Request a Demo
Once relegated to futuristic sci-fi movies about killer machines, machine learning is now permeating all aspects of our everyday lives, from optimizing Google searches to Netflix recommendations to the Nest learning thermostat. In the smart building space, machine learning is contributing to improvements in every facet of building mechanics and the occupant experience. It doesn’t take a Ph.D. to utilize the functions of machine learning, but since we have a few on staff, we thought we'd share some basics.
1. Machine learning is a subset of artificial intelligence that automates data mining.
Machine learning, artificial intelligence (AI), data mining—what’s the difference? These three terms are sometimes used interchangeably and experts can easily spend hours debating over where to draw the line between machine learning and AI. (Believe us. We've sat through some of those conversations.)
AI is loosely defined as machines thinking like humans. The human brain is an amazing computing machine. At any given minute, we are capturing tens of thousands of data points from our five senses, recalling memories from past experiences, drawing conclusions between cause and effect, and making informed decisions. We quickly learn to recognize patterns, but even geniuses have their limits.
You can think of machine learning as a more automated and continuous version of data mining. Data mining can often detect patterns in data sets that no human would be able to find. Machine learning is capable of generalizing information from large and dynamically changing data sets, and then detecting and extrapolating patterns in order to apply that information to new solutions and actions.
In the smart building space, machine learning enables a building to run as efficiently as possible while responding intuitively to occupants' changing needs. For example, consider the difference between how a scheduling program in a smart building might handle a recurring board meeting and how a machine learning application could do even more. A simple scheduling system could be programed to adjust the temperature in the conference room to 72 degrees on the specified day of the board meeting in advance. However, machine learning algorithms can make sense of thousands of variables, like time of year, daylight glare, indoor temperature, indoor air quality, lighting controls, audio-visual needs, occupancy in a conference room, and historical readings of personal preferences, to instantly create the ideal lighting and thermal environment for a presentation or conference call.
2. Machine learning still requires humans.
Machine learning's ability to automate, anticipate, and evolve is powerful, but that doesn't mean computers will take over the world. Machine learning still requires human operators to provide context, to set parameters of operation, and to continue to improve the algorithms.
Machine learning discovers patterns that are inhumanly possible to see, and makes subsequent adjustments throughout the system. However, it’s not always good at finding out why those patterns exist. For example, a lot of smart building functions are created with the intent to make the people inside more productive. But we can’t simply tell the building to “make people productive.” A human must set up those definitions and rules for the building to follow.
On the flip side, data alone cannot explain when outliers or anomalies occur. For example, an algorithm can take note and respond when people in a specific work area continually request it to be ten degrees warmer than anywhere else in a building. However, the algorithm would not be able to tell the building operator that this is happening because of a broken thermostat. Machine learning helped the building operator uncover this issue, but at the end of the day, skilled people are still needed.
3. There are two different kinds of machine learning.
There is supervised and unsupervised machine learning. Smart buildings will likely incorporate both. Here's a very simple example of what supervised vs. unsupervised machine learning looks like: Suppose you want to teach a computer to recognize an elephant. With a supervised approach, you could tell the computer that an elephant is "a large mammal with a prehensile trunk, long curved ivory tusks, and large ears, native to Africa and southern Asia." With an unsupervised approach, you might show the computer a group of different animals and tell the computer which is an elephant, then another group of different animals and ask the computer which is an elephant, and repeatedly show multiple groups and correct the computer, until it learns that specific features (like a trunk, big ears, and ivory tusks) define an elephant.
In the smart buildings space, both supervised and unsupervised machine learning are used. As the cost of building sensors drops and more occupant-facing apps are developed, occupants are able to provide more feedback and continuously "correct" the smart building as it learns how to create optimal conditions.
4. We are currently experiencing a renaissance in machine learning.
During the 1980s, when rapid advancements in computers and computing power emerged, there was an enormous amount of excitement (and fear) around the potential of computers, artificial intelligence, and machine learning to solve all of the world's ailments—freedom from diseases to freedom from household drudgery. As AI and machine learning began to develop as formal fields of studies, actually turning these hopes and ideas into reality were much more difficult to achieve and AI retreated into the world of fantasy and theory. However, in the last decade, advances in computing and data storage have changed the game yet again. Tasks that were considered "too difficult" for machines to learn, like advanced voice recognition and language cognition, are becoming a reality.
5. You don’t need to have a Ph.D. to benefit from machine learning.
Writing machine learning algorithms is quite different from learning to use them. After all, you don’t need to know how to program an app to use your iPhone. The best platforms abstract away the obscure to present business users with interfaces that require minimal training. If you know your use case and the basic concepts of machine learning, you are ready to go. The technical expertise to fine tune which algorithms make the most sense for a particular use case is left to data scientists. Users don’t need to know the math, just their business domain.
Machine learning is here, and it's growing quickly. Buildings are already using machine learning in a variety of ways to make the existing infrastructure function better, as well as to enhance the experience of the occupants inside. From an energy usage standpoint, buildings are learning occupants’ needs and optimizing usage of heating, cooling, and other mechanics—intricately negotiating human desire with overall system efficiency.
But, what will this mean for us moving forward? Inevitably, it will mean most things will happen without us having to ask for them. Going beyond temperature and lighting, machine learning could prepare rooms for meetings—adjusting screens, shades, conference calls—or signal elevators and shuttles. Machine learning means a future state of many levels and layers of automation, all adjusting the environment based on your current activity.
