Vendors are building different types of AI products. AI-enhanced analytical tools are being built, as well as the hardware components required for building AI infrastructure. I was fortunate to be a delegate at the inaugural AI Field Day. Over three days, we heard from vendors who are bringing both types of AI IT products to market. This post reviews AI-powered networking tools. Additionally, it provides evaluation recommendations.

AI Enhanced Analytical tools

AI, or more precisely machine learning, can be used  automate all sorts of IT operational tasks. Gartner calls this AIops:

 “Artificial Intelligence for IT operations combines big data and machine learning to automate IT operations processes, including event correlation, anomaly detection and causality determination”. – Gartner 2019

The 3 networking companies who presented to us have been working with machine learning technology to make their networking analytical tools smarter. Here are the high points of each presentation.

Virtual Network Assistant for Network Ops

What if you could build an AI tool that could keep an eye on your network, alerting you proactively about problems or threats? This is exactly why the Mist Systems team (now Juniper) built Marvis. Marvis is a Virtual Network Assistant (VNA) solution. It is a conversational assistant that uses NLP (natural language processing).

The concept is Marvis actually becomes another member of the networking ops team. The hardest part may be getting the human team to trust their new VNA co-worker. Ultimately, the goal is to get all data from an access point (AP) and could feed Marvis what it needs to become a self-driving action framework.

screen shot of Marvis by Mist Systems

Marvis reportedly can reduce network ticket average resolution time from 48 hours to 2 hours. Those numbers are pretty hard to ignore. However, could you train and trust a VNA as you would a new hire?

AIOps from Aruba

Aruba’s  AI-enhanced analytical tools include an AI platform named ESP. No, not this one. Aruba’s cleverly named tool stands for Edge Services Platform. They created this model to harvest data shed by networking devices to provide AIOps for networking teams:

 

Aruba also reports that their AIOps platform reduces mean time to resolution, trouble tickets, and eliminates manual, time-intensive troubleshooting. Additionally, they told us that the engine’s recommendations provide greater than 90% accuracy.

AI-Enhanced Analytical Tools for Security 

Cisco told us their how their AI products helps with security. Their AI-enhanced analytical tools for security uses ML-based classification and ML-behavioral spoofing and attack detection. Their algorithms focus on anomalies. So their tools start paying attention when a device starts to act differently. JP Vasseur was an excellent presenter, explaining complex topics with ease.

 

Questions to Ask Vendors

These tools are powerful. However, deploying them to your environment puts your infrastructure in the hands of automated products and virtual assistants. Therefore it is vital to perform due diligence, don’t just ask the usual product evaluation questions! Here are three things to consider:

  1. Most of these tools are SaaS services. Therefore, you must scrutinize the security practices of the service. It’s not enough for vendor to only send metadata to their cloud, and not data packets.  When metadata is combined with other data it can rehydrated and reveal the original data. Ask the vendor for details of how the metadata is handled.
  2. Secondly, what is the vendor doing with metadata? Software vendors have always combined analytics their customer base. This information helps them identify trends to make the product better as well as general market trends. However, is your vendor combining data from all of their customers into one dataset? Additionally, will these data sets be sold to third parties or partners? Or will they provide the datasets as an educational resource to outside sources? This isn’t unprecedented. Cambridge Analytica acquired their data set from Facebook under the guise of “academic research”.
  3. Finally, how was the vendor’s machine learning model built? It is common practice to use a publicly available model to jumpstart the creation of a new learning model. However, we’re discovering that many publicly available models have inherit or even explicit bias. Will a biased model impact your organization in the future?

Real Talk

Without a doubt, machine learning and AI will help operators by automating mundane tasks. These AI-enhanced analytical tools put the data a machine sheds to a good use! Additionally, these tools can also remove the most common cause of IT errors: human error. Eventually, these tools will also be self-healing systems, freeing up the human admins to do higher-order tasks. However, it is important to do the due diligence and analyze the fine print before giving up control of your IT estate.

What are you seeing out there? Are you using any of these tools to help you protect your environments?