Behind the Scenes: How Real AI Drives Real Results for the Enterprise

Juniper Networks’ recent global survey on AI readiness and adoption shows that 75% of APAC respondents now consider AI to be a strategic priority for their businesses. This comes as no surprise, with the deployment and operation of an entire network becoming more complex.

Our recent blog by Lee Ming Kai, Head of Systems Engineering, APAC, looks at how the Juniper AI-Driven Enterprise leverages AI to deliver experience-first networking and rapid time to value with networks that “just work”, from Day 0 onwards.

In this blog, we take a look at what’s happening behind the scenes to deliver experience-first networking for service providers, enterprises, IT teams and their end users.

Taming the Complexity of Growing IT Networks

What started as straightforward Ethernet CSMA/CD best effort connectivity must now support increasingly complex requirements. Demanding services that are sensitive to latency and loss – such as Voice over Internet Protocol (VoIP) and video conferencing – need to be delivered simultaneously with the high endpoint scale requirements of smart devices and the internet of things, all of which must achieved across both wired and wireless networks.

Adding features and functionality to the network has dramatically increased this complexity and stretched operational resources. Yet, even as older generations of networks have drawn closer to the limits of their capability, the network for the next decade has already arrived: the Juniper AI-Driven Enterprise.

With the AI-Driven Enterprise, Juniper is not only delivering experience-driven networking across wired, wireless, SD-WAN and client to cloud, we’re also extending the use of AI beyond wireless networks into wired and wide area networks. This is a significant development that has proved to be quite a revelation for some customers, who are now recognizing the root cause of problems that had become part of everyday life. For example, after deploying Marvis to their wired network, one organization found that multiple switches were misconfigured, which explained why voice calls had been dropping in a certain area of their building for some time.

Setting new standards in the Cloud + 5G era, Juniper’s AI-Driven Enterprise provides versatility, scalability and flexibility, while making the network easier to operate. Everything is simpler and easier to manage, thanks to automation, artificial intelligence and machine learning.

“As an organization built on a network-based work environment, we expect to evolve and diversify our offerings in the future, which will result in a demand for higher speeds. We are confident that the various AI-driven network upgrades by Juniper will drive value in every corner of our organization and put us in a strong position for continued growth.”

Young Ki Kim, Director, JOINS JoongAng

Why Talk About Real AI in Networking?

In recent years, artificial intelligence has been hyped by many and there can be confusion about what it actually is and what it does. Trying to get to the root of how AI is being used can be like peeling back the layers of an onion: it’s accompanied by lots of tears and blurry vision. For example, many companies are merely using statistical methods for anomaly detection while claiming to be using AI.

Statistical analysis using normal distributions and correlations often results in poor accuracy in anomaly detection, which can lead to missed actual positives and trigger a large volume of false positives that a human operator must investigate. Using more sophisticated techniques, such as autoregressive integrated moving average (ARIMA), anomaly detection is a better way of predicting an expected outcome, but its accuracy is still around 80%. However, this is a big step up from simple standard deviation analysis, which is often less than 50% accurate.

Mist AI’s anomaly detection is more than 95% accurate. It uses augmented long short-term memory recurrent neural network (LSTM RNN) to accurately detect network problems.

Not only does Mist AI predict, but it also automatically responds and remediates. It enables network systems to self-correct for maximum uptime and provides prescriptive actions for fixing any problems. For example, if Marvis (Mist’s inbuilt virtual network assistant) detects that an access point (AP) is going to fail, it can trigger a request to send a new AP to the customer. The radio resource management (RRM) – enhanced by reinforcement learning – will track for persistent and transient interference while automatically increasing the signal strength, changing the channels and potentially adding a new spectrum from other neighboring APs to compensate. The user experience remains seamless.

In a real world example, SaaS organization ServiceNow wanted to be able to detect and resolve a Wi-Fi problem before a user knew it existed. By switching to Juniper, they eliminated over 90% of their user-generated trouble tickets.

It’s this highly accurate prediction and response capability, which Juniper engineers have developed over years, supported by a broad collection of data and a bespoke mix of AI and machine learning models that makes Juniper’s Mist AI real AI – able to provide users with an assured, seamless and secure experience wherever they are.

Built-in Location-Based Services Meet a Multitude of Needs

Mist AI also powers Juniper’s unique Indoor Location Services. An array of antennae is built into every Juniper access point, providing a virtualized Bluetooth low energy (vBLE) network overlay that delivers exceptional locational accuracy of one to three meters, and often less. There is no longer any need to install hundreds or thousands of battery beacons, no need to worry about them becoming lost, damaged or mysteriously “relocated”, and definitely no need to spend days changing batteries every six months or so. This means that location services can now be a practical reality for Juniper customers.

As location services in Mist are AI-driven, there’s no need for calibration. Instead, Marvis Virtual Network Assistant (VNA) looks at signal strength and uses deep learning to work out where it is coming from. It can then pinpoint the user or device’s location on a floor plan or map. Many enterprises are already using this capability to locate devices to manage room and venue capacity in a way that supports social distancing. Mist AI is able to track the number of people there are in a given area and send out an alert when the maximum is reached.

Location services can also be used for proximity tracing. This can have a significant impact in reducing the number of people who may need to self-isolate following a positive COVID-19 test. Only the people who have been in close proximity to the affected person will need to quarantine, instead of the entire organization. Mist AI can also report on the duration and proximity of the contact, allowing organizations to implement different levels of quarantine and isolation for their employees.

There are many other uses for location-based services, including supporting user engagement, indoor navigation (wayfinding), on-site safety and security measures, reduction in energy use, better space utilization and facilities management. It’s also extremely useful for locating portable equipment that has been lost or misplaced, even across the largest of factory, hospital or university sites.

AI Simplifies Operations

Marvis uses Mist AI’s natural language processing to simplify the operation of an increasingly complex network. For example, it’s no longer necessary for a user to have special training to troubleshoot and isolate an issue. They can simply ask, “Why is Yeong Hui having issues?” Marvis will respond by delivering a root cause analysis for the devices that Yeong Hui is using and provide recommended fixes with supporting graphs and logs.

Because Marvis runs on Mist AI, it doesn’t need to be trained from scratch, as it has already accumulated years of learning that are continuously being updated. None of the information it holds is used directly, and all data is anonymized for training purposes. And while discussions about AI will often bring up the subject of bias, this isn’t a real issue when dealing with devices, as long as the sample sizes have been sufficiently large.

AI That Repays the Market’s Trust

AI is a growing part of our everyday lives, whether we realize it or not. Thanks to increasing AI adoption and the work by the governments of Japan, South Korea, China, India and Singapore (which introduced the Model AI Governance Framework at the World Economic Forum of 2019), organizations and individuals in APAC already put a high level of trust in AI. This trust is only set to increase, once the opportunities and benefits are better-known.

Juniper was one of the first vendors to recognize that the IT service model must change in order to keep up with growing complexity and address the needs of the future. AI is the way forward for a better network experience now, and in the years to come.