What Labeling Does for Machine Learning/AI

"To connect the dots, you must first label them" - Sun Tsu (probably)

In a world where new AI breakthroughs are in headlines every day, it's important to remember that these new advancements are all built on the same foundation - points of data and the relationships between them. Sometimes the relationships are direct and obvious like a father and son, other times they're events in time where cause leads to effect. Without going too deep into math, an AI model at its core is simply a set of parameters that define the nature of these relationships and how strongly they're linked.

One part of machine learning is collecting high quality data, but the actual hard part is ensuring they're labeled in a way that can encapsulate the context. Human intelligence works the same way. For example, if instead of father and son two people were introduced as John and John Jr. you would inherently assume they were father and son, because in English speaking countries Jr. is a label with a known context.

Another example would be an email with a Word file attached that is called, for example, final_v4_final_presentation_a1_1_send_version(3).docx. That may have been a good-enough file name right before that email was sent, but six months later that file will be long buried in someone’s documents folder and difficult to search for. Luckily, the sent mail folder will come in handy to find that file since the email it was attached to will most likely have additional context for searching.

Identifying and tagging relationships from a building's sensors is never as obvious as those examples, so we built software tools to make it less painful. At Resolute Building Intelligence, our systems collect and label billions of data points each week, not only making it easy to identify and resolve issues when faults occur but also preparing the data for intelligent agents. The hard part is already done so you’re not left searching for answers.

---