Artificial intelligence progression occurs in three steps, according to Mazin Gilbert, vice president of advanced technology at AT&T Labs: building automated systems, implementing automation and hyper-automation that includes advanced analytics and machine learning.
As crucial as that final step is, most companies have not yet been able to implement it due to major challenges.
Now there is the promise of progress for artificial intelligence (AI) via the combination of AT&T's own work in closed-loop automation and collaboration with other companies through the Open Network Automation Platform (ONAP) open source initiative. ONAP, which was created by the Linux Foundation earlier this year by combining AT&T's ECOMP with OPEN-O, will be the software-based platform that will increase the capabilities of AI and, eventually, automation. (See AT&T's Chris Rice Details ECOMP & OPEN-O Merger.)
Telco Transformation: What is AT&T's goal in utilizing AI?
Mazin Gilbert: At AT&T Inc. (NYSE: T) we think of AI as a smart way for us to do data-powered automation. Think about playing a chess game; every move is planned with the goal of winning the game. For us, the goal is automation, and machine learning is really about trying to drive it at every single step. That is already in effect in businesses where automated software systems have replaced manual processes thanks to data-poweredn AI-like speech recognition.
TT: That indicates that the first two stages of AI -- building automated systems and implementing automation -- are underway. What about hyper-automation?
MG: Very few companies in the world today are using hyper automation. In such a network, AI agents are continuously and dynamically learning on the fly with no human intervention. There are very few of those. It's a big research challenge that researchers have been focusing on for the past several years. This is a very tough problem: how to scale, how to find the right technology.
TT: Why is hyper-automation essential to AI deployment?
MG: Try to use any of the AI assistants today and you'll see that if you use it now, one minute from now, or ten minutes from now, it will give you the same response. That's because the system is not learning; it's not adapting on its own. If these systems are not learning, then after a while they become obsolete.
For example, we have AI systems in our network that detect patterns that could be security threats to the network. When you implement an AI system to defend against a cybersecurity attack, in many cases you don't know what they look like. They could change every day, every minute of the day. That's why you need it to have the capability of detecting things its never seen before.
In hyper-automation mode, the systems self-learns and self-evolves. Until you solve this, we will have a tough time in making AI mainstream across the board. If every deployment of AI requires an army of PhDs to feed it and improve it month-by-month and year-by-year, you could only have limited deployments.
Get real-world answers to virtualization challenges from industry leaders. Join us for the NFV & Carrier SDN event in Denver. Register now for this exclusive opportunity to learn from and network with industry experts --
communications service providers get in free!
TT: In a previous Telco Transformation Q&A in January, you visualized a "zero touch" form of automation that was as yet out of reach. (See Gilbert: Marrying AI, SDN Creates a Powerful Platform for AT&T.) Has there been progress on this front?
MG: There has been significant progress since January. Within AT&T, we've deployed aspects of ECOMP in closed-loop automation. A closed-loop means that it is fully automated with no human intervention required. Those are systems that collect data in the network, analyze the data, predict events that are about to happen, take action on those events and continue to monitor them. One way we use that is in detecting if something may go down in our networks and have the system automatically take remedial action.
Another driver for progress is ONAP, for which I serve as the chairman of the technical steering committee. (See ONAP Takes Center Stage at ONS.) Over the past several months, we've built up a huge ecosystem of operators, service providers and integrators. The best people in the industry who really understand the network and the cloud are part of this consortium. We knew that for us to do something really big in the network globally, we have to move everything to software. The next phase is to move this to AI, which you'll hear more about in the next few months.
Now we're at the crawling stages, and it's good to start out there to learn how things works and how to make them better. With the help of the 50-plus companies participating in open source, we will be advancing to the "walk" stage within the next three years. We'll get to "run" when the network become self-healing.
TT: How do you balance working with competitors in open source while also distinguishing AT&T from its competitors?
MG: It's for our own customers and our users to ensure that the best brains worldwide from the best companies working on it. We want to standardize the basic capabilities and the platform. Think of it as the skeleton of a car. Every car works off the same basic skeleton, yet different car manufacturers compete in a very healthy way. How these companies distinguish ourselves and compete is on the customers, data and how to use the technology on the network.
AT AT&T, we have several years head start on this. We compete on how we apply that and how quickly those closed-loop systems operate, what actions they take. That's our "secrete sauce" that we keep at AT&T.
TT: How would you characterize the way the AI journey has played out?
MG: First: we're going through generational evolution in AI that I consider "Gen One." We've used AI to grab a lot of data and accomplish some fundamental tasks like speech and image recognition. The work that went into developing that started three, four decades ago, and now we see its commercialization.
Now we're evolving into a new area, driving global change and network transformation via ONAP. To truly arrive at hyper-automation, though we need to solve some challenging problems.
In order to expand AI deployment, we have to lower the barrier to entry. It shouldn't be reserved for PhDs making up a small group but open to thousands. People in middle school and high school should be able to create these systems. Employers like us should be able to have every employee able to operate such systems.
We will progress from crawling to walking in the hyper-automation space, and so we have to plan for new AI applications that people may not have even dreamed of before. AT&T is a big company with the scale to apply new, compelling opportunities for AI.
TT: Can you address concerns about the potential danger of AI raised by people like Elon Musk?
MG: As we get 150 petabytes of data in every day, there is no way for a human to review it all. AI systems are very good at taking in a lot of data and inferring what the data is telling you. It is up to you as a human expert to tell the system what to do in response to "go shut down the IP address because when you see that, it's a security threat." It will not take actions that I have not trained it to do. That's not what we're building, not what we're deploying, not what these AI systems are. Their purpose is to help us humans do our jobs better.
— Ariella Brown, Contributing Writer, Telco Transformation