My Animosity Toward Artificial Intelligence

In my previous posts, I talked about the basics of AI in relation to network and systems management as well as why I love it. This post isn’t going to be as optimistic about the current state of affairs when it comes to AI. It isn’t the tech that bothers me. So what is my big problem with AI? The answer is simple and comes down to basically two things: marketing and money. There are a lot of products touting the latest and greatest in artificial intelligence. They all claim to be the only thing that can save your infrastructure from the impending doom of manual work. The literature, webinars, and sales calls all start off with the horrors of running a network without AI and how inefficient it is. There is no shortage of FUD (fear, uncertainty, doubt) on the topic, all spewed in the name of selling you the latest and greatest cloud-based widget that’s smarter and faster than you are. “How can you possibly, responsibly, safely run a network without our algorithms looking over your shoulder?” They show off demos of their system finding impossible correlations in a veritable ocean of data and maybe even show the system reacting automatically while telling nice stories of administrators and engineers sleeping soundly while the mighty AI saves the day. The biggest question to ask there is “Where did the data come from?” Was it created specifically for the demo? Was it taken from a customer’s live network? Is that network identical to yours? The training that goes into building a reasonably sophisticated AI is intense. Especially when you’re talking about monitoring dozens, if not hundreds, of applications across a massive network. Does it understand your needs? Does it adapt? Does it dream of electric sheep? Marketing aside, the topic of money is usually saved until the target (you) has decided they can’t live another minute without their product. Let’s assume for a minute that you’ve found a company that is building an AI suited perfectly for your needs. It ingests massive amounts of data from all over your network into a box (or boxes) that crunch all that info actionable results for you. It spits out pretty reports and integrates with all your systems. Where does that box (or boxes) live? On-prem? That going to take up a lot of rack space, power, and cooling. Plus those machines aren’t going to be your run-of-the-mill 1U servers. They will be power hungry, face-melting, GPU-packed behemoths. Cloud-based? Now you’ve just moved those same beasts to someone else's data center. Granted, the cost will decrease a bit because the vendor is likely counting on scale and flexibility to process multiple customers on the same hardware just as effectively. But add to that the bandwidth requirements of sending all your application logs, all your network logs, and anything else you can through at it up to the cloud, constantly. There are a ton of costs that go into running a system that can think like a person, and licensing is only part of it. The marketing is strong when it comes to these products and it can be a bit misleading. AI is not going to save you from something that you couldn’t do yourself. AI isn’t going to magically fix all your user issues. AI is definitely not going to put your network and systems on autopilot. AI is going to be expensive and time-consuming to set up properly. However, AI can save you money in the long run by becoming a force multiplier for your IT staff. AI can add in efficiencies that you weren’t able to even come close to before, and maybe even start making some money for your company (depending on your industry) by aligning your data better and creating new avenues to generate income from the systems already in place.