Blog AI & ML

CoPilot: Reduce Risk and Save Time Using Explainable AI

Jeevan Patil Senior Director, Product Management Published 8 Jun 2021

plane controls with pilot copilot

Hopefully, everyone reading this blog is in the process of getting vaccinated and itching to fly to your favorite destination for a summer vacation. I know, I am planning to take one as soon as ExtremeCloud IQ CoPilot launches! Imagine you walk into the airplane, turn to your left, and don’t see a pilot or a copilot. Would you be worried? Of course, you would!

The job of a Pilot and CoPilot

For the last few decades, most airplanes have been designed to take off, adjust to turbulence, fly themselves based upon a recommended path, and support a smooth landing. So, the pilot’s job is to enter the destination, feed the routes, interrupt everyone eating lunch when the altitude is above 10,000 feet, and offer cheesy jokes. On a more serious note, the pilot’s job is to be the manual override in case of need and provide all of us peace of mind. On the other hand, the copilot assists the pilot with radio communications, navigation, completing critical operational checklists. And most importantly, the copilot acts as an extra pair of eyes and ears to increase their situational awareness, ultimately helping the pilot focus on flying the plane smoothly.

Enter ExtremeCloud IQ CoPilot

Networks are complex, and most IT administrators, unfortunately, spend their time flying solo as they aim to monitor, react to alarms, and implement best practices. Therefore, administrators use several applications, multiple screens, and too many alerts that desensitize them. Cloud networking offers a powerful new way to manage networks. 

While cloud management inherently delivers flexibility and efficiency, machine learning (ML) and artificial intelligence (AI) compound the advantages. However, when you combine powerful analytics with the human touch of Extreme GTAC support, trends and anomalies can be detected and addressed rapidly. As a result, ExtremeCloud IQ CoPilot helps you save time and reduce risk.  

Beware of AI washing

Even worse, some vendors will engage in AI-washing, claiming that AI can solve every problem or that customers should automatically trust the recommendation from an AI algorithm. Anyone with data science background knows that these statements are just not true. There are several limitations of ML/AI, but only the relevant ones are covered here.

  1. Lack of Good Data – Data is the modern-day currency. The bigger the dataset, the higher the probability of determining patterns accurately without false positives. After data is collected, information needs to be cleaned and confirmed for good quality. All the cloud data lakes should ensure business continuity by performing periodic backups in geographically redundant locations. And data must be secure when it is in motion and at rest. With more data, your results get better. You should know that Extreme is the 2nd largest and fastest-growing cloud networking company and the only cloud networking company to offer you unlimited data.
  2. Privacy of your data – The next time your vendor markets any capability with ML/AI, ask them how many employees are allowed permission to access your data. Managing customer data in the cloud is a huge responsibility. Maintaining the highest levels of information systems and data protection, management, and compliance is essential. Furthermore, customers should understand that there is a repeatable process standardized across companies that manage sensitive data. ExtremeCloud IQ is the only cloud network management solution triple-threat, with CSA Star, ISO 27001, ISO 27017, and ISO 27701 certifications. For more details, check this out.
  3. The ability to trust – Just search for how many people have blindly driven straight into a lake while following their GPS instructions. If your bonus depends upon network uptime, would you blindly follow a vendor’s ML/AI recommendation? Therefore, the next AI superpower is Explainability which is driven primarily by the choice of algorithms that are used to predict and manage anomalies. To learn more, see this blog.
  4. Model drift – Over time, the relationship between the dependent variable and the independent variables changes over time. Remember how fashion changes over time. If you are unable to stay up to date with this, then the garment companies will suffer. The Extreme Networks engineering approach continues to protect against model drift by relentlessly checking against negative test cases.

Problem-solving with Extreme

Extreme takes a bifurcated approach, utilizing observability for health checks and advanced ML for anomaly detection. Health checks for entire networks, devices, and clients are visualized in the Network360, Device360, and Client360 dashboards within ExtremeCloud IQ. Machine learning algorithms achieve the calculation of these health scores. You might be wondering about the difference between the terms monitoring and observability. Monitoring answers ‘what’ questions (as in “what is the status of your client device?”). On the other hand, observability answers ‘why’ questions (as in “why is the client performance so slow?”).

Now let’s move to the anomalies. Given the complexities of IT networks and the scope of the anomalies observed, CoPilot further subdivides observations into two areas: network-wide issues and localized network issues. The focus is on critical yet often-seen problems with DHCP, DNS, and RADIUS which might disrupt wireless communications. Ethernet cable, PoE and multicast data anomalies may also impact the user experience. As shown in the figure below, multiple Wi-Fi issues are observed which might impact capacity and efficiency. The answers to these problems can be found using statistical anomalies with machine learning algorithms.

multiple Wi-Fi issues are observed which might impact capacity and efficiency

ExtremeCloud IQ CoPilot introduces explainable machine learning (ML) and artificial intelligence (AI) that an IT administrator can use to view and verify the data behind an insight.  When ML/AI alerts and resolution recommendations are auditable and explainable, ambiguity is removed, and trust is built.

But how do we know it really works?

As anyone in the ML/AI arena will attest, the idea is to run continuous experiments and verify. Therefore, at Extreme Networks, we use a multi-step process:

  1. Identify real issues identified from GTAC
  2. Define the problem crisply
  3. Acquire relevant data
  4. Calculate normal per customer
  5. Identify the anomalies and share the data
  6. Prioritize the anomalies per occurrence and severity
  7. Share actionable Insights with explanations
  8. Recommendation to customers
  9. Run a private beta to verify with customers
  10. Enable all of our Pilot customers availability via a public beta

Over the last 6+ months, we have been running a private beta with hundreds of customers on these anomalies. We had a school experience a Wi-Fi Efficiency issue, a church developed multicast problems, and a retail store experienced Wi-Fi capacity problems. Backed with intelligence from our collected data and our data scientists, our GTAC team reached out to the customers, verified, and addressed the issues. Now that we know we are onto something real, we decided to scale the solution. Starting next week, we begin a CoPilot public beta, we are rolling out these insights to all of our cloud customers. Come join us in for a test-flight – Everyone with a Pilot license is welcome!

extreme cloud iq copilot ad "delivering actionable insights"

Get the latest stories sent straight to your inbox!

Related Stories