Many people talk about Artificial Intelligence and Machine Learning as the Holy Grail for the Internet of Things. Are these people right? They are. They are, except when they’re not. Therein lies the nuance in terms of IoT and the importance of the data.
First, though, let’s step back a few years. In the early days, IoT systems were just a cool new thing. They were mainly smart devices that were connected via the Internet to become “IoT-enabled” devices. As people slowly begin to figure out the importance of data, some of the early players described the “Holy Grail of IoT” as predictive analytics. At that point, the market was primarily still trying to figure out where this was all headed, and predictive maintenance had the attribute of a more substantial return on investment than most other dimensions of IoT. In that sense, it probably was the Holy Grail. On a personal level, I never thought that was the case. Not that predictive analytics is not important. It certainly is and continues to be contributing the strongest ROI for most IoT applications, especially ones in business and industrial settings.
That said, predictive analytics is but one of several components in the analytic stack. Other elements would include operational analytics, investigative analytics, and AI and Machine Learning. Operational analytics are the grids and graphs that we usually associate with business intelligence. Operational analytics answer the question, “what is happening?”. Because instrumented assets provide a much more granular signature, the operational analytics resulting from IoT provides a much more insightful and more precise picture to address the question as to what is happening. Likewise, the role of investigative analytics addresses the question, “why is it happening?”. Investigative analytics is engaged using toolsets that allow for data scientists to munge through data, dicing, and slicing to perform forensic analysis and determine root causes. Predictive analytics is often derived from the investigative analytics that creates these insights that form the basis for the predictive models. Of course, predictive analytics is answering the question, “what will be happening?”. This creates an opportunity for significant savings in the form of reduced downtime, streamlined maintenance, and more productive operations in numerous environments.
But the forward progression of IoT points to truly adaptive systems. It is clear to everyone that the state of a given environment, be it related to energy use, mobility, healthcare, agriculture, or virtually any domain, will change. Some change instantaneously, some change more slowly, but the world around us is continuously changing. The “Holy Grail”, at least in my mind, is the ability for systems to change and adapt without human intervention. If we look forward, the promise of Artificial Intelligence and Machine Learning brings us to that point. The word “autonomous” is not only becoming increasingly associated with vehicles; it is becoming increasingly associated with everything. Well, almost everything. And that opens all kinds of doors.
But there is a significant caveat in this thesis. The best AI and ML models in the world are still a function of the underlying datasets. If the IoT data is a function of a poorly architected ecosystem, where the data is dirty, or undermined by the inclusion of private data without permission, or unenriched or not contextualized, or unsecured, then the dataset is less than optimal. Robust datasets for IoT need to be curated. This means the inconsequential data is filtered out. This means that the data is likely contextualized with other IoT data. But it also means it might be enriched with enterprise data, or third party data to create additional dimensionality that yields greater insights powered by the very models in question. And looking forward, it means that the data curation contemplates security, privacy constraints, and data ownership and governance, all of which ultimately contribute to creating the best possible dataset made available to the right constituent in the right representation at the right time.
But in reality, we probably have a long way to go, in most instances, to get to that point. So is AI and Machine Learning the Holy Grail for IoT? Aspirationally, it is.
