Are AI-Enabled Networks Enough?

Understanding Core Components and Capabilities

One thing I’ve learned over the years is that technology is always evolving. I’ve often told my colleagues and friends, “I used to say I’d have to reinvent myself every five years, but now it’s happening every two.” What’s driving this accelerated change? Creative technological innovations. Many clients have sought my opinion on AI and AI-powered solutions, particularly in areas like networking and security. While I believe these advancements can help organizations achieve more with fewer resources, I also advise exercising caution. No matter the vendor, most AI-enabled networks tend to incorporate the following components: 

Machine Learning: A branch of artificial intelligence that leverages algorithms to analyze data sets and build models, enabling machines to perform tasks that were once exclusive to humans. Essentially, it’s the ability to make predictions (or provide answers) based on the data it has processed. Think of it as advanced pattern recognition on steroids. 

Deep Learning: A subset of machine learning and a key element of data science. It powers applications and services that enhance automation, allowing systems to perform both analytical and physical tasks without human intervention. 

Natural Language Processing (NLP): A field of artificial intelligence that employs machine learning to enable computers to comprehend and interact using human language. 

Generative AI: A form of artificial intelligence that allows machines to generate content in response to prompts. This can include text, images, video, audio, or even software code. 

Autonomous Network: A new AI-driven network model that is self-managing and self-healing, requiring minimal human involvement. The goal is to create a self-regulating network that delivers greater efficiency, performance, and reliability, reducing costs and minimizing human error. Key pillars of such networks are typically: Sense, Think, and Act. 

The common objective I hear is that AI-enabled networks should make networks more intelligent, self-adaptive, efficient, and reliable. Manufacturers claim that an AI-enabled network can dynamically adjust workloads based on real-time data, ensuring optimal performance even during periods of high demand. The phrase I commonly hear is “Autonomous Networks” as the ultimate goal (though I have yet to see a solution that is fully there), or “AI-Powered Network Operations.” Is this adaptability a game-changer for managing the constantly shifting needs of modern applications and services? If it performs as advertised, absolutely.  

The Role of Data and Limits of AI in Network Operations

One key lesson I’ve learned about AI is that its accuracy depends heavily on the data it processes and the selection of the right data models for decision-making. Many people don’t realize that training an AI model to make informed decisions can take time, which is why there are so many AI-enabled products. Think of AI-enabled network operations as pre-trained models, similar to popular version of AI (v2, v3, v4) in the market today. With each version of AI, it improves based on the data it has ingested, creating a more refined product with every iteration. 

What are the “limits” of what AI can do on the network? That’s a great question, and although I dislike saying this, the answer is often, “it depends.” However, what I can say for now is that AI’s capabilities are mostly focused on network operations. As AI becomes integrated into networks, several dependencies must be considered. These include, but are not limited to: 

– Manufacturer/Product capabilities 

– Existing Infrastructure 

– Version or Generation of AI model/Platform 

– Compatibility & API 

– Your comfort & level of trust 

– Data Collection/Accuracy of Data 

As stated above, most AI-enabled networks primarily focus on network operations and automation. For example, consider a scenario where users are experiencing issues with “time-sensitive apps” crashing on the network.  Based upon the “Machine Learning” module, the AI may be able to recognize common variables that are causing this (such as ports not configured correctly), then utilize the “Deep Learning” module to correct this issue, then use the “NPL” module to create/update a ticket as to what the issue was and how it was resolved, then finally use the “Generative AI” module to ensure that an initial check is done so that anything that applies to this situation doesn’t happen again.   

An important factor to consider in this example is, what happens if the data isn’t accurate?

The Impact of Inaccurate Data on AI-Enabled Networks

Lets explore the Data Accuracy question. If you follow me on LinkedIn, you may remember a four-part series I wrote called: “Is Artificial Intelligence the Emerging Frontier?“. In that series, I discussed the potential of AI using the context of AI learning how to play Super Mario Brothers. I had a curious thought about AI and bad data that I decided to explore. So I restored the solution from my backups for this experiment and intentionally fed the AI incorrect or random data that had nothing to do with playing Super Mario Brothers. Unsurprisingly, as I kept inputting bad data or nonsensical information, the AI model struggled to perform. What happened was a series of odd or awkward movements, and in most cases, it couldn’t complete the level (the timer would run out). In a way, you can think of bad data in AI as technical debt, the more bad data it consumes, the less accurate and capable it becomes. 

If bad data, such as mislabeled, unavailable, uncollected, or improperly polled data, enters the system, it could slow down the process or prevent the AI from identifying the issue entirely.  

Someone might ask, “Wouldn’t the AI platform detect this and fix this automatically?” My initial thought is that the answer depends largely on your level of trust and comfort with the solution. If trust and comfort are in place, it could be possible. But does the AI solution have the necessary visibility, data, and authorization to make such decisions? 

When troubleshooting an issue or redesigning a network, what level of visibility would you require? This is a critical factor to consider when architecting an AI-enabled network solution. Another key consideration is which APIs from network-connected devices should be integrated into the solution. Many devices, such as compute, storage, applications, appliances, IoT devices and desktop applications come with APIs that allow for integration. I recommend having a structured discussion with your trusted technical advisor to ensure that, when the time comes, you have a clear understanding of what this could or should look like and what are the potential use cases. 

Securing AI-Enabled Networks: Key Steps and Considerations

How do you secure an AI-enabled network? This is a challenging question, and I may not be able to fully answer it without more context, but I can offer some high-level guidance and considerations. I’ve shared a few posts on security on KNZ’s website and LinkedIn. At the end of the day, AI is still an application with dependencies. Most AI-enabled applications require high-speed connections to their dependencies to function, but not necessarily to the devices they pull data from. A good starting point would be to create a dependency map, identifying what the AI-enabled network application connects to. 

A useful approach to consider is the principle of least privilege or even a minimal version of zero trust. It’s important to keep in mind that if an AI-enabled network solution is compromised, serious issues could arise. As I often advise, begin by defining the problem statement or scenario and work backwards from there. This is a solid strategy to ensure your organization is taking the necessary steps to secure any environment. 

For example, imagine that an AI-enabled network solution gets compromised, and some parameters are altered, resulting in the system ingesting false data, such as information pointing to the CEO of the organization. This could serve as an effective distraction while intellectual property is being stolen, and the organization might not notice until it’s too late. How could you mitigate this risk? In addition to relying on the partner or manufacturer of the solution, here are a few steps to consider: 

– Ensure data integrity through regular validation and monitoring 

– Restrict access to critical parameters within the platform 

– Implement secure access controls to the platform 

– Limit wherever possible what the AI enabled solution can access/change.  

Remember, when it comes to security, we all play a role in safeguarding our solutions. It’s essential not to solely rely on the manufacturer or partner to secure the environment.

Returning to the question posed in the title: Are AI-enabled networks enough? In my opinion, not quite yet. They still require human intervention. However, implementing AI-enabled solutions can pave the way for future autonomous networks and help address current network issues. To illustrate this, think of an AI-enabled network as a Tier 1 support agent. As the agent gains experience and knowledge, they progress to higher roles in their career. I believe the same is true for AI. Just as a Tier 1 support agent isn’t perfect, neither is AI, but with time and data, it will improve. As our understanding of data and analytics advances, AI-enabled solutions will continue to evolve and become more powerful, leading to more transformational events on networks. For example, imagine a user needing help with a application issue. They could open an AI-enabled network chatbot, describe the issue in plain language, and have it resolved in real-time. Could this happen today? Possibly. Will it happen in the future? Absolutely. 

About the Author:
Chris Price Headshot

Chris Price is an experienced executive deeply committed to nurturing and empowering team members to realize their fullest potential. My passion lies in technology thought leadership, and my career has been dedicated to providing guidance and leadership in aligning technology with business objectives. In recent years, we’ve observed a significant evolution in technology, particularly in digital solutions, which have the potential to differentiate businesses and confer a competitive advantage in their respective industries. In this new era of digital business, organizations must embrace transformation. Within my team, we possess the expertise to guide organizations through the disruptions brought by digital innovations, offering innovative ideas and state-of-the-art technology to navigate these changes effectively.