Artificial intelligence helps solve networking problems

With the public release of ChatGPT and Microsoft’s $10-billion investment into OpenAI, artificial intelligence (AI) is quickly gaining mainstream acceptance. For enterprise networking professionals, this means there is a very real possibility that AI traffic will affect their networks in major ways, both positive and negative.

As AI becomes a core feature in mission-critical software, how should network teams and networking professionals adjust to stay ahead of the trend?

Andrew Coward, GM of software defined networking at IBM, argues that the enterprise has already lost control of its networks. The shift to the cloud has left the traditional enterprise network stranded, and AI and automation are required if enterprises hope to regain control. “The center of gravity has shifted from the corporate data center to a hybrid multicloud environment, but the network was designed for a world where all traffic still flows to the data center. This means that many of the network elements that dictate traffic flow and policy are now beyond the reach and control of the enterprise’s networking teams,” Coward said.

Recent research from Enterprise Management Associates (EMA) supports Coward’s observations. According to EMA’s 2022 Network Management Megatrends report, while 99% of enterprises have adopted at least one public-cloud service and 72% have a multicloud strategy, only 18% of the 400 IT organizations surveyed believed that their existing tools are effective at monitoring public clouds.   

AI can help monitor networks

AI is stressing networks in both obvious and nonobvious ways. It’s no secret that organizations that use cloud-based AI tools, such as OpenAI, IBM Watson, or AWS DeepLens, must accommodate heavy traffic between cloud and enterprise data centers to train the tools. Training AI and keeping it current requires shuttling massive amounts of data back and forth.  

What’s less obvious is that AI enters the enterprise through side doors, sneaking in through capabilities built into other tools. AI adds intelligence to everything from content creation tools to anti-spam engines to video surveillance software to edge devices, and many of those tools constantly communicate over the WAN to enterprise data centers. This can create traffic surges and latency issues, among a range of other problems.

On the positive side of the ledger, AI-powered traffic-management and monitoring tools are starting to help resource-constrained network teams cope with the complexity and fragility of multi-cloud, distributed networks. At the same time, modern network services such as SD-WAN, SASE, and 5G also now rely on AI for such things as intelligent routing, load balancing, and network slicing.

But as AI takes over more network functions, is it wise for enterprise leaders to trust this technology?

Is it wise to trust AI for mission-critical networking?

The professionals who will be tasked with using AI to enable next-generation networking are understandably skeptical of the many overheated claims of AI vendors.

“Network operations manage what many perceive to be a complex, fragile environment. So, many teams are fearful of using AI to drive decision-making because of potential network disruptions,” said Jason Normandin, a netops product manager for Broadcom Software.

Operation teams that don’t understand or have access to the underlying AI model’s logic will be hard to win over. “To ensure buy-in from network operations teams, it is critical to keep human oversight over the AI-enabled devices and systems,” Normandin said.

To trust AI, networking professionals require “explainable AI,” or AI that is not a black box but that reveals its inner workings. “Building trust in AI as a reliable companion starts with understanding its capabilities and limitations and testing it in a controlled environment before deployment,” said Dr. Adnan Masood, Chief AI Architect at digital transformation company UST.

Explainable and interpretable AI allows network teams to understand how AI arrives at its decisions, while key metrics allow network teams to track its performance. “Continuously monitoring AI’s performance and gathering feedback from team members is also an important way to build trust,” Masood added. “Trust in AI is not about blind-faith but rather understanding its capabilities and using it as a valuable tool to enhance your team’s performance.”

Broadcom’s Normandin notes that while networking experts may be reluctant to “give up the wheel” to AI, there is a middle way. “Recommendation engines can be a good compromise between manual and fully automated systems,” he said. “Such solutions let human experts ultimately make decisions of their own while offering users to rate recommendations provided. This approach enables a continuous training feedback loop, giving the opportunity to dynamically improve the models by using operators’ input.”

AI can assist network support with natural-language chat

As enterprise networks become more complicated, distributed, and congested, AI is helping resource-strapped network teams keep up. “The need for instantaneous, elastic connectivity across the enterprise is no longer just an option; it is table stakes for a successful business,” Coward from IBM said. “That’s why the industry is looking to apply AI and intelligent automation solutions to the network.”

The fact is that AI-powered tools are already spreading throughout cloud and enterprise networks, and the number of tools that feature AI will continue to rise for the foreseeable future. Enterprise networking has been one of the sectors most aggressively adopting AI and automation. AI is currently being used for a wide range of network functions, including performance monitoring, alarm suppression, root-cause analysis, and anomaly detection.

For instance, Cisco’s Meraki Insight analyzes network performance issues and helps with troubleshooting; Juniper’s Mist AI automates network configuration and handles optimization; and IBM’s Watson AIOps automates IT operations and improves service delivery.

AI is also being used to improve customer experiences. “AI’s ability to adapt and learn the client-to-cloud connection as it changes will make AI ideal for the most dynamic network use cases,” said Bob Friday, Chief AI Officer at Juniper Networks. Friday said that as society becomes more mobile, the wireless user experience gets ever more complex. That’s a problem because wireless networks are now critical to the daily lives of employees, especially in the age of work-from-home, which forces IT to support users in environments over which IT has little to no control.

This is why AI-powered support is one of the most popular early use cases.

“AI is enabling the next era of search and chatbots,” Friday said. “The end goal is an environment where users enjoy steady, consistent performance and no longer need to spend precious IT resources on mountains of support tickets.”

Chatbots and virtual assistants built with Natural Language Processing (NLP) and Natural Language Understanding (NLU) can understand questions that users ask in their own words. The system responds with specific insights and recommendations based on observations made across the LAN, WLAN, and WAN.

“Where this client-to-cloud insight and automation simply was not possible just a few years ago, today’s chatbots can utilize NLP capabilities to provide context and meaning to user inputs, allowing AI to come up with the best response,” Friday said. “This far surpasses the simple ‘yes’ or ‘no’ responses that originally came from traditional chatbots. With better NLP capabilities, chatbots can progress to become more intuitive, to the point where users will have a hard time telling the difference between a bot and a human.”

The early stages of this vision are already underway. AI is currently being used to help Fortune 500 companies accomplish such things as managing end-to-end user connectivity and enabling the delivery of new 5G services.

Gap turns to AI-powered operations and support

Retail giant Gap’s in-store WLAN networks were originally designed to accommodate a handful of mobile devices. Now these networks are used not only for employee connections to centralized resources, but also to connect shoppers’ devices and an increasing array of retail IoT devices across thousands of stores.

“Wireless in retail is really tough,” said Snehal Patel, global network architect for Gap

Inc. As more clients connected to Gap WLANs, a string of problems emerged. “Stores need enough wireless capacity to support innovation, and the network operations team needs better visibility into issues when they arise,” Patel said.

Gap’s IT team searched for a WLAN technology that would leverage the scale and resiliency of public clouds, but the team also wanted a platform that included tools like AI and automation that would enable their networks to scale to meet future demand.

Gap eventually settled on a set of tools from Juniper. Gap deployed Juniper’s Mist AI, an AI-powered network operations and support platform, Marvis VNA, a virtual network assistant designed to work with Mist AI, and Juniper’s SD-WAN service.

Gap’s operations team can now ask Marvis questions, and not only will it tell them what’s wrong with the network, but it will also recommend the next steps to remediate the problem.

“Before Mist, we spent a lot more time troubleshooting,” Patel said. Now, Mist continuously measures baseline performance, and if there’s a deviation, Marvis helps the operation team identify the problem. With enhanced visibility into network health and root-cause analysis of network issues, Gap has been reduced technical-staff visits to stores by 85%.

DISH taps AI to scale 5G for enterprise customers

Another Fortune 500 company that has adopted AI to modernize networking is DISH Network, which has deployed AI to enable new 5G services. DISH was seeing increasing demand for enterprise 5G services but was having a hard time optimizing its infrastructure to meet that demand.

Enterprise customers were seeking 5G services to enable new use cases, such as smart cities, agricultural drone networks, and smart factories. However, those use cases require secure, private, low-latency, stable connections over shared resources.

DISH knew that it needed to modernize its networking stack, and it sought tools that would help it deliver private 5G networks to enterprise customers on demand and with guaranteed SLAs. This was not possible using legacy tools.

DISH turned to IBM for help. IBM’s AI-powered automation and network orchestration software and services enable DISH to bring 5G network orchestration to both business and operations platforms. Intent-driven orchestration, a software-powered automation process, and AI now underpin DISH’s cloud-native 5G network architecture.

DISH also intends to use IBM Cloud Pak for Network Automation, an AI and machine-learning-powered network automation and orchestration software suite, to unlock new revenue streams, such as the on-demand delivery of private 5G network services.

Cloud Pak automates the complicated, cumbersome process of creating 5G network slices, which can then be provisioned as private networks. By automating the process, DISH can create enterprise-class private networks on 5G slices as soon as demand materializes, complete with SLAs.

 AI-powered advanced network slicing allows DISH to offer 5G services that are customized to each business. Businesses are able to set service levels for each device on their network, so, for example, an autonomous vehicle can receive a very low-latency connection, while an HD video camera can be allocated high bandwidth. 

“Our 5G build is unique in that we are truly creating a network of networks where each enterprise can custom-tailor a network slice or group of slices to achieve their specific business needs,” said Marc Rouanne, chief network officer, DISH Wireless. IBM’s orchestration solutions leverage AI, automation, and machine learning to not only make these private 5G slices possible, but also to ensure they adapt over time as customer use evolves.

How IT pros should prepare for AI

As AI, machine learning, and automation power an increasing array of networking software and gear, how should individual network professionals prepare to deal with their new artificial colleagues?

While few professionals will miss the mundane, repetitive chores that AI excels at, many also worry that AI will eventually displace them entirely.

“While AI is developing exponentially, it is inevitable network teams will be exposed to AI-enabled devices and systems,” Broadcom’s Normandin said. “As network experts are not meant to become AI specialists, a cultural change is probably more likely to happen than anything else.”

Masood of UST agrees that a cultural change is in order. “Network teams are rapidly evolving from just managing networks to managing networks with a brain,” he said. “Within the context of networking, these teams will need to develop the ability to work collaboratively with data scientists, software engineers, and other experts to build, deploy, and maintain AI systems in production.”

Network professionals will need to level-up existing skills in network management and optimization so they can accomplish such tasks as using machine-learning algorithms to predict network congestion and to improve network performance. They will also need to develop new skills in data analysis and visualization, NLP, outlier analysis, anomaly detection, and optimization algorithms. “I am not suggesting they become an AI developer or a data scientist,” Masood said, “but deeper understanding of the underlying algorithms and statistical models used to build networking specific AI systems will definitely give them a competitive edge over their non-AI-literate counterparts.”

Normandin said that a new role, NetDevOps, will emerge to manage AI-orchestrated networks. “Successful NetDevOps initiatives will look like fully automated environments that can deploy changes across networks, ready to be consumed in a DevOps approach all along the [continuous integration/continuous delivery] pipeline,” he said.

Programmable, software-defined, and cloud-based network environments have made NetDevOps possible through infrastructure-as-code and automation. “Now, network operations teams have to make their Agile revolution and accept the risk of more frequent changes and more automation,” Normandin said. “As a consequence, they’ll need to refocus on the main outcome: monitoring and guaranteeing the digital experience delivered by the networks.”

(Jeff Vance is a Network World contributing writer and the founder of Startup50.com, a site that discovers, analyzes, and ranks tech startups. Follow him on Twitter, @JWVance, or connect with him on LinkedIn.)