Can AI Help Network Staffs Manage Network Complexity?

As IoT, cloud-based, and on-premises networks converge, data on network performance is pouring in from all points. How can network staffs keep up? Is there a place for AI in network monitoring, and where should you draw the line?

AI net management
(Credit: Jinda Noipho / Alamy Stock Photo)

Between now and 2032, the IoT (Internet of Things) market is expected to grow by a CAGR of 24.3%, reaching $4,062.34 billion by 2032. The primary growth drivers are the movement of more business operations to remote locations, the ability of more powerful IoT devices to do more IT on their own, and the ability of IoT to find a place in almost every business use case that companies develop.

At the same time, new network protocols like Wi-Fi 6 are dramatically increasing the number of devices that networks can carry.

Both trends lay the groundwork for corporate network expansions, but the complexity of having to monitor all of these network nodes and devices also expands exponentially for network staffs. Even with current network monitoring and remediation tools, how will network professionals be able to catch every emerging performance or security issue?

The industry answer is by adopting artificial intelligence (AI) for network monitoring, maintenance, and remediation. AI has the potential to automate a large share of work in these areas that staff must do manually today—with the added advantage of being able to rapidly process and assess incoming real time data so the AI can act quickly. This is what makes AI a key component of AIOps (AI for IT operations).

Here’s what network AI tools can do

In a baseline startup, network AI needs a ruleset with which to operate. It's up to the network staff to define and input a complete set of rules that cover network performance parameters and monitoring, security threat detection, governance, etc. These are the rules that the AI engine will use for its daily monitoring. Additionally, the data coming into the AI data repository from network sources must be clean (e.g., no jitter or useless metadata) and of high quality. It is up to network staff to ensure that incoming data meets these standards.

An AI “model” is constructed from the rules of monitoring, etc., that the IT network staff provides. Once these rules are in place and the incoming network data is clean, AI network monitoring can begin.

From this point on, AI should be able to deliver the following capabilities that IT network staffs don’t have:

  • Built-in and automated machine learning that detects new patterns and anomalies in network data and assesses their impact on network rulesets

  • The ability to detect potential issues and trends from real-time data

  • The potential to automate routine network operations, such as using an AI RPA (robotic process automation) function to automate the provisioning of physical and virtual devices in the network.

Advantages like these give network staffs the ability to automate routine daily work that otherwise would eat into schedules. AI’s ability to predict network trends and issues also gives network staffs ways to anticipate and intercept these issues before they ever manifest themselves in a network service degradation or outage.

The challenges of AI adoption

The “catch” to a total adoption of network AI is that it’s still in early stages as a tool. AI has its shortcomings, and its relative newness makes it highly unlikely that any network staff will plan to relinquish ultimate human control over the network anytime soon.

Here are the challenges:

AI must know everything—and it might not

If a user adds a network or a network device without IT’s knowledge, the AI (like other network monitoring tools) could remain unaware if the new network isn’t linked into the AI. One way to prevent this is through corporate policy (i.e., no one outside of IT can install a new network without IT’s knowledge), or by zero trust networks, which can detect an addition, subtraction or modification to any network or network device.

The incoming data AI uses must be perfect

If the data isn’t perfect, the AI is going to form conclusions on faulty information. Consequently, at the onset of every AI project, data cleaning and vetting must be performed so maximum data quality can be achieved. Data (and data quality standards) will change over time, so IT needs to periodically assess the quality of data to see if it maintains its quality standards. If there is data quality degradation, adjustments will need to be made.

AI must be linked into everything

Existing network monitoring tools, networks, and devices must have APIs (application programming interfaces) that can easily integrate with the AI. If the AI fails to connect with even one network device or monitoring tool, its assessments could be off.

The AI model requires periodic reviews

If the AI begins to deliver results that are less than accurate and that vary from what a network expert would conclude, it’s time to investigate and potentially tweak the AI model or ruleset. Over time, there is bound to be drift (and outcomes) from initial results.

Use your own predictive powers, too

An AI strength is that it can process a great deal of data and detect new network trends (such as an emerging security threat) from data patterns. Security hackers know this, too. That’s why they’re always trying to come up with a new element of surprise that isn’t so detectable by AI.

Because of this, it’s a good idea for network staff to research emerging network and security trends on their own. The AI is likely to miss some of this information.

What's a sound middle-ground approach?

A sound middle ground approach for the network is to continue to use the monitoring tools that you’ve been using, but also to look for ways to introduce AI. In many cases, your vendors will do this for you as they move more of of their tools into an AIOps environment.

As this happens, it’s important to keep in mind that network tools (including AI) aren’t perfect.

A modicum of human control over the entirety of network operations will always be needed, including the ability to override what the automation might want to do in a particular situation.

About the Author(s)

Mary E. Shacklett, President, Transworld Data

Mary E. Shacklett is an internationally recognized technology commentator and President of Transworld Data, a marketing and technology services firm.

Stay informed! Sign up to get expert advice and insight delivered direct to your inbox

You May Also Like

More Insights