Sergey - stock.adobe.com
Published: 19 Apr 2018
There's no shortage of opinion on how AI will shape the future. Prominent scientists and Silicon Valley entrepreneurs can't seem to agree on whether AI will ultimately solve all of humanity's problems or simply destroy us. Whatever AI has in store, it's already changing the way we work.
Over the last year, I've seen a surge in the number of IT vendors touting intelligent, AI-powered foftware products called AIOps tools. These products run the gamut from smart servers to network intrusion detection, but some also leave me scratching my head. I can't help but wonder whether the intelligence in some of these products represents a new interpretation of what AI means or an exaggeration applied to automation -- yet another buzzword corrupted for marketing purposes.
Cynicism aside, there are real and powerful examples of AI creeping into all levels of the IT stack. Development shops deploy armies of intelligent bots to help test software. Cloud management vendors, including Densify and YotaScale, apply AI algorithms to identify inefficiencies. Systems and app performance management tools from BMC, New Relic and others use AI to correlate events and head off mistakes before they become problems. Google used DeepMind to improve its data center efficiency, cutting cooling costs by 40%. And, of course, the emergence of cloud-hosted AI developer tools and powerful new instance types means organizations can begin to experiment with building their own apps and AIOps tools.
The advantages are obvious. Bots click through regression tests and help identify problems much faster than humans. Algorithms recognize infrastructure or cloud use patterns that aren't easy for people to identify. It doesn't have to stop there. AI bots now write code. They communicate with other bots. With proper integration and tooling, it's not hard to imagine an IT stack that essentially drives itself, from code tests to production deployment and infrastructure performance tuning.
This might be science fiction right now, but many of the capabilities already exist, and the missing pieces are within reach. That doesn't mean IT will be a hands-off job anytime soon.
AI and machine learning still require massive, often expensive computing power. Will you be able to justify dedicated hardware resources, cloud instances or (potentially expensive) SaaS subscriptions -- all in the name of working more efficiently? APIs might help you string the pieces together, but, barring some major vendor consolidation, shops that want to integrate AI across the entire stack would for the foreseeable future need to build it themselves.
But perhaps the largest challenge will be overcoming our fear of AI. Yes, fear.
The most publically visible application of AI has been the self-driving car -- a challenge that requires thousands of sensors to interpret and act on real-time data. And the stakes are high. In March, a self-driving car struck and killed a pedestrian in Arizona. This first fatality led Uber to halt its test program, and it renewed questions about the safety of autonomous vehicles. Right-sizing a cloud instance seems almost simplistic by comparison -- yet many IT pros would sooner buy an autonomous car than allow a bot to drive their cloud.
I've heard a lot lately from IT vendors with AI hooks who say customer demand is surging. And the data seems to back them up. In TechTarget's 2018 IT Priorities Survey, nearly as many people said they would adopt some form of AI (13.7%) as would embrace DevOps (14.5%).
However, there's a capability absent in many of these AIOps tools: autonomy to act on their analysis. Most of today's tools aim to provide teams with actionable insight rather than the tools acting themselves, and companies say customers aren't looking to give up this oversight. In some ways, this is a natural first step. You wouldn't cede authority to a tool or service with which you had no experience. And much of today's AI analysis is a black box of algorithms that customers don't expect to understand.
Yet, if we want to squeeze the most out of AIOps tools -- and out of IT -- we'll need to let go of the wheel. Under the right circumstances, machines can simply move faster without us. Today's software delivery trends emphasize speed and continuous feedback but face predictable bottlenecks: testing and provisioning.
Why wait for a human to select the appropriate instance size when an AI tool already analyzed what would maximize performance and minimize cost? We already know -- and the proliferation of third-party optimization tools should prove -- that humans are inherently bad at this type of decision. Sure, AIOps tools could make a mistake or fail to account for some unseen anomaly, but it's likely a human would make similar costly mistakes. Unlike autonomous vehicles, the stakes for letting AI drive IT aren't life and death.
Over the past several years, companies have willingly surrendered control over portions of their IT stack, and the trend is only accelerating as organizations consider serverless and SaaS deployments. Barring another AI winter, companies may soon find themselves weighing the benefits and risks of embracing this next evolution of abstraction by algorithm. Will you let your IT stack drive itself?