As we continue to meet with service providers around the world to discuss their transformation initiatives, the topic of “troubleshooting AI” is frequently raised. Generally, their operations team is excited, and little anxious, about a new vendor sales pitch where “everything will be automated” using AI. They are concerned about adopting a new technology that can be difficult to understand by itself, and are even more concerned about how to troubleshoot issues that are presented by the AI. What do they do if the issues the AI raises don’t make sense? How do they troubleshoot a “black box”?
Most of the time, the operations team lives a relatively normal life. That is until they are faced with a severity 1 problem that needs immediate remediation. That’s when the wheels can begin to feel like they are falling off. In a traditional automated environment, the approach of guiding the personnel through the troubleshooting process is challenging. And, as much as I love AI and deep learning and all the cool things we can build with them, we must be extremely careful here before presenting AI as The Solution. AI may be part, but not all, of a much broader mix.
tools, or the wrong tools, won´t make it any easier. AI can be effectively used in some instances, but it won’t solve all their problems, and it may even create some new, unexpected ones. AI can be effectively used for now in data analytics, as a supporting tool, or for RAN optimization with 300-400 parameters. But I believe we should be cautious, and take baby steps as we begin the digital transformation journey, and education is certainly the best first step.