Confused by Artificial Intelligence and Machine Learning?
That’s hardly surprising, cynical marketing departments have hyped AI and ML in networking spheres for their own purposes and muddied the waters. Well established vendors and startups alike are engaging in the kind of hyperbole designed to get boardroom bigwigs foaming at the mouth for some of that sweet buzzword action.
I’ve found that whilst all of the big names have prominent landing pages for AI in networking, there are few products to reinforce all this chest-beating. The ones that do offer something of substance only fulfill one or two criteria of what I would call AI. It reminds me of the 2017 cryptocurrency ICO madness where a fancy website = $$ yo. Take this web page by Extreme Networks, there’s no product to explore. So networking companies…if the game is indeed changing then give us the products to make us more efficient, more scalable, and less error-prone. Don’t just tell us there’s an awesome shiny thing that we need – coming soon.
What is Machine Learning?
ML is the recognition of patterns in a data-set by a computer that has not been specifically told what to look for. Machine learning algorithms “learn” that when x happens then statistically y should be the outcome given historical occurrences. This can be useful to humans when we have vast data-sets with hundreds or thousands of parameters to trawl through. After all, the brain has its limits.
Machine learning could be helpful to baseline a functioning network. If abnormalities beyond the standard deviation occur then us humans can be notified to further investigate. ML is looking for patterns, but by itself it doesn’t know what the issue is or how to fix it.
What is AI?
AI is a deeper field of computer science than ML alone, and a tricky one to pin down an exact definition for. The best I’ve come up with is the slightly wordy: the ability for a non-biological form to exhibit thought, planning and self-awareness in a mental capacity and the ability to process and spontaneously (ie. not pre-programmed behaviour) manipulate the physical world.
That’s my attempt at distilling what is a complex and constantly evolving scientific field into a sentence. I’m sure there are better. What it means to be an intelligent being is reviewed and revised as we learn more about our brains, and therefore what a machine is required to do in order for it to be classed as artificially so, is also in a state of flux. Machine learning is just one branch of technology that makes up AI. Here are a selection of the others:
- Natural Language Processing (NLP)
- Deep Learning
What makes AI as a marketing strategy in networking so contentious is that really, most are just doing ML… and some aren’t even doing that. I’ve not seen anything that I would consider to be anywhere close to the definition of AI I’ve outlined. Perhaps my definition is misguided, but I’d argue that at this moment in time, “AI” does not exist. We only have disparate elements of it in isolation.
AI in networking
Some companies such as the wireless vendor “Mist” (now part of Juniper) are using some aspects of NLP in addition to ML to combine elements into a more powerful tool. However, having seen this demonstrated it does still feel slightly gimmicky as the product functions perfectly fine without it. I also haven’t heard Mist setting the world alight in bettering the end-user experience when compared with more traditional products. I’ll get round to reviewing this for myself in due course.
As anyone who has wrangled with Siri’s bullshit or has raged at a chatbot knows; we’re still unable to get reliable NLP results without wanting to frustratedly dropkick our iPhone into the nearest active volcano. The point is, there is still a syntax even if it is a looser syntax than say, writing a wireshark display filter. We can’t yet talk naturally to devices without misunderstandings, but the same could be said of off-shore call centres.
That’s a lovely buzzword soup, want some “intent” with that?
Intent is another industry flavour of the month which I keep seeing crop up. Although I’ve not used an intent based system I understand it to be a policy, configuration, and state enforcement engine. We tell the system what we want to achieve, and it figures out the detail to get to that point. The same way that a network engineer has to know the network well enough to understand whether changes impact other areas.
An example of intent could be that the accounting package needs priority over everything else on the network at month-end. We could give this intent in English (an improvement in NLP perhaps), and the system hides all of the complexity. It checks the existing policies for conflicts, creates the configuration to classify packets, and verifies that the networks overall intent (to get data from A to B) has not been impaired by the change.
Terrifying. Will the righteous router overlords take our jobs?
I’m not going to lie, in time it’s possible that they will. Once upon a time, not too long ago people trampled on dyed cloth barefoot in vats of urine to make colours set in the fabric. With that beautiful image, you can see that occupations which existed for centuries can be replaced in short order. With innovations in networking we also may have to adapt; but unlike our ancestors we may come out smelling of roses.
The tedious aspects of networking are the first pieces up for grabs by the AI usurpers. Engineers already frequently do automate bulk configuration updates, and SD-WAN has simplified the edge so it’s reasonably trivial to manage. I hope next-gen networking can bring the confidence that enterprise-wide changes will succeed by having the computer analyse state fabric-wide. We’ll all sleep a lot easier.