KOHb - Getty Images
AI has become a hot topic -- especially the subcategory of generative AI, with tools such as OpenAI's ChatGPT and Google Bard captivating people across virtually all industries. While these technologies create vast opportunities, we want to focus on the AI and automation opportunity for IT operations and network operations teams specifically.
If you've been in IT long enough, you've seen what happens when consumerization of a technology creates rapid adoption. In this case, generative AI usage grew to hundreds of millions of users in just a few months. It stands to reason that enterprise CEOs and their boards might ask how generative AI applies to their companies and start throwing money at it. In addition to the excitement, there's a lot of confusion as it creates a wave of hype and "generative AI washing" across all industries.
While the excitement around generative AI is fun to watch, it's important to remember that AI and machine learning (ML) have existed for decades. Anyone remember IBM's Watson competing on Jeopardy!? That was in 2011. And its development began five years prior to that. In fact, generative AI has existed for nearly 10 years at this point.
While the headlines talk about AI replacing workers, it's been more about workers using AI to make their jobs easier. The CEO of IBM, Arvind Krishna, recently said, "You won't lose your job to AI, you will lose your job to the person using AI."
Why is AI needed for the network?
Research findings from "2023 SASE Series: A Network Perspective on SASE and SD-WAN" by TechTarget's Enterprise Strategy Group highlighted that 73% of organizations believe their network environment has become more complex over the last two years. So, services that use AI to find problems quickly, send alerts and even provide recommended fixes will be increasingly in demand to help simplify network operations.
Network vendors have been using cloud-based management tools to anonymously collect and analyze network data from all the customers using their products. Over time, this provides a lot of valuable information. Vendors can study this data and write algorithms to identify, alert and provide recommendations. Keep in mind, these are not the large language models in ChatGPT or Google Bard that scrape information from the internet. Rather, think of these services as small language models, or SLMs, that comprise only detailed network data from a particular vendor. While most of these AI/ML models couldn't compete on Jeopardy!, they can drive significant operational efficiencies for organizations.
Today, many network vendors include AI-centric capabilities, such as AIOps, that use the data collected from a network and provide real-time intelligent alerts and recommendations to solve a problem, even automating responses in some cases.
But just because you can automate based on AI, should you?
What is holding AI adoption back?
Network operations teams tend to be conservative -- and there's a reason for that. If the network goes down, it could cost a business millions of dollars per minute. As a result, these operations teams have learned not to rush out and deploy the latest technology just for the sake of having new technology. Consider how long it still takes to deploy new, but proven, technology from an existing and trusted vendor. This is because the network team's top priority is ensuring the network is always available, secure and delivering positive experiences.
For most network operations teams, the concern isn't about being replaced by AI, but more about job loss because they over-relied on AI to a point that something was done automatically, creating a fault that brought the network down. Enterprise Strategy Group's "End-to-end Network Visibility and Management Trends" research supports this approach: The majority of respondents (82%) said they would like to use AI/ML services to create either intelligent alerts or recommendations to fix a problem. But confidence in AI is still in question, as IT teams said they prefer to remediate the problem themselves. According to the research, only 18% are interested in fully automating the network environment.
It's important to remember the difference between using automated runbooks as part of the response and having a fully automated response. The former would involve the team member manually kicking off an Ansible playbook or equivalent, while the latter automatically makes changes without human involvement.
So, a big part of building trust in AI services for operations teams is vendors incorporating closed-loop systems that offer team members an opportunity to provide feedback regarding the alerts and recommendations generated. For example, operators can confirm if the recommendation was correct. If it's not, then the operators can supply the system with the correct fix to continuously improve the underlying recommendation algorithm.
This closed-loop system is critical for network operations teams as it enables them to better understand the efficacy of the AI service and ultimately become comfortable letting the AI software automate responses based on specific use cases that they approve. Organizations adopting AI in their network environments are in the process of defining what that time to comfort is -- the amount of time it takes for operations teams to validate that the recommendations provided by the AI programs are accurate and trustworthy. You can expect those times will vary based on the team and the severity of the alerts.
Where does generative AI fit in?
Generative AI tools can enable operations teams to use natural language prompts to manage the network. A query as simple as "How is the network performing today?" or "Were there any problems last night?" could dramatically simplify network operations. But is that really differentiated from what currently exists? Do network operations teams need generative AI?
Given that generative AI is all about net new content generation, is it required to let you know a specific metric has crossed a threshold? Not really; those alerts are not open for interpretation. But what if a network operator could ask generative AI to create a synthetic data flow that introduces an anomaly it hasn't experienced yet, then gauge what alerts and recommendations the system would deliver. When used in conjunction with a digital twin, organizations could create a number of resiliency or what-if scenarios to test the network and provide significant insights and benefits to operations teams and the business.
The future of AI looks bright
The use of AI and automation in the network and overall IT environment will only continue to grow. It will become a requirement, especially as network environments continue to support more highly distributed and complex environments. Generative AI services could take network operations to the next level, especially when combined with digital twin technology.
For organizations getting started with AIOps, it's important to verify that the data used to train the model is relevant, trusted and of high quality. This means it uses network data collected by the vendor of your choice, not data scraped from the internet.
Remember to give your experienced network operations teams the time to validate and become comfortable with the alerts and recommendations generated. It would also be prudent to work with vendors to understand their AI roadmaps, including generative AI, and create programs to test new capabilities as they are introduced. At the end of the day, this is all about keeping the network up, optimized and secure. Resiliency is foundational to all these goals, and generative AI can enable it.
AI is not going away, so network operator/AI augmentation is the future. Now is the time to embrace it and create a bright future for network operations.
Enterprise Strategy Group is a division of TechTarget. Its analysts have business relationships with technology vendors.