Should I Leverage AI to run MY network? A Skeptics View.

As with any new technology, there are many who are skeptical about AI and in particular how AI will impact IT generally and networking specifically. To address the concerns of these skeptics head on, the Juniper Networks team coordinated a series of conversations. They called these chats the “AI Skeptics” series and I think they did a great job addressing many of the most common concerns, doubts, and just plain unknowns in the world of AIOps and how it will (and already is) impacting NetOps.

As in the past, Juniper came to me to review the material and provide my take as a recap to the video series. I initially intended to discuss the videos one at a time, providing a by the book recap of what went down in each conversation. But as I watched the videos again, a few themes surfaced across all three. So, if you want the blow-by-blow of each video – I recommend you simply watch them! They are short and direct with an all-star cast of moderators and participants. And if you want to know what I found most important about what was said, well, in that case, read on!

AI Makes Your Job Easier

The first conversation in Juniper Network’s AI Skeptics series focused on the big question that so many IT professionals are asking, “Is AI going to take my job?” And Darryl Alder, Lead Network Architect for Aston Martin was quick to squash that fear in his opening comments. He emphasized that AI is not about replacing humans with robots, but more so about taking the robot out of the human. I really enjoy this turn of phrase as it captures the reality so well. Darryl illustrated his point with the example of satellite navigation solutions, which did not replace human drivers. Instead it has augmented them, making us all better drivers – or, at least drivers who get lost far less often.

AI in IT is similar. Properly executed AI enhances, enables, and empowers all of us who design, deploy, and operate networks. Shamus McGillicuddy, VP of Research at Enterprise Management Associates, shared that his research showed that teams who are actually using AI show little concern for AI taking their jobs. Instead, they report that AI brings a lot of value. From turning information into insight to empowering teams and thus lowering the number of escalations, the folks who are already using AI find it far less scary than we might think.

This theme of AI making our jobs easier was reinforced in the second AI Skeptics conversation, which focused on the impact of AI on channel partners. Mark Thames, Business Development Director for Juniper Networks pointed out that while AI can improve the experience of both users and operators, it cannot deploy itself. Someone needs to design and deploy the network, and AI simply makes it easier for those doing so to exceed expectations and wow their clients. Jason Guynn, Senior Sales Engineer at Juniper Networks demonstrated this with a great example. He was part of a three-person team who recently deployed 4,000 Mist APs across 70 buildings at a large university. What was striking for him was how much operational overhead the AI removed. Because Mist AI took care of all the channel selection and power tuning, they didn’t have to go back to tune anything after the initial deployment.

The consensus across all three conversations is that AI can help you deliver faster while also doing more with the same amount of people – which, counter to the concern that AI may take your job, actually makes every individual more valuable. As Zeus Kerravala, Founder and Principal Analyst of ZK Research said in the third AI Skeptics video: “AI makes you better.”

AI is More than a Tool, it’s a Colleague

It turns out that how AI makes your job easier is just as interesting, or maybe even more interesting than the simple fact that it does make things easier. And that’s because AI and AI-enabled solutions are different than what came before in a very striking way. They are much more like co-workers than they are like traditional IT tooling. Tom Hollingsworth stated this very directly in the first chat, when he summed up by saying “[AI] is not a tool, you are hiring an AI.” A very similar conclusion was reached at the end of the third chat as well, when Zeus Kerravala advised to “be patient with AI” because it often needs time to learn and progress. He pointed out that this growth over time is a hallmark of “real AI” versus rules-based systems or other AI in name only. And Christian Scholz, Senior Consultant at Axians Networks & Solutions doubled down on this point in his closing remarks, stating that you must approach AI like you would a child or a newly hired colleague, teaching it how you want things to work.

This is exactly how I have been telling folks to think about AI for years now. And I was not the first. I heard Tim O’Reilly say something very similar in a talk six or seven years ago when he said that scripts are workers and developers are managers. Having gone through the transition from individual contributor to manager several times in my career so far that resonates. At the IC level, it can be easy to be threatened by a co-worker that you see as more valuable than you and be worried they may take your job. But at the manager level, you learn that the best thing possible is to have your team full of folks smarter than you. You want the people working on problems alongside you to be as good or better than you. Now, with the age of ubiquitous AI seemingly upon us, it is time to rephrase that – I want not just the best people working with me, I want the best robots too.

It’s kind of like the difference between a broom, which is a tool, and Micky Mouse’s magic broom from Fantasia, which is much more of a co-worker. Sharon Mandell, CIO at Juniper Networks acknowledged that in IT there has always been more work to do than people to do it. Which means that the helping hand of an AI co-worker should be a welcome addition to any team. After all, who among us has never needed a second (or third) set of hands?

AI Comes in Many Formats

Just as networks come in all shapes and sizes, AI can be implemented in many different ways. And as you might have guessed, how AI is implemented can make all the difference.

On one side of the spectrum is do it yourself (DIY) AI. To deploy a completely DIY AI solution, you need machine learning experts, data scientists, a ton of clean data, and often an army of people to build and train the ultimate solution. Of course this has the benefit of ensuring that the resulting AI solves exactly your problem in exactly the way you want. Still, for most this is a project doomed to failure for lack of resources and know how.

The other end of that spectrum is where we find pre-built AI solutions. The upside here is that we don’t have to be or become complete AI experts to deploy a targeted and pre-built solution. There are however a few common pitfalls to watch out for. The first is the potential for a mismatched solution. As Sharon Mendell pointed out in the third AI Skeptics conversation, you must first deeply understand the problem you need to solve, and then understand the problem that the vendor built their solution to solve. If they do not align it’s very hard to force a fit because of the training involved in creating a pre-built solution. Another common issue, pointed out by Zeus Kerravala is that so many current pre-built solutions are overly specific and do not work across the entire end-to-end network. As he stated, siloed solutions lead to siloed data which leads to siloed insights. Sharon reinforced this challenge by illuminating the potential for a “war of the bots” that can happen when AI solutions are each trying to optimize within their narrow scope and affecting each other in the process.

The ideal solution, as discussed by Andre Kindness, Principal Analyst at Forrester Research in the second AI Skeptics video chat, should simplify network operations by providing service level assurance, event correlation, root cause analysis, and anomaly detection. And it needs to do this across your entire Enterprise network, both wired and wireless.