Artificial Intelligence in Enterprise Networking

Written by: Allyn Crowe, Senior Security Engineer
Connect with Allyn on LinkedIn

The world seems to be on fire for artificial intelligence (AI) these days. The label of “AI” is being slapped on just about everything. And the networking industry is no different. However, many questions exist about everyone putting an AI label on their product. How can you figure out if it’s actually AI or just machine learning (ML)? Honestly, what’s the difference between the two? How does AI help us in enterprise networking? Ultimately, how do I ensure a product labeled AI is helpful?


AI and ML Defined

AI is defined (by the Oxford Dictionary) as “the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.”

Machine Learning is “the use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyze and draw inferences from patterns in data.”

As we look at the various definitions of AI and ML, we see that AI is a broad term. In its definition, ML is a subset of the more extensive AI definition. As we dive in, we see other tighter subsets like deep learning or branches like fuzzy logic and natural language processing (NLP). We also see cross-cutting fields (some may argue that it’s a larger set) like data science. So, in its basest sense, ML is AI, but AI doesn’t have to be ML.


The Differences

As we’re seeing in current industry marketing, there seems to be less difference between the two. Many things that once were described as ML have been rebranded as “AI.” The term AI tends to conjure up images of computer voices, answering complex queries with minimal time or effort. ML tends to evoke thoughts of computers chugging away at data sets for days to try to give you some relation between them. And with the release of some awe-inspiring generative AI systems that sound very convincing and less “robotic,” we see AI everywhere.

The critical difference, to me at least, is that ML is a tool that is great at doing one thing (like correlating data sets), whereas what I want from a higher-level AI is to take those correlations and provide guidance. We can see some differences if we look at some of the typical “AI”-described systems these days. Generative AI systems, for example, can generate some very human-sounding blocks of text. But is there any “truth” in that text? A generative AI doesn’t understand the meaning of what it’s generating, just that the generation fits its current model. This drives us into some of the fun of model training, self-learning, etc.


What This Means for You

While we could spend many articles discussing the differences between the various pieces and parts, we’re here to focus on what AI can do for Enterprise Networking. Setting aside the mental pictures of your wireless controller magically making everyone’s Wi-Fi perfect, there are some actual use cases where AI can assist. Use cases such as expounding on correlative ML models to provide potential solutions to a long-running lower-level issue masked by different symptoms. For example, a maximum transmission unit (MTU) issue can hide behind high retransmit rates or dropped packets. Some of the AI systems in networking can detect some of the various symptoms, run the tests, report that the issue (that you didn’t know about) is an MTU problem, and tell you where to fix it in your configuration. In the future, we may even trust AI enough to fix it and then let us know.

Another use case we see is using generative AI and large language models (LLM) to respond to our queries more intuitively. Rather than learning a special query language to go into a query interface and iteratively tighten our data set until we discover the critical piece of information, we can now have a conversation with a chatbot that uses its data set to try to provide us an answer to the fundamental question we’re asking. By moving from a specialized language and privileged interface into a more general and user-friendly interface, we can help enable our users and front-line technicians to fix their problems. And if they can’t solve the problem, we have a more detailed and data-rich source of information for the escalated ticket.


How Do You Know It’s Correct?

Trusting an AI-based system is critical when placing our network in its digital hands. We have seen how false positives and negatives affect our responses to various situations. So, how do we make sure we can trust an AI system? Explainable AI (XAI) aims to help guide this thought process by creating a framework to provide human operators with mechanisms to understand and trust the AI outputs. The draft Internal Report by the National Institute of Standards and Technology (NIST) 8312 (NISTIR 8312) describes this framework.

By building the system to garner this trust, we can now see when a system uses AI and where it’s more of a branding play. By explaining the system in language, we understand the curtain is drawn and can tell what’s behind it. If a vendor can’t explain how and why their AI provides the results it does, it’s a red flag.


How Do You Know It’s Helpful?

Unfortunately, because of the rather flagrant use of AI as a branding term, we must do some due diligence to understand what is behind the AI branding of a product.

Luckily, just because the use of the AI term is new (or revitalized, at least), it doesn’t mean we have to develop entirely new tools to understand the feature set represented by the term or even dive deep into AI theory. We can, of course, reuse industry sources, references, and reviews. We can ask questions that leverage XAI principles. And we can push for clarity in the vendor communications with where they apply particular fields of AI (such as ML, DL, NLP, etc.), and we can dig into how they collect their data to feed into their systems. Because we know that bad data in means bad results out.

Our partner, Juniper Networks, has provided an excellent, easy-to-follow guide called “Bringing AI to Enterprise Networking.” While some of the information is Juniper-specific (especially around their Mist platform), a good amount of the content can guide you no matter the current vendor you’re using, especially as you look toward future network upgrades or changes.

Download the Free Guide Here


How Do I Start?

A great way to get started is to engage with a trusted partner to help determine your requirements, evaluate industry products, and help you fit the right technology to your business and technology needs. The key behind this partner is having expertise in the various areas of technology that leverage AI. Nexum is here to help. With our pool of expert engineers, we can help provide input and guidance as you move forward in your choice of AI-based tools.


Check Out More Resources

Event News

AI-Native Now

Join Juniper Networks on June 5th for a LinkedIn Live exclusive discussion on “Leveraging AIOps for Maximum Impact.” 

Read More »