What does IT Ops have to do with self-driving cars and steering wheels? A lot, actually. A Business Insider article on the subject discusses the human factor in autonomous cars:
“…car makers implementing these advanced driver assistance systems (ADAS) have always maintained that drivers are still required to monitor the driving situation and keep their hands on the wheel. … [Self-driving cars that lack steering wheels] will only be able to operate in limited geofenced areas within an urban area”.
Which brings me to Machine Learning and the human factor in general.
Basically, Machine Learning by itself is dumb, whether its in cars or in ITOps. It needs context and intent to be beneficial. In IT Ops, that means that it requires business context to work properly.
For example, if you have a host naming convention with the site code embedded somewhere in the hostname, ML might be able to recognize codes like ATL and NYC as Atlanta and New York City, but without humans providing the architectural context that these are your two primary data centers, it has no business context and does not know how to use this information.
In other words, simply pointing ML at a data set is not going to discover your business architecture…which you need in order to get the most out of your ML-powered IT Ops tools.
The reason I feel strongly about this is because I’m a TOGAF (The Open Group Architecture Framework) certified architect, and in TOGAF, you start with the business architecture and then dive down into the data, application, and technical architectures, all while making choices based on what serves the business architecture. This adds and applies much-needed context and intent, strategy and logistics to technical design and operations.
What’s my point?
“Black box” or non-transparent machine learning is, and always will be, very limited in its capabilities. To really be able to implement ML – you need Open Box Machine Learning, which takes into consideration your human input.
If you’re implementing microservices in a cloud provider for example, and have an elastic infrastructure with random hostnames and containers with extremely short lifetimes, neither machine learning nor humans alone will be able to manage this dynamic environment. Only a combination of the two can allow the implementation of a distributed discovery and a contextual accompanying KV store that are both needed to in order to automate.
To summarize: If you want to use Machine Learning without constraints, you’re going to need a steering wheel. That is, you’re going to have to apply your business and tribal knowledge to the ML system. Open-Box Machine Learning is that steering wheel, giving you the visibility and control needed to automate your IT operations scalably and cost-effectively, so you and your IT Ops team can enjoy all the benefits of Machine Learning.