Building Trust in Machine Learning with AIOps

Dev Nag is the Senior Director, Office of the CTO, for VMware.

It takes time to build trust – whether it be with people, brands, or technologies. In the early days of email, for example, people would often call an intended recipient to ensure that the message was actually delivered because they didn’t completely trust the technology. But over time, and with familiarity, doubts about email faded away.

Today, organizations are being asked to trust another technology – machine learning (ML) – to do data-crunching at a scale that that humans simply cannot do, whether it’s gathering, organizing or analyzing data, or even taking action.

However, for the machine to learn, it needs a human to help guide it. This includes defining the organizational priorities and acceptable parameters so that the ML process can produce results that matter to the organization. That’s where AIOps – or Artificial Intelligence for Operations – comes in. AIOPs uses ML to provide an abstraction of the operational system in a way that’s simplified for human engagement. It also provides the interface that allows human and machine to interact at larger scales and with more context than ever before.

What that abstraction looks like is up to the human. While the data is meant to be presented in a way that’s both intuitive and relevant to the user, it has to be flexible enough for humans to prioritize the data to meet business needs. AIOps can help to solve many problems, but it can’t define which problems should be solved. That human touch goes a long way in building trust in AIOps technologies for all stakeholders.

Building Internal Trust

From IT to engineering to executives, there are a number of internal stakeholders whose trust is needed for AIOps to be most effective and successful in an organization. The machine has to be able to do more than compute over large datasets. It has to be able to manipulate the data in a way that adds value to the business, through previously unreachable insights or real-time operational adjustments that deliver big savings while maintaining or improving service levels.

Defining those parameters for the data is essential in the early phases of AIOps adoption, and it requires a human to set them. AIOps uses suggestion phasing, where the machine – in the midst of analyzing data – presents a suggestion on what it might do with the data, based on what it’s learned. Rather than taking action directly, AIOps prompts the user with a suggestion and waits for acceptance or rejection. The user can then decide to move to ‘autopilot’ mode when they feel like they trust the recommendations.

With suggestions that might be rejected, AIOps learns to correct and refine processes to adapt to customers’ preferences, essentially customizing the AIOps interface. And because suggestion phasing is an ongoing process, it builds trust at a gradual pace, in a human-centric way that reflects traditional trust-building exercises.

Building External Trust

Suggestion phasing does more than build internal trust. It aims to build trust among external stakeholders, as well, including the increasingly watchful eye of government regulators around the globe. How data is handled – or has been handled in the past – is increasingly a concern across a spectrum of people, from consumers to companies to regulators. As AIOps enables the machine to learn through accepted or rejected suggestions, it’s also mapping the decision-making process that should leave no doubt about how the data is being handled or how decisions were made.

AIOps captures the models and the data used to make decisions along the way, allowing for the auditability that businesses need to meet requirements such as Service and Organization Controls 2, or SOC2, and General Data Protection Regulation, or GDPR. Likewise, AIOps works with policy engines at multiple levels. For example, some geographies may place restrictions on acceptable encryption algorithms for data, or boundaries on the sizing of certain clusters. Not only does AIOps need to recognize those restrictions, it also needs to learn how to optimize within those complex – and dynamic – constraints.

AIOps Evolves with Business Needs

As the tool that links the human and the machine – in some ways, the teacher and the student – AIOps uses the combination of machine learning and human decision-making to provide a more adaptable, customized service.

Over time, AIOps learns both sides of the equation – how the application and infrastructure react to operational interventions, and how the human operators define performance. As such, AIOps directly represents an abstraction from the actions on a system to the goals of the operators, from the "how" to the "why."

This means that humans will always play a role in AIOps, ensuring that the ML process continues to be aligned with the organization’s goals. When those goals shift, the AIOps teams need to be in the loop, ready to guide the AIOps toward new objectives.

The right AIOps solution provider knows the complexities of customer adoption and how to respond to them, and the balance between automation and manual oversight. AIOps represents the best of both worlds – ML’s ability to handle enormous datasets in real-time, and the deep human understanding of both the behavior of their systems as well as the ultimate business goals of their organization.

Opinions expressed in the article above do not necessarily reflect the opinions of Data Center Knowledge and Informa.

Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating.