Is your network AI as smart as you think?

Network operators told me that in the future, artificial intelligence will manage their networks. They also told me that the supplier had communicated the same message to them. The good news is that "some aspects" are true. The bad news is the same.

Network operators told me that in the future, artificial intelligence will manage their networks. They also told me that the supplier had communicated the same message to them. The good news is that "some aspects" are true. The bad news is the same. The focus here is on "some aspects." In order to get the most benefit from artificial intelligence network management, you must figure out the vague "some aspects". For this, you can realize this by imagining "ants and farmers".

 

Ants can build extremely complex anthills with various interconnected tunnels and checkpoints. Do these worker ants have some powerful ant engineers to guide the completion of this process? The answer is no! They each perform their own simple tasks wholeheartedly and complete tasks instinctively. In fact, there is an "ant engineer", but it is their own DNA that organizes their work to achieve their goals. This is a bit like the way most network AIs work.

The network is composed of a bunch of technology "collections", each of which is a bit like an anthill. There are collections based on supplier, device type, physical location, and connection relationship. If you look closely at today's network AI, you will find that it mainly runs on "collections." Maybe it manages Wi-Fi or it may manage edge elements such as SD-WAN or SASE. AI applications used to manage a "collection" will incorporate management goals into their DNA (that is, their design). Simply put, if we are a Wi-Fi supplier, we know how Wi-Fi works and incorporate this knowledge into our AI management.

The challenge comes when we no longer think of collections as independent elements and start to think of networks as "collections of collections". The network is not an anthill, but the entire ecosystem where the anthill is located, including trees, cows, and many other things. The tree knows how to become a tree, and the cow understands the habits of the cow, but who understands the entire ecosystem? A farm is a farm, not any combination of trees, cows and anthills. The person who understands what a farm should be is the farmer, not the elements of the farm or the suppliers of these elements, in your network, dear network operators, the farmer is you!

In the early stages, the developers of artificial intelligence clearly admitted that the knowledge engineers who build the AI ​​framework are separate from the subject matter experts (whose knowledge shapes the framework). In software, especially DevOps, management tools are designed to achieve a target state, that is, in our farm analogy, it describes the location of cows, trees, and ants. If the current state is not the target state, they will do something or move something to move closer to the target state. This is a great concept, but to make it work, we must know what the goal is. At the corporate network level, we need our Wi-Fi experts to subconsciously introduce knowledge into Wi-Fi AI management tools. If AI vendors don’t know how this knowledge was acquired, their AI will be useless.

If your hopes for AI are dashed, it is still too early to say! Many network operators are still very satisfied with the AI ​​that manages the technology collection that makes up their network. After all, when nothing happens to one set can be remedied by adjusting the other, why worry about coordinating Wi-Fi and SD-WAN management? If this set-AI model can meet your needs, you're done NS.

To understand whether it is feasible to be an "ant" (at least in terms of network AI), the best way is to ask whether your technology collection is really "atomic"—that is, completely independent and self-contained. This boils down to the visibility and control range of your AI. Basically, the set-specific AIs are all independent. Ideally, you need your AI to gather ants to do their own things instead of intervening in each other's activities. You don't want the AI ​​in one place to view another set without coordination and react to conditions, or the two AI set processes deal with the same problem at the same time.

If the remedy for a problem in one set might involve doing something to another set, then you need your AI to be able to cover the combination. Therefore, if you feel that the network operations center used to manage the ecosystem is too expensive and overloaded, and want to deploy AI so that everyone can relax, then you need to understand the supplier's AI proposition more deeply.

This is not easy for companies, because of the people I interviewed this year, more than three-quarters said that they do not have much (if any) artificial intelligence expertise within them. Many people feel that they are at the mercy of suppliers because they promise to provide great things, but they don't seem to fully meet expectations. Is there nothing that companies can do about this phenomenon?

The easiest way to master the use of AI in the entire network ecosystem is to find a "manager of managers, MoM" (manager of managers, MoM, which means that the fund manager does not directly manage the investment of the fund, but rather Entrust fund assets to some other fund managers for management) methods. In modern terms, you can call this "intent modeling". If each of your technology collections can be regarded as a "black box", its behavior is modeled according to its own SLA, and its AI process can execute the SLA, then all you need is to make these collections of AI tools Each of them generates a fault report to a higher level package. Then, if there is a problem beyond a single set of technologies, or if one set fails and a higher-level fix must be considered, then the package can decide what to do.

The challenge here is to find the target state and how to recover when something goes wrong. Remember the "subject experts" and "knowledge engineers" mentioned above? It is difficult for us to build AI solutions for networks because all networks are slightly different, and only users themselves know what they think is "good" or "bad" . Some AI tools may provide machine learning (ML) capabilities to let your network operations center (NOC) personnel understand the situation and know what to do; others may use what the network provider knows usually represents normal options and common remedies Baseline.

However, both of these methods have some problems. Machine learning may take time, and when your AI system is performing its mission, it will further consume your NOC resources. When your network is mainly composed of equipment from one supplier, the supplier baseline is most effective. Both can be adjusted, but both may conflict with adaptive network behavior.

IP networks essentially use topology to discover and do their own thing. Even for NOCs, it is difficult to influence routing; they usually need to plan new MPLS routes for traffic engineering, and AI is unlikely to do this. Some companies (including Google) have turned to software-defined networking (SDN) to provide central control of routing, and AI can then control the network by controlling the SDN controller.

AI in network operations can be traced back to a combination of events to send out abnormal signals, and it can also be used as a way to implement effective responses. At any level, your potential AI supplier should be able to discuss with you how their products collect information and how to realize their insights. Delve into the details of these two things, because no matter how magical AI claims, it won't work without these two ingredients. Finally, please remember: to be a farmer, not an ant, your position will never be lost due to the emergence and development of AI.