Generative AI Poised to Redefine Wi-Fi Network Management – How Should Industry Players React?

Rapid advancements in Generative Artificial Intelligence (GAI) have the potential to disrupt and reshape entire industries, and the Wireless Local Access Network (WLAN) market is no exception. In the coming years, the development of GAI will transform WLAN network management by simplifying and streamlining the process through which users access the data they need. Existing tools for taking user commands and sharing data, based around Graphical User Interfaces (GUI) and text, will give way to a conversational approach based on a Conversational User Interface (CUI) leveraging GAI. Accompanying this will be the expansion of data collection capabilities of AI models and the convergence of multiple data sources, enabling a generation of more granular insights and personalized recommendations. With market disruption brought by GAI now imminent, the question becomes how should industry players ensure they are in a position to successfully exploit the new opportunities GAI offers?

The application of AI and Large Language Models (LLM) to WLAN networking is nothing new. Back in 2017, the small-scale WLAN vendor Mist Systems introduced the industry’s first AI-driven Virtual Network Assistant (VNA) with LLM capabilities, named Marvis. Alongside implementing cutting-edge AI for advanced network automation and analytics, Marvis also simplified and facilitated network management with its CUI, which allowed network administrators to ask queries and receive insights and recommendations in a natural language conversational format. Confident that advanced AI capabilities would help give them a competitive advantage over the competition, Juniper Networks snapped up Mist Systems for US$405 million in 2019 and leveraged their technology for the basis of the Juniper Mist cloud networking platform. While Juniper Networks was not alone in recognizing the potential of CUI’s in the late 2010s, they have overseen one of the best implementations of the technology. This is primarily because Mist AI was built from the ground up on a controller-free microservices architecture, specifically for the purpose of enabling unified wireless, wired, and Software-Defined-Wide Access Network (SD-WAN) management and monitoring. For contrast, in 2018, leading enterprise WLAN vendor Cisco acquired conversational AI startup MindMeld for US$125 million but has since struggled to successfully integrate it across its multiple (somewhat fractured) network management platforms. It was thanks Juniper Network’s industry leadership in AI that the company was ranked second for innovation in ABI Research’s recent enterprise WLAN vendor competitive assessment.

Recent developments in GAI, most notably the meteoric rise of ChatGPT, have intensified attention on and hopes for the technology, and several leading WLAN vendors have begun jostling to prove their GAI credentials and position themselves at the forefront GAI’s implementation for WLAN management. During Cisco’s most recent quarterly earnings call, CEO Chuck Robbins identified generative AI and security as key drivers of long-term growth, and they laid the groundwork for this with several significant announcements earlier this year. In March 2023, Cisco introduced additional AI-powered enhancements, such as automated chat summaries and the generation of actionable insights from customer interactions to its web conferencing and contact center-as-a-service platform Webex. This was followed by the May 2023 purchase of Generative AI startup Armorblox, which has pioneered the use of natural language understanding to increase cybersecurity for email, cloud office applications, and enterprise communications. This acquisition (alongside recent acquisitions of Valtix and Lightspin) will help build up Cisco’s AI-first Security Cloud. Despite these proactive moves, as demonstrated with Cisco’s struggle with MindMeld, bolting acquired technologies onto legacy systems will pose significant interoperability and consistency challenges.

Determined not to rest on their laurels, Juniper Networks is again pushing the boundaries of AI in WLAN networking, and in May 2023, announced two important updates to their VNA Marvis. The first was an integration with ChatGPT to deliver advanced LLM summarization and more authentic natural language conversation. Upon receiving a prompt, Marvis will first search Juniper Networks’ indexed internal documentation, and should this fail to provide an answer, the AI will then turn to ChatGPT. Over time it is envisaged that this ChatGPT-enhanced Marvis will grow to not only become more familiar with the customer’s specific network, but also the user’s language style, allowing it to provide increasingly bespoke answers. The second new fundamental component of Marvis is the incorporation of user experience feedback for reinforcement learning. This will bring considerable improvements to troubleshooting as the AI’s data collection capabilities are extended from the network layer (i.e., wired and wireless) into the application layer, first with Zoom, then later this year with Teams. Zoom cloud data will be natively integrated into the Mist cloud, so that real-time data from Zoom on metrics such as latency, packet loss, and jitter will be combined with data on network features including LAN, WLAN, client, and time data. Machine Learning (ML) analysis of this aggregated data will be used to predict the expected performance outcome, and then using Shapley values, Marvis will be able to identify and explain which network features are responsible for subpar performance. In practice, this will enable Marvis to directly inform the IT team of the cause of Zoom’s subpar performance—whether that be anything from wireless interference to an overloaded CPU on a laptop—and offer recommendations to improve performance. As with the ChatGPT integration, this global ML model will be refined through continuous user experience learning. This update will be extremely valuable for enterprises globally, as collaboration programs like Zoom are currently one of the major sources of support ticket problems, and these applications are already present on virtually all enterprise-related client devices, meaning customers are not required to install an additional agent on devices.

HPE Aruba Networking, the second largest enterprise WLAN vendor by scale (and overall leader in ABI Research’s recent enterprise WLAN vendor competitive assessment), has so far adopted a different approach towards GAI. Over the past twelve months, Aruba has released a slew of new AIOps capabilities designed to assist network operators in network management. Most recently, in April, the company unveiled new tools to help businesses deal with shortages of networking experts by applying ML models to identify precisely what tech generalists need to know about the network, and then automate the required processes. Although new visualization tools have been introduced to illustrate the state of the network, no CUI has yet been announced. While this may in part reflect caution towards what is a relatively new and untested technology, Aruba is also likely to have come to the conclusion that there are other AI-based capabilities which are in greater demand from their clients and will deliver a better Return on Investment (ROI). In other words, despite the high levels of hype around CUI, enterprises in the current economic climate are perhaps more likely to be concerned with staff shortages (which Aruba’s new tools can help resolve) than changing how they access data (especially if they believe legacy methods were working just fine).

Vendors looking to leverage GAI for Wi-Fi Network Management should consider some of the following factors:

Company Restructuring: To accommodate a greater focus on GAI it may be necessary to expand certain departments, for example, by adding new members to the data science team. Likewise, integrating departments might be key to the successful implementation of a GAI strategy. This could mean the combining of the data science team with the support team to facilitate the development and operation of a CUI-based support assistant.

Third-Party Interoperability: Enterprise customers are increasingly wary of being locked into a ‘walled garden’ ecosystem and are seeking unified management of equipment from multiple vendors within their network. Thus, for an AI assistant to be truly optimized to provide actionable analytics and recommendations for a customer’s unique network, the AI should have at the bare minimum visibility of third-party equipment on the network.

Client Compatibility: To ensure ease of access for customers, it is also important for a conversational interface to be available on a diverse array of clients supporting different platforms, for example, on both Windows and Android devices.

Real-Time Network Data Exploration: For GAI to be the most responsive and optimized for a customer’s specific network, access to real-time network data is key, and this requires visibility across the entire network and integration with core applications. Vendors looking to target a certain vertical should set as a core element of their strategy the integration with leading applications in that vertical. For example, vendors focusing on large public venues should look to bring tighter integrations with Wi-Fi analytics and indoor location services applications.  

Be Wary of Hype: One of the main advantages of a conversational interface is that it revolutionizes how data is accessed, and in the process, lowers the barrier for users to effectively manage the network. However, this may not be a priority for clients. As detailed above, whereas Cisco and Juniper are going full steam ahead with CUI, Aruba has instead focused its resources on other areas. It is important for companies to correctly identify what the priorities of clients are and to understand the true ROI possible from the introduction of CUI before they proceed with costly and lengthy development.

Growing Pains: While GAI models are built to learn over time, they may not be optimized nor fully capable for complex network management from the outset. For example, at present, ChatGPT only has the ability to answer junior network engineering questions, an issue that users of Juniper’s ChatGPT-enhanced Marvis may face. Extensive development on top of the GAI model via an API and extensive beta training may reduce this threat to an extent, but there is still the prospect of less-than-optimal results upon launch.

Security Vulnerabilities: One of the greatest concerns organizations have towards GAI integrations is security, particularly around whether company data will be captured by the model, and who will then be able to access this data. To address these concerns, vendors looking to utilize GAI may wish to consider developing their own solutions which have strict security controls, as opposed to implementing a third-party offering.

Regulatory Uncertainty: The rapid development of GAI has outpaced the ability of most governments to establish a policy framework to regulate the technology, but given the widespread transformation it will bring, further restrictions and even national bans can be expected. Mainland Chinese WLAN vendors in particular face tough regulatory challenges, as the draft policy on generative AI released in April by the Cyberspace Administration of China (CAC) includes provisions limiting their access to open-source GAI platforms, and closed-source ones like ChatGPT are also out of bounds in the country. This will mean that custom models will be required. Don’t assume that western markets will be spared, though. In April, Italy became the first Western country to ban ChatGPT due to noncompliance with GDPR. Vendors looking to introduce GAI should bear these risks in mind during development.