“A fully-automated, zero-touch deployment…self-configuring, self-healing, self-optimizing, self-evolving…That’s the nirvana,” of telco AI according to Red Hat Senior Director of Technology and Architecture Azhar Sayeed. Speaking recently with RCR Wireless News, he laid out that long-term vision then looked at the incremental steps operators can take to move towards “hyper automation,” which will ultimately be a necessity given the rapidly increasing complexity of networks and services.
In pursuit of that end state though, there are a number of infrastructure challenges that need to be addressed, including the sheer scale of distributed infrastructure, disaggregation of hardware and software, a huge ramp in data being generated by network-connected devices and the network itself, a dynamic service landscape that requires near-constant configuration changes, then baking-in automation from day zero. “Telco infrastructure is becoming massive,” Sayeed said. “Net/net the effect is that data center hardware is being deployed much, much closer to the subscriber…We need to think about how do you actually manage an ever-changing infrastructure…Automation really becomes step zero for implementing anything AI.”
There are three kind of big pillars that underlie closed-loop infrastructure management:
- Hyper automation of everything–servers, networks, storage, capacity planning, business processes and infrastructure-as-code.
- Analytics and AI–data analysis, AI modeling and training, deep learning, decision-making accuracy and variance analysis.
- Autonomy and governance–policy/operational decisions, thresholds for actions, bias removal, privacy and security.
These disciplines support closed-loop infrastructure management where configuration changes are based on observations from instrumentation, analysis of that data, and established triggers, or reactions, all working in harmony to create an eventually circular cycle of autonomy. “You instrument what you observe,” Sayeed said. “You gather that particular data, then you can analyze…You create certain triggers…then you can react based on those triggers…Observability is key here because that’s what provides you [with] the quality of data.”
The long-term trajectory here is from tactical automation to process automation and on to hyper automation, he explained. As AI applied to this evolution, “It’s about really taking a complete look top-down and building that particular trusted data environment for your AI/machine learning to operate” in. “The environment is actually already changing from your on-prem to multi-cloud. You need to have hyper automation really to be able to actually automate both your on-prem and public cloud environments, to be able to gather that kind of data, create that trusted data environment, then apply AIOps to it.”
For operators, tactical automation gets rid of low-value, high-volume tasks and provides a fairly straightforward opportunity to learn from mistakes and asses costs and benefits. Process automation is used to unify siloed processes and facilitate faster innovation; this also requires the organizational overhaul that runs through telco AI discourse. Sayeed called it an “automation-first culture.” Then hyper automation, or advanced automation, draws on trusted data, applied AI and ML then takes action based on event-driven triggers.
The good news is that operators are already effectively doing what Sayeed characterized as “predictive AI” use cases, things like predictive maintenance, network optimization, customized marketing campaigns, fraud detection, network security, traffic analysis and network planning. So that sets a good stage for further advancement and, he added, “only now the computing power has come to a point where you can actually do this near-real time or real-time and actually be effective.”
From that current state, operators can move towards phase one of implementing AIOps which largely relates to the closed loop infrastructure management process outlined above. “Data science with artificial intelligence combined can get you to some of these capabilities,” Sayeed said. And while some AI use cases are well established, gen AI is new for telcos.
With a current focus on customer-facing gen AI tools largely around self-service chatbots and customer care, Sayeed sees as a path to using gen AI for engineering support wherein the intuitive interface helps them more easily parse network data. “Over a period of time, you can get into all the other things such as service management, text-to-code, all sorts of inter-domain communication and conversation and integration…That’s really where I see the role generative AI and how it’s coming up.”
For Sayeed’s full session, and more telco AI content, register for the virtual Telco AI Forum on demand.
Click here for more on how telco AI supports network automation and 5G monetization.
