AI and machine learning are already key components in the closed-loop control of BT's programmable network.

James Crawshaw, Principal Analyst, Service Provider Operations and IT, Omdia

May 10, 2019

5 Min Read
How BT Is Applying AI to Support Its Programmable Network

At the Augmented Intelligence event organized earlier this year by BT's Global Services division, the operator outlined to its enterprise customers how they could benefit from AI and what are the human, process, legal and regulatory challenges.

While AI ethics and AI governance are important topics, my attention was drawn to the presentation on AI in Networks by Dr. Simon Thompson, BT's Principal Investigator for AI. I spoke with Thompson after the event to find out how BT is applying AI in its networks today and what benefits it is yielding.

AI and machine learning are already key components in the closed-loop control of BT's programmable network, according to Thompson. They are used to detect anomalies and take preemptive action rather than relying on resilience (redundancy, failover) or restoration/repair. BT is also using AI/ML to ensure efficient, fair and pragmatic resource allocation between users and between the services themselves.

Figure 1: Dr. Simon Thompson, Principal Investigator for AI at BT. Dr. Simon Thompson, Principal Investigator for AI at BT.

Moore's Law and big data have paved the way for AI
Although BT has been using AI techniques for more than 20 years, the combination of cheap, fast processing power and large data sets has opened up many more applications of late. Around 2010, Thompson led the creation of BT's big data initiative. In the networking domain, big data analytics gave BT a much better understanding of its fixed broadband access network. "We went from a situation where we relied on physical telemetry and alarms to one where we could look at every line and know what conditions existed at a given time," says Thompson. "By using big data we were able to see and understand the whole system. That led to better decision making about the allocation of resources and actions to improve customer experience."

Dealing with data uncertainty
Applying AI techniques such as deep learning to the network data proved a challenge, however, due to the cumulative errors in inventory records and the complex topological relationships (between exchanges, street cabinets, telegraph poles, etc.). By using a Bayesian probabilistic approach to modelling the access network, BT was able to overcome the noise and uncertainty in the data to extract knowledge and value. Thompson explains: "We were able to correct mistakes in the network inventory, to determine more accurately the grade and type of cabling, and that helped us to better direct our investments."

But BT isn't just using AI to manage its copper infrastructure more efficiently. It is also looking to exploit AI in electronic transmission and switching infrastructure. Thompson sees the application of AI in network infrastructure as an opportunity for disruptive innovation. "As the network vendor ecosystem evolves we're excited about startups providing new network technology that we can bring to our customers," he says.

The importance of explainability
The use of AI has to be "explainable," however, as Thompson, erm, explains. "We need to onboard AI technology in networking in a managed, transparent and auditable way so that our customers can understand how and why our network is behaving as it is. To do this we need to be able to understand and control the AI components within all our network equipment. We need to be able to understand all the decisioning systems as they interact with each other across different vendors' equipment." To do this Thompson points out that BT's researchers are working with the TM Forum on the creation of standards around AI management as part of the AI & Data Analytics Project.

As well as using AI to optimize the planning and building of its network, BT is using AI techniques to improve service delivery and operations, specifically in areas such as alarm management and anomaly detection. "We've done a lot of work using AI to classify faults in the access network, translating telemetry readings into fault diagnosis. And that has then fed into the engineering workflows. For example, currently we are trialing anomaly detection as a way to reduce the workload on operational teams who are often overwhelmed by alarm volumes. By using AI we are able to filter out more of the unnecessary alarms than earlier thresholding and statistical techniques. This enables operations to focus on the alarms that matter."

Translating customer intent into network configuration
BT is also looking to use AI for service creation and management. Currently, creating a service such as an enterprise wide area network requires a lot of manual design and configuration of network elements. "We see more opportunity to apply AI to improve and extend how we automate a lot of this process by translating the customer's intent into a set of network instructions. Our SNAP platform has the first tranche of technology for automated network setup, but there's a big opportunity for AI to do more there," notes Thompson. BT's Service and Network Automation Platform (SNAP) allows it to integrate solutions from partners to offer its customers a choice of SD-WAN and NFV managed services.

While the AI term is often misused and abused by technology marketers, Thompson is convinced that real AI can add significant value to a communication service provider such as BT in multiple areas, including networking. "AI is a portfolio of technologies which have evolved over the last 60 years and are useful as tools to solve problems. Because of Moore's Law the availability of computer processing power has increased to a level where we can now apply machine reasoning to many more data-rich problems."

To find out more about how BT is applying AI in networking you can download the presentations and watch the videos from BT's Augmented Intelligence event here.

— James Crawshaw, Senior Analyst, Heavy Reading

Read more about:

EuropeOmdia

About the Author(s)

James Crawshaw

Principal Analyst, Service Provider Operations and IT, Omdia

James Crawshaw is a contributing analyst to Heavy Reading's Insider reports series. He has more than 15 years of experience as an analyst covering technology and telecom companies for investment banks and industry research firms. He previously worked as a fund manager and a management consultant in industry.

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like