Google Preps Machine-Learning-as-a-Service for Networks

Google has started applying its artificial intelligence (AI) expertise to network operations and expects to make its tools available to companies building virtual networks on its global cloud platform.

That could be a troubling sign for network technology vendors such as Ericsson AB (Nasdaq: ERIC), Huawei Technologies Co. Ltd. and Nokia Corp. (NYSE: NOK), which now see AI in the network environment as a potential differentiator and growth opportunity.

The Internet giant has been using AI to improve its search engine and enhance other consumer-facing products and services. In 2014, it acquired a UK-based AI startup called DeepMind, which subsequently made headlines when it was taught to play Go, a Chinese boardgame of fiendish complexity, and managed to beat Lee Sedol, one of the world's best players. (See AI Threat Is Tech's Fart in the Room.)

Drawing heavily on its capabilities in machine learning, a branch of AI, Google (Nasdaq: GOOG) is now developing more automated systems to cope with the growing scale of its global network, which comprises data centers and backbone infrastructure.

Google already uses software-defined network (SDN) technology as the bedrock of this infrastructure and last week revealed details of an in-development "Google Assistant for Networking" tool, designed to further minimize human intervention in network processes. (See Google Has Intent to Cut Humans Out of Network.)

That tool would feature various data models to handle tasks related to network topology, configuration, telemetry and policy. It is based on the concept of intent-based networking, whereby a technician defines what he or she wants to achieve but leaves the heavy lifting to automated processes.

Machine learning could take automation to yet another level as Google continues to extend its network in response to soaring adoption of Internet services and new customer needs.

"It is humanly impossible to manage this network and so we are relying on automation," said Sam Aldrin, a network architect at Google's network operations division, during a presentation at the AI Net conference in Paris last week. "The cloud has been the primary triggering factor to go in this direction."

Google's latest vision is of a third wave of cloud computing that will take humans out of the network equation. "We are now looking at cloud 3.0 where disaggregation is fundamental," said Aldrin. "To do that we need to have a lot of things in place."

If "cloud 1.0" was about the rollout of virtualization technology in data centers, then "cloud 2.0" involves greater use of the public Internet for enterprise services, according to Aldrin. Most organizations are still wrestling with this challenge, but Google has put "cloud 3.0" firmly at the top of its networks agenda.

Want to hear more about the leading operator use cases for AI technologies? Join us in Austin from May 14-16 at the fifth annual Big Communications Event. There's still time to register and communications service providers get in free!

However, using AI in a network environment is far from straightforward. For one thing, branches of AI such as natural language processing and machine vision would seem to have little relevance in the data network.

Moreover, when it comes to machine learning, much of the information flowing around Google's systems is unstructured data that is difficult to process.

To address that challenge, Google has built a "recommender" tool that examines new events against historical data and finds similarities. By training and fine-tuning its data models, it hopes to be able to make accurate predictions of network events in future.

Asked if Google might offer its AI tools as a managed service for telcos, Aldrin hinted at a forthcoming offer for companies operating on its Google Cloud Platform (GCP).

"Machine-learning-as-a-service is already available," said Aldrin. "Specific to networks, I don't think it is out there yet. But if you are going to build virtual networks on GCP you should be able to get this sooner or later."

Dean Bubley, the founder of analyst firm Disruptive Wireless and a moderator at the AI Net conference, said the update would be "interestingly scary for anyone from Nokia, Ericsson or Huawei."

All three equipment vendors regard AI as a sales opportunity and are developing tools and services for network management that include machine learning technologies: They will be troubled by any sign that Internet giants might eventually become rivals in this market. (See Robot Wars: Telecom's Looming AI Tussle.)

But for those who work in the telecom sector, the overriding concern is that AI and other automation technologies are poised to claim thousands of jobs in the next ten to 15 years.

Google is not the only company to have recently talked about managing networks with minimal human involvement. Finnish mobile service provider Elisa Corp. revealed last month that it now operates a "zero-person" network operations center thanks to automation and the use of software robots. (See The Zero-Person Network Operations Center Is Here (in Finland) and Finland's Elisa is selling its automation smarts to other telcos.)

China's Huawei reckons the number of employees dedicated to network management could fall by 90% in the next five years if service providers take advantage of its new intent-based networking and AI technologies. (See Huawei Can Help Cut 90% of Networks Operations Jobs, Says Senior Exec .)

Even if that is not exaggerating what is technically possible, such a dramatic cull seems highly unlikely due to cultural, legal and strategic factors -- some operators want to be able to shift network operations staff away from "firefighting" duties to focus on more challenging tasks. However, operators clearly see headcount reductions as a way of reducing operating costs and boosting profitability while sales remain flat.

The world's 20 largest telcos with headquarters in Europe and North America cut more than 63,000 jobs in 2016, or about 3.5% of total staff numbers. Average revenues per employee rose 4.5% that year, to about $431,000. (See Efficiency Drive by Major Telcos Has Claimed 74K Jobs Since 2015.)

— Iain Morris, News Editor, Light Reading

Phil_Britt 4/16/2018 | 9:06:04 AM
Machines Only not the Answer Humanly impossible? Yes. But to rely solely on the machines, without some human intervention to ensure machines not spewing out wrong information and are indeed doing what is expected (initial programming is still on point) is a mistake, too.
Sign In