write a program to optimize a task, such as identifying broadband outages in a particular area. "Most of what we see today is narrow AI," says Crawshaw.
Nevertheless, while Watson may have a much higher profile in the world of chatbots and digital assistants, it is also being used for network management and operations, says IBM. At the time of publication, the company had not responded to a request for details of its network customers. But in marketing literature it says an operator could use Watson to automate processes at network operations centers. The benefits, it claims, include lower operational costs and better network and service availability.
Can a network vendor like Ericsson, with no real track record in AI, seriously challenge one of AI's original pioneers? Jägare is sanguine. Much like Crawshaw, she thinks the effort of taming Watson for a network environment could be overwhelming. "A lot can be managed further upstream but there are few [companies] that can work across the network," she says. "If you look at what IBM is offering, there is extra work you have to do. You have to teach the system."
For Google, the focus is likely to be very different. Although the search engine giant has put the brake on some of its telco activities, it continues to run a mobile virtual network operator (MVNO) in the US. Unlike most MVNOs, Project Fi, as it is known, has deals with more than one network, and can automatically switch between them to satisfy customer needs. According to a well-placed industry source, Google has also been developing its own virtual core network. Google is already using DeepMind, the AI it bought in 2014, to improve other services and technologies. It does not take a great leap of imagination to envisage a role for DeepMind within Project Fi and this virtual core.
AI-powered virtual networks and fully automated pipes are unlikely to be with us for many years. As analysts and commentators are fond of reminding the industry, excessive hype always surrounds the discussion about forthcoming technologies like 5G and AI. But while 5G seems destined to underwhelm, AI is already at a stage few would have thought possible just a few years ago. DeepMind can beat the world's best player of one its most complex board games. Annie, another AI, composes music that audiences cannot distinguish from the classical greats. It is no longer a question of whether AI will take over, but of how fast it will happen, and who pulls the strings. (See Automation's Advocates in Downsizing Denial.)
— Iain Morris, News Editor, Light Reading