Artificial intelligence (AI) and machine learning (ML) are getting a lot of attention and for good reason. Recent improvements in the speed and cost of data processing have made AI/ML broadly available in a host of business and industrial products and services, and so it seems being powered by AI/ML is showing up in every imaginable context. Throw in terms like deep learning, neural networking, and it gets confusing for those making buying decisions.
What is artificial intelligence?
Artificial intelligence is the AI in AIOps (which stands for artificial intelligence for IT operations). But it is important to know that artificial intelligence and machine learning are not the same thing. All machine learning is artificial intelligence, but not all artificial intelligence is machine learning. That’s because AI is a broad category describing all the non-biological or computer-assisted ways data is analyzed for decision making. Simple AI has been around for decades. It is a fast, accurate, and efficient way of making “either-or” determinations, but it is limited by the data and range of outcomes it is given. It does not recognize nuance, and it can’t handle new information without being re-programmed.
What is machine learning?
Machine learning is a subset of artificial intelligence based in pattern recognition and the ability to autonomously “learn” and enrich its understanding of its environment. ML algorithms are what give artificial intelligence the ability to ingest and adapt to new information on its own. In IT operations, machine learning is what sets AIOps apart from legacy ITOps. Without machine learning, an AIOps platform offers only an incremental improvement over traditional ITOps. With machine learning, however, AIOps takes a quantum leap forward because it continuously adapts to new information to better understand changes in its environment, expanding its capabilities, and making more accurate decisions.
Machine learning automates analytical model building. It uses complex algorithmic methods to find hidden insights in data without being explicitly programmed where to look or what to conclude. And therefore, replaces the human analysis often required to diagnose problems in IT. Where more conventional, deterministic programming is designed to perform repetitive tasks quickly, and with consistent results, machine learning is designed to enrich its understanding of an environment and produce more precise insights. As an ML-based system ingests more data and observes operations over time, it will find optimal solutions to the problems it encounters.
What are the differences?
A good way to understand the differences between these terms is to consider anomaly detection in the context of IT operations. There are many ways to detect anomalies in IT operations, including the use of simplistic forms of artificial intelligence, but they require someone to program the system with predetermined parameters. Because anomalies are so common in IT operations monitoring, if you try to apply these methods at scale to the number of signals generated by today’s modern networks, you’ll quickly overwhelm your staff. The remedy is often to recalibrate to generate fewer signals, but then you may miss critical signs of declining service health or a failure somewhere in the system. Instead of simply saying, “This signal is outside of your prescribed band, look into it,” machine learning will take all relevant data into consideration to understand why a signal is different and determine whether or not action is required. If action is required, it can use that knowledge to inform an efficient course of remediation.
What’s more, ML is the power behind using data to understand the relational context that matters to your unique environment and services. When you have machine learning driving your AIOps, you have the power to understand the relationships between your data and operations, recognize and solve problems, rapidly sort through (not be overwhelmed by) your big data to find correlations, autonomously train your systems to respond to changes and, eventually, develop a rich enough understanding of your environment to achieve true automated IT operations.
With machine learning, IT processes can progress beyond responsive automation and become predictive, forecasting likely events, and initiating the preventive actions necessary to avert a crisis rather than waiting for one to occur. using machine learning and AIOps in conjunction with the accurate intake of all sources of data makes it possible to build trust in your monitoring platform’s ability to know what actions to take based on an event, and reach new conclusions by interpreting your big data in ways that people are not capable of doing. Machine learning makes it possible to discover and generate the kinds of correlations between data and events that lift the overwhelming burden of always being in reaction mode; thereby, enabling innovation.
Without machine learning, an AIOps platform can only do what it is told to do. And without it, you can’t truly automate. The platform may be exceedingly proficient at whatever routine tasks it faces, but it cannot perform tasks it hasn’t been programmed to tackle, and so will fail when confronted by events it doesn’t understand. That’s because conventional programming lacks the ability to recognize and adapt to unfamiliar inputs the way that artificial intelligence can in general, and the way that machine learning does specifically, even if it encounters the same conditions repeatedly.
What do companies use machine learning for?
To solve problems with big data without becoming overwhelmed by big data, your IT operations platform needs machine learning, which can handle the data from every input and turn that data into knowledge on its own. Without ML, until a person identifies an issue, solves the problem, and updates the programming accordingly, the outcome will be the same. Every time. And that is one of its primary benefits. ML is a tool that does not replace human intelligence, but supplements and complements it.
Machine learning frees skilled IT operations staff from inefficient, repetitive drudgery, and from the impossible task of keeping pace with the overwhelming volume and velocity of data in today’s complex IT environments. By solving a class of problems typically assigned to humans but that occur at a crushing scale, machine learning lets those valued people focus their skills on more creative endeavors.
At human scale, we do well, but our capacity has limitations. We are, after all, only human. In IT, where things operate at machine-speed and cloud-scale, pattern recognition and correlation are impossible. That’s because today’s IT environments are large, complex, and fast-changing. The patterns that emerge from the data generated by the wide array of endpoints, applications, cloud instances, servers, third-party inputs, IoT monitors, and more tell a story of health and performance. The patterns also signal risk of trouble ahead—but are easy to miss under traditional ITOps approaches. We need help. We need machine learning. We need AIOps.
What does it mean to “use machine learning”?
Now that you understand what machine learning is (a subset of the much broader category of artificial intelligence that adapts its behavior in response to changing inputs), it is worth stopping for a moment to consider what it means when a business claims to “have ML” or they “use ML.”
Machine learning is not just one type of algorithm or one type of problem. So, without knowing how machine learning is applied and used, it’s like saying a building “uses” wood. Yes, it’s factually true, but it doesn’t tell you anything about how it’s put together. (Is it prefab or stick-built? Can it survive a flood or hurricane or other natural disasters?) Machine learning is a grouping of ingredients that can be combined in a multitude of ways to create a dazzling array of different solutions, in the same way flour, butter, and sugar can be combined to make anything from a light and flaky croissant to a heavy, luxurious black forest cake.
When a company says they “use ML,” you should as the question “How do you use it”? What is it that they have made by combining different components of ML in a specific configuration? Is it a fully finished product that you as a consumer simply turn on? Is it a series of integrated tools that require some input or setup from the user? Or is it just open-source machine learning libraries that have been made accessible within their product, passing the labor of data science and architecting down to you? Each of these examples requires a different level of effort to properly leverage and will scale differently depending upon these important details as to how machine learning is used.
In the next article of this series, we will discuss the how of ScienceLogic’s current applications for machine learning, how those applications are integrated into the rest of SL1, and why it is that regardless of your level of expertise with machine learning, those integrated solutions can be applied with minimal effort at maximum scale.
Ready to learn more about how AIOps can help your business? Read this analyst report »