Image: metamorworks, Getty Images/iStockphoto

On the surface, it would seem that artificial intelligence (AI) is widespread in the telecom industry. For years we’ve been familiar with voice-activated menu systems that respond to your verbal commands.

However, the potential for AI in the telecom arena goes much deeper than voice controls, albeit with some unique challenges. I chatted about the topic with Tom Footit, VP of Product Management at Accedian, a performance assurance solutions provider, Kailem Anderson, vice president of Portfolio and Engineering at Blue Planet, an intelligent automation provider and Eric Braun, chief commercial officer at MobiledgeX, an edge computing company.

SEE: 5G: What it means for IoT (free PDF) (TechRepublic)

Scott Matteson: What are the opportunities for AI in the telecom space?

Tom Footit: Telecom networks generate an enormous amount of data, and as a result there are a lot of opportunities for AI in this space. The opportunities break down into three main broad categories:

  • Using data to understand and predict performance and security of networks and applications.
  • Using data to understand and predict customer behavior.
  • Using data to assist customers when they encounter issues (customer support).

To get the most out of AI, telecom companies will have to leverage technology that is capable of providing the most granular, high-quality (and clean) data related to performance, end-user experience, and an organization’s security.

Eric Braun: The opportunities for AI in telecom are many. There are three opportunity spaces where AI in telecom can be transformational:

Telecom’s own operational and business systems maximize personalized customer experience opportunities while increasing efficiencies of delivery, performance and pre-emptive maintenance. Telecom operational and business systems generate large volumes of clean, valuable data–the ideal input to machine learning analysis and the use of the inference decision making that AI systems generate. Telecom today is already the largest operational Internet of Things (IoT) solution on the planet.

Due to telecom’s unique distribution, the infrastructure can be available to host third party purchase provider’s AI engines that can act on large volumes of third party publishing data in real-time that otherwise could not be used for third party purchase provider insights. This is the main opportunity that 5G and edge computing can capture from the third party purchase provider perspective, as all businesses become service providers in their own rights toward their own customers.

SEE: GlobalData predicts 3G/4G networks will dominate huge Indonesian telecoms market by 2023 (TechRepublic)

In the future, there is also the opportunity to combine both telecom AI insights and third party purchase provider insights to further improve the experience and operations because of their presence inside the telecom edge network. One example is predictable radio performance where applications can pre-empt changes in connectivity capability and adjust their performance in advance of changes.

Kailem Anderson: AI is quintessential to operating and controlling not only today’s networks but future ones. As service providers continue to grow their networks to support the ever-growing demand for new products and services, they build them on top of existing systems–mixing old with new, vendor A with vendor B, and becoming increasingly complicated overall.

Managing this network complexity is overwhelming network technicians and putting service providers on the back foot reacting to their network rather than being proactive. Bringing AI to networks presents an incredible opportunity to get in front of service issues, reduce manual tasks, and improve operational efficiency, which ultimately allows resources to be re-directed toward creating a better, more tailored customer experience. This is the end goal of a network that can adapt–AI-driven automation.

Scott Matteson: What can AI provide that current implementations cannot?

Tom Footit: Telecom networks traditionally were very static and very manual to design, deploy and maintain. More recent demands on operators’ networks have required them to become much more dynamic [leading to the rise of software-defined networks (SDN), network virtualization (NFV) and adjacent technologies] but the process of designing, deploying, and maintaining them have struggled to keep up with this rise in complexity, and largely remain manual processes for most telecom operators.

In a nutshell, AI offers the promise of being able to better automate some of the design/deploy/maintain lifecycle, allowing the increase in network complexity to occur without a corresponding increase in operational expenditures to run the network.

Eric Braun: Historically, telecom systems have been carefully designed and implemented based on long experience with the nature of telephony calls. With 4G/LTE and 5G, the network loads are rapidly converting to data connections, which are entirely different in nature from voice phone calls requiring telecom service providers to quickly and dramatically change many of their design and operation principles as a result. Machine learning and data already available provide the means to do this, as well as the basis for network operation automation.

SEE: How AI is impacting the workplace (TechRepublic)

Machine learning based on this data is also a way by which network operators can gain invaluable insight into how their networks are being used and where the value is being generated.

Kailem Anderson: To put it simply, AI has the potential to improve every step of the network operations journey. AI can learn faster than humans can, so by leveraging the massive amount of data that these networks produce, AI solutions are able to identify patterns and potential network issues before they even materialize, find a solution, and implement it prior to a service disruption. For example, Blue Planet Proactive Network Operations solution leverages AI to foresee many unplanned network outages, pinpointing the cause and prescribing the necessary actions to preemptively resolve the issue.

Scott Matteson: How easy are these opportunities to implement?

Tom Footit: On one hand, these AIOps opportunities are easier to implement because we’re talking about networks that have traditionally generated a lot of data about how they operate. On the other hand, collecting that data and doing something intelligent with it is not a process that most telecom operators have started; they are largely using the data to generate alarms and operate in a break/fix mentality. Implementing systems and processes can be a challenge.

Eric Braun: Changing design and business practices developed over a century of experience is difficult at best. These issues are compounded by the newness of ML/AI and the new knowledge required, and by the changes required to automate operations. The potential value is enormous, but capturing that value requires change at all levels of an operator from management and business structure down to the mathematical competence required.

If looking ahead at 5G and edge computing, and by using the 4 edge model explained here, then it is clear that telecom edge exists today but the only applications that take advantage of the presence of the distributed infrastructure are telecom functions themselves, such as radio controllers, packet cores and border network functions. The biggest challenge faced by telecom companies to fully capitalizing on enabling AI for itself and for third party purchase providers is enabling programmable access to its infrastructure and systems and making such access as easy and “cloud-native” as possible.

Kailem Anderson: Leveraging AI in the network is not as simple as a snap of a finger but it is a wholly worthwhile endeavor. The first step for any service provider is defining what automation, and therefore AI, looks like in their network–where it’s needed and where it can best help provide value. From there, they can begin phasing AI into operations. This is crucial as, although they do so quickly, AI solutions still need time to learn, which means adopting AI is not immediate.

SEE: 4 ways to fine-tune your AI and machine learning deployments (TechRepublic)

In fact, providers will still rely heavily on human expertise, especially in early stages: Yes, the network operations team will use AI to identify potential root cause issues, but those team members won’t have the AI close the loop and automate the fix. Only when the AI solution has developed a track record and the operations team has a high degree of confidence in its recommendations and ability to prescribe and act appropriately should an AI solution be used to fully automate processes.

Data is the underlying current that runs through this entire conversation, and it’s that data that will dictate the speed with which AI functionality is possible. These systems need data to learn. They rely on training and data tagging, which require investment. Service providers that embrace the importance of data early on will see pay-off in the long run as their AI capabilities come to fruition at an accelerated rate.

Scott Matteson: What’s causing these delays and impediments?

Tom Footit: The technology to leverage data for AIOps in telecom networks is there today and it is based largely on well-established open source technologies and well-established data science that has been developed in other fields and industries. The largest impediment that we see is change: The ability to get people and processes to adopt the changes they need to move from a largely manual way of working to one that is much more automated takes time. And organizations need time to learn to trust the data and trust that the AI can effectively automate multiple tasks, freeing humans up to do the more complex and higher value operations.

Eric Braun: Change of this magnitude must be driven top-down, based on a clear corporate strategic goal or mandated and with clear executive support to overcome the expectable resistance of “we don’t do it that way!” Adoption of cloud-native operations, processes, and talent is key, and discovery must move from academic architectures to intelligent learning in reality.

As our CMO once said, clever telecoms are placing multiple bets. The good thing about edge computing is that it is going to happen because every single player has recognized it needs to happen. That includes hyperscalers, telecom mobile operators and all the vendors. So the only question is how should it happen. And it’s not a one-size-fits-all market.

SEE: Significant advances in 5G, AI, and edge computing among the top tech predictions for 2020 (TechRepublic)

What we are saying to all operators is that the only wrong strategy is to do nothing. Please do not academically talk about this for two years because in the next two years the market will have formed in the real word and you will have no leverage or understanding. If I were a CSO I would be very scared to choose one strategy when no one is really sure how this market will turn out. Place a couple of bets and learn from them.

Scott Matteson: What should organizations do to alleviate the roadblocks?

Tom Footit: AIOps is all about incremental change: Starting small and leveraging data to make better decisions, while still leaving the final decision in the hands of humans, is a good way to start. The technology is available to do “closed loop” automation without involving humans in the decision-making process; for example, detecting and fixing an issue in a network automatically–but that doesn’t have to be the first step.

Scott Matteson: What should providers do to alleviate the roadblocks?

Tom Footit: Stay in lock-step with their customers. If AIOps is about incremental change, then providers need to stop overstating the benefits of it and work with their customers to start implementing it. For most operators, change will come by evolution, not revolution.

Scott Matteson: What opportunities will be available in the future?

Tom Footit: Because telecom networks are so data rich, there are many opportunities. The concept of using closed-loop automation for self-healing networks is the target everyone is aiming for, and it is something that is certainly attainable at scale in the future.