What is AI networking? Use cases, benefits and challenges

 

 

AI networking overview

“AI networking offers great potential to disrupt long-standing traditional networking operations to create a massive productivity increase.” Gartner, Innovation Insight: AI Networking Has the Potential to Revolutionize Network Operations

Artificial Intelligence for Network Operations (AIopss) is expanding into AI networking, focusing on ongoing “day 2” network administration, maintenance, and optimization. Networking infrastructure and AI are combined to automate and optimize IT operations.

AIOps is more broadly focused on the information and operations (I&O) infrastructure level, while AI networking is specific to the networking domain (data center switching, wired, wireless, LAN, WAN, SD-WAN, multicloud).

Though the concept has previously existed under several titles that effectively related to the same job, Gartner coined the phrase in 2023. It has been referred to by vendors as intent-based networking, autonomous networks, self-driving networks, and self-healing networks.

Use cases for AI networking

 

IT infrastructure is critical to today’s enterprise, but it can be complex and difficult to manage, and IT teams often require specific, high-level skills to identify, troubleshoot and solve network problems. Additionally, network managers are bombarded with alerts from all angles that can be difficult to sift through and prioritize. All of this is complicated by the ongoing talent shortage of IT workers, which makes automation an urgent matter.

AI networking seeks to transform traditional IT operations and make networks more intelligent, self-adaptive, efficient and reliable. The technology uses machine learning, deep learning, natural language processing (NLP), generative AI (genAI) and other methods to monitor, troubleshoot and secure networks.

 

Core capabilities include:

 

Automation of networks

Tasks like network configuration, monitoring, and troubleshooting are automated by AI networking. As a result, downtime is decreased, performance is enhanced, and resource allocation is optimized. Recommendation and response are automated, as are configuration and problem management, software updates, and other tasks.

ITSM that is optimized

AI networking can optimize IT service management (ITSM) by handling the most basic level 1 and level 2 support issues (like password resets or hardware glitches). Leveraging NLP, chatbots and virtual agents can field the most common and simple service desk inquiries and help users troubleshoot. AI can also identify higher-level issues that go beyond step-by-step instructions and pass them along for human support.

AI networking can also help reduce trouble ticket false-positives by approving or rejecting tickets before they are acted on by the IT help desk. This can reduce the probability that human workers will chase tickets that either weren’t real problems in the first place, were mistakenly submitted or duplicated or were already resolved.

These AI handoffs can improve response times and reduce IT staff workloads, allowing them to focus on strategy and more advanced tasks. AI networking can also enhance operational efficiency and reduce human error caused by alert burnout.

 

Improved network management and performance

AI can analyze large amounts of network data and traffic and perform predictive network maintenance. Algorithms can identify patterns, anomalies and trends to anticipate potential issues before they degrade performance or cause unexpected network outages. IT teams can then act on these to prevent or at least minimize disruption.

AI networking systems can also identify bottlenecks, latency issues and congestion areas. Through ongoing analysis of workloads, resource utilization and demand forecasts, AI can allocate network resources, scale infrastructure, reroute traffic where needed and improve quality of service (Quality-of-service (QOS)).

AI networking can also accomplish the following:

• Determine probability of hardware failure — such as a faulty CPU or flash drive — and resolve it when convenient.

• Correlate multiple datasets to determine source of latency or other issues.

• Respond to spikes in demand by requesting more bandwidth or rerouting to alternative channels based on noise, interference or congestion.

• Determine the cause of high server response times.

• Generate vendor-neutral troubleshooting or provisioning so that the IT team doesn’t have to be versed in every vendor’s specific platforms or jargon.

Incident management

Through correlation analysis, pattern recognition and other methods, AI algorithms can target incident causes and suggest remediation actions. This can reduce the IT team’s time and effort in identifying, diagnosing and resolving issues.

Intelligent security

AI can look at traffic, user behavior and system logs to pinpoint anomalies and flag potential security breaches or attacks. This can support proactive threat detection, response time, mitigation and network protection. AI can also respond to cybersecurity issues in real time.

Day 0 and day 1 functions

While it homes in on day 2 operations, AI networking also supports day 0 and day 1 functions, including network design, setup and recommendations to optimize network performance.

Main components of AI networking

AI networking incorporates numerous technologies, including the following:

Predictive analytics

Through data analysis and statistical models, AI can learn to understand a network and its policies. It can study and process predefined metrics, traffic flows, trends and patterns and compare them against established baselines.

Trend analysis and pattern recognition

Algorithms can analyze trends and use pattern recognition to make sense out of real-time and historical data. Through monitoring and observability, AI can process event and telemetry data to detect incidents as they occur.

The system does this by creating a baseline from historic data, then continually learning and refining — with or without human-in-the-loop — patterns of events based on data as well as human operator input, guidelines, reaction and interaction.

Event correlation

Using baseline models, time-series and topology information, AI can compress and correlate events across telemetry domains and group-related events, thus reducing the need for human intervention.

Closed-loop problem resolution

AI networking continuously learns and improves associations between events and human responses, whether through explicit actions, guidance or simple observation. This process might trigger a system to offer recommendations or take action itself based on its training and parameters.

GenAI

AI networking will increasingly use genAI and large language models (LLMs), which can offer suggestions or create specific, catered plans of action.

For instance, Gartner posits, an engineer could ask a ChatGPT-like interface to design a leaf-spine network (consisting of two switching layers) that could support 400 servers using Vendor A. Using data (both public and organization- and industry-specific) the platform could then generate the required configurations for this specific prompt.

Digital twins

Using a simulated nonproduction environment, enterprises can validate the impacts of network changes before they are deployed in the physical world. A combination of AI and digital twins can also work into a continuous integration/continuous delivery (CI/CD) pipeline that can allow for “what if” scenarios and to ensure that the network is operating as expected.

Gartner predicts that by 2026, 50% of networking vendors will offer a digital twin capability in their tools, up from 10% in 2023.

AI networking comparison to AIOps

AI networking and AIOps are closely intertwined as they both fuse AI and ML with networking. But there are important differences.

AIOps has been around longer as a term and concept. Coined by Gartner in 2017, the technology enhances decision-making across I&O by aggregating and contextualizing large amounts of operational data. Its stages include initial data collection, model training, automation, anomaly detection and continuous learning.

As opposed to AI networking, which starts at the more strategic day 2 maintaining, monitoring and optimizing of the network, AIOps begins with day 0 planning and design, including defining business strategies and outcomes and identifying customer needs.

The system then ingests historical and real-time streaming data, filters out “noisy” data and identifies patterns in data. As it evolves into day 1 automation and remediation, the platform’s functions grow increasingly sophisticated as the system collects more and more data and continues to learn.

Day 2 and 2-plus capabilities then address user experience  through such capabilities as infrastructure and device health metrics, application-based context and pre/post connection performance  and provide more comprehensive AI-driven support based on closed-loop automation and self-remediation.

AIOps is touted for its impact on user, operations and DevOps/app experiences and location services.

 

Some AIOps capabilities include the following:

• At-scale automation

• Automated detection and resolution of anomalies before they have an impact

• Ongoing performance analysis and optimization

• Aggregation of operations data from multiple, disparate IT environments

• Reduction of false alarms

• Increased detection of malware traffic and vulnerabilities

• Causation determination through ML-driven root-cause analysis

• Improved operator efficiency and user experience due to conversational interfaces powered by NLP

Benefits of AI networking

Gartner asserts that AI networking can drive operational management savings by up to 25%. This is because it can reduce support calls, allow for improved troubleshooting, increase network availability and optimize end-user experience “that can’t reasonably be achieved by scaling manual resources,” the tech firm says.

Notably, AI networking simplifies network management, security and application infrastructures even as they become more complex due to disparate data center, multicloud, colocation and edge environments, as well as increasing abstraction layers (Kubernetes or containers).

AI networking can also achieve the following:

• Quickly respond to problems before humans discover them and before failure occurs

• Predict and prevent network problems

• Correlate data sources to centralize problem identification via typology and comprehension of contextual network correlations

• Determine if an appropriate resolution is available for a certain issue and generate the data flow required for additional investigation

• Optimize resource allocation and streamline ITSM processes

• Strengthen security

• Provide insights to humans and enable data-driven decision-making

Ultimately, Gartner says it has seen enterprises experience savings of more than 50% in areas including troubleshooting and install time.

Furthermore, because AI networking simplifies network management, workers don’t need to have deep network configuration and troubleshooting skills. Enterprises can automate via genAI and remove manual human setup.

And, with fewer workers needed to manage the network, organizations struggling with a skills/experience gap can manage networks in-house rather than outsourcing.

 

Challenges of AI networking

Still, ambiguity around AI networking  what it means and how it works remains, as the definition is new and fuses different concepts that vendors have been promoting for some time now.

This lack of clarity and mixing of terminology has hampered implementation: Gartner estimates that the AI networking adoption rate is less than 10%. This indicates that enterprises are interested, but need more clarification on the technology and what it does.

 

Additional concerns include the following:

• Inaccurate AI recommendations leading to incorrect network configurations, creating unnecessary complexity or causing outages or other issues. This can stem from incorrect prompts from users, or occur when a system hasn’t been trained correctly or with enough data.

• Tool sprawl: Enterprises are increasingly concerned with “technical debt” and the associated costs.

• Culture and buy-in: Network management personnel can be risk-averse and may not trust AI tools or unproven recommendations. Similarly, some workers may eschew the technology for fear of it replacing their jobs, or because they are content with the status quo. These factors can limit the value of AI investment and its potential benefits.

• Enterprises may lack sufficient, quality data to provide proper insights or resolve issues.

• Technical skills for areas such as prompt engineering may require additional time and resource investment. While the idea is to automate workflows so that human workers can focus their talents on more strategic, high-level tasks, new data science skills may emerge as systems evolve. For instance, users may need to become versed in prompt engineering, or equipped with skills to effectively analyze AI outputs.

• Inflated expectations: There’s a lot of hype around AI. And because the technology is so new, there is little standardization and vendors may overstate their capabilities. This can lead to oversetting expectations that exceed reality, and enterprises may not get the value they’d hoped for, with systems providing incremental or negligible benefits.

Recommendations moving forward with AI networking

Gartner predicts that by 2027, 90% of enterprises will use some AI to automate day 2 network operations. Similarly, the firm says that by 2026, genAI technology will account for 20% of initial network configuration, up from near zero in 2023.

Moving forward, AI networking will be offered via numerous methods. These include the following:

• AIOps platforms that take a horizontal approach at a broad infrastructure level.

• Network vendors that incorporate AI into existing tools including SD-WAN, access points and switches.

• Multivendor tools that provide AI networking across several tools.

• Providers of a managed service incorporating AI networking.

As with any new or evolving technology, enterprises should proceed with care and due diligence. Gartner and other experts make numerous suggestions as enterprises explore AI networking technologies, their use cases and benefits. These include the following:

• Starting small and performing proof of concept (PoC) and testing before adopting tools and rolling them out into production. When developing a strategy, enterprises should act on the AI’s recommendations and predictions, then incrementally rely on automation as the system proves itself (or doesn’t) and human trust goes up.

• Identifying what type of AI networking system — AIOps platforms, network vendors, multivendors or managed service providers — work best for their particular enterprise based on resources and needs.

• Selecting AI networking vendors after determining whether their operational model is do-it-yourself (DIY) or managed network services (MNS), and also whether the networking environment is single vendor or multivendor. This will help companies determine which type of vendor is the right option for them.

• Requiring vendors to provide specific details and offer complete breakdowns of their tools and what they can (and cannot) do and deliver. Organizations should ensure that these potential future partners provide timelines. This should be from implementation to one to two years out (or longer).

• Identifying how network operations will change and evolving roles from network management. IT teams should be armed with data consumption capabilities and be able to analyze and act on AI recommendations.

• Justifying adoption by calculating cost savings and benefits when it comes to resource efficiency, network availability, performance and improved experiences.

Key takeaways for AI networking

• AI networking combines AI with networking infrastructure to automate and optimize IT operations.

• AI networking has the capability to transform network management by automating maintenance, troubleshooting and incident management. Algorithms have the enhanced capability to make data-driven predictions that improve security, performance and operations.

• AI networking incorporates a number of different technologies, including predictive and trend analysis, pattern recognition, digital twins and genAI.

• The benefits of AI networking include operational management savings of up to 25% through reduced support calls, improved troubleshooting, increased network availability and optimized end-user experience, among other things.

• Risks include inflated expectations and hype, inaccurate outputs, insufficient data to power systems and cultural buy-in. New skills may also be required, including prompt engineering and analysis of AI output.

• While AI networking and AIOps are sure to be disruptive, enterprises must be deliberate in their approach and determine the best platform for their business needs and strategies.

Synonyms, acronyms, abbreviations

• Artificial intelligence networking

• AI networking

• AI network

• AI networks

• Intent-based networking

• Autonomous networks

• Self-driving networks

• Self-healing networks