You Don't Need AI, You Need an Algorithm

A few weeks ago, I participated in a working group tasked with identifying areas where artificial intelligence and machine learning could improve network utilization. What, specifically, “improve” and “utilization” meant were some of the first questions we addressed. During that conversation, though, one of the participants made an insightful observation that the group promptly ignored. He told a brief story about how someone had asked him for a networking device that could aggregate multiple uplinks and then “use AI to choose the best one.” When he offered a simple WAN failover solution, a technology that has existed for decades and comes built-in to open source firewalls like pfSense, commercial appliances from the likes of Cisco, and every modern mobile phone, the customer told him, “No.” Like our working group, that customer person had a solution (artificial intelligence) driving their requirements (an AI-based WAN failover), not the other way around.

The question my working group grappled with was not, “How can we improve X?” — a question that presumes X needed improving, which in our case it did not — but rather, “How can AI and ML do the improving?” As the world continues to wring its hands about “explainable” and “transparent” AI, it’s worth considering the possibility that in many cases — like in the WAN failover scenario — the solution might be driving the requirements; in many cases, what we actually need is something much less complex, something that looks a lot less like AI and a lot more like a basic algorithm.