"Machine Learning" — Gives "computers the ability to learn without being explicitly programmed." — Arthur Samuel
Two popular approaches to ML development involve supervised and unsupervised learning.
Supervised learning works like our education system: there's a teacher "supervising" the learning process. The algorithm is asked a question and makes a prediction of the answer based on the labeled training data it has been exposed to. It is "corrected" by the teacher when the prediction is wrong, and learns iteratively from its mistakes. When the algorithm's predictions achieve an acceptable level of performance, it graduates with a diploma (although it may continue to need supervision).
Unsupervised learning on the other hand has no correct answers and no teacher. Algorithms are simply fed unlabeled data and left to structure the data in some new, interesting way.
Unsupervised learning garners enthusiasm because of its seemingly endless possibilities for creativity, but concrete business applications for this approach remain limited.
Supervised learning has proven a better ROI, since it can answer specific prescriptive questions, like "has that roof suffered water damage?" This approach does, however, require more investment upfront, because the algorithm usually requires large amounts of labeled example data to make sense of the world it will operate in.
For a deeper dive into supervised vs unsupervised learning , check out our white paper.
And if you want the abbreviated version, download the Executive Summary.