The training phase of artificial intelligence (AI). Machine learning systems "learn" about a subject by being fed a huge amount of data samples, which may be identified and labeled or not (see
supervised learning).
The machine learning software, which is implemented using a "neural network" architecture, keeps building and modifying its own data relationships in the training stages in order to improve the data recognition capability of the resulting AI system used in production. "Deep learning" is an advanced form of machine learning, which uses many layers of recognition. See
deep learning and
neural network.
Pattern Recognition and Virtual Assistants
Machine learning (ML) is used to develop pattern recognition systems (face, handwriting, voice, etc.) in many areas, including search engines, medical diagnosis, ad serving, spam filtering and sales forecasting. Today's virtual assistants are the result of both machine learning and "handcrafting," the latter providing predefined frameworks for responses.
Not Like Regular Programming
The final algorithm from the machine learning phase is a lot more difficult, if not downright impossible, to flow chart and debug compared to the routine "if this-do that" logic in regular data processing applications. As more samples become available and more fine tuning is applied, the resulting recognition system becomes more accurate. See
AI and
computer vision.
The Hierarchy
Machine learning (ML) is a subset of AI, and deep learning is a more elaborate form of ML.