Since just last fall, we’ve been hearing about machine learning and artificial intelligence from all corners of our industry. Examples that Gabe and I covered include Citrix, ExtraHop, Lakeside, Centrify, Egnyte, and Microsoft EMS. There was a lot of talk about it at conferences like RSA and the Cloud Identity Summit; and it’s a main focus of cloud compute platforms like Google and Azure. In other words, if you have a pulse and follow any type of technology, you’ve heard the machine learning buzz and know it’s everywhere.
Gartner said last August that machine learning was at the peak of the hype cycle, but clearly it’s not there yet and it’s still going up. However, we know that the trough of disillusionment is coming eventually. So how much should we worry about the (possibly overinflated) hype? And how should we think about machine learning from an IT perspective?
I know I’ve been thinking about these questions a lot. I wrote a few thoughts about machine learning at the beginning of the year; six months later, I think some key points are becoming more clear. At the risk of repeating a few things I already wrote, here’s what I’m thinking about machine learning and artificial intelligence right now:
First, after more time parsing all these new announcements, it’s clear that at this point, they’re about business logic. There are no new application package formats, OS platforms, or devices coming out of machine learning right now, so we’re not going to have the type of user-caused headaches that we had with BYOD, consumerization, FUIT, and mobility. (As I’ve long said, consumerization was a one time switch over, and many of the ill effects have been attenuated by the advancement of corresponding applications and IT tools.)
Machine learning is naturally coming to IT management and security products, but again, these aren’t being pushed on us by users—instead, they’ll be carefully vetted and controlled, just like any other upgrade cycle. We’re assuming any products that use machine learning will fail safely and have auditing modes. For example, if a product users machine learning to automatically configure user access permissions for various resources, there should be a step for human review. Or if we’re using machine learning to do user behavior-based authentication and access control it should be configurable to be more conservative and fail safely by throwing up a multi factor authentication challenge. If a product doesn’t have these fail safe mechanisms, it should be discovered during the vetting process—or natural market forces should hopefully keep unsafe products from getting released anyway.
Another thing to keep in mind is that some of these products do their “learning” based on your own data set or users. They might not work well (or just be okay-ish) during an initial trial, but then get better over time once they’re at productions scale. Or vice versa, issues may be not arise until they’re at production scale.
Finally, in some ways, we already have plenty of apps with business logic that (speaking at a high level) takes in data and then tries to use it to make seemingly more intelligent decisions. The underlying algorithms and technology may be changing, but it’s the same forward progress. The interesting thing will be to look back in several years and see in context how much the machine learning wave of 2017 sticks out as a turning point that enabled new applications.
Getting back to the hype—is there too much hype about machine learning.? Maybe. But I’m not bothered by it, for the reasons outlined here. For what it’s worth, from the IT software perspective, I’m still hearing almost nothing but excitement (or hype, if you prefer that term). We’ll get to the trough of disillusionment at some point, but then, to continue on the hype cycle model, the rise to maturity will follow. At that point, machine learning will be something that’s “just there.”
With all this hype about machine learning and artificial intelligence, I wanted to list a few more places for further reading. Interestingly, these all come from Andreessen Horowitz:
- Here’s their enterprise-oriented Artificial Intelligence Playbook and intro blog post.
- Computer vision—the idea that we’ll be able to get as much data out of an image as we can out of text—is one of the most fascinating developments. Check out this post on the topic from Benedict Evans.
- Here’s a primer on AI, deep learning, and machine learning.