The importance of Machine Learning is actually quite easy to grasp. Overall, it is mainly used by businesses to identify key insights in data that it would be too difficult or too time-consuming for human experts to find on their own. Past this initial understanding of ML however, everything typically feels much more convoluted. For many, the confusion starts with the split of the Machine Learning field into “Supervised” and “Unsupervised” methods of learning.
In past articles, we have attempted to illustrate the differences between these two fields to you, with the hope that current and future work in the Machine Learning space may become easier to understand. In the interest of being as clear as possible, we left out many of the technical terms that pervade the ML niche, which may have caused some of you to misunderstand the primary difference between Unsupervised and Supervised learning frameworks.
In short, we explained this as if in Unsupervised Learning, AI systems do not start with any past data to structure their learning as they go live. Truthfully, they do, but it is not labelled. In Supervised Learning, human engineers typically tell a system how to understand certain data sets before it goes live. To use a popular analogy in the space, imagine that these workers show an AI what a cat is, instead of requiring the AI to learn that fact on its’ own. With these ideas in mind, we can then say that if an AI is structured based on Unsupervised Learning principles, then it essentially is meant to learn on its’ own.
Circling back to our point about what the relationship is between Unsupervised Learning and what is called “training data,” this means that such an AI would receive a large amount of data as it goes live that it essentially does not understand at all. Considering this, it would be logical to ask: why would any AI team want to do this? Furthermore, wouldn’t such a framework be needlessly time consuming?
To truly understand the answers to these questions, it’s important to understand the benefits of using Unsupervised Learning methods, through the eyes of an ML professional. In our next piece, we’ll detail some of those for you, in connection with the risks that come with opting for unlabelled data. For now, keep in mind that the general consensus appears to be that Unsupervised Learning might actually lead us to the Age of Artificial General Intelligence, due to its’ ability to teach an AI to be a data scientist in a basic sense.
