Unsupervised learning is an ML approach applied when no labeled data is required or available, meaning there is no way to internally validate the results produced by the model. The goal of unsupervised tasks is exploratory: to uncover patterns in the data, find natural groupings based on similarity, or unveil the underlying structure of the data.

The primary unsupervised tasks discussed include Clustering (grouping data points by similarity, such as topic modeling) and Dimensionality Reduction (simplifying or transforming data while retaining core information). Content transformation, such as writing meta descriptions, is also considered an unsupervised task characteristic because validation is not straightforward.