What is Machine Learning?

This guide often uses the terms AI and ML interchangeably. The overall taxonomy of the AI/ML space can be confusing, and definitions vary based on points of view. However, AI is generally considered to be a broad topic with several sub-fields, methodologies, and practitioners.

One of the key distinctions is the difference between general and specialty AI (Figure 1). General AI – or artificial general intelligence (AGI) – involves the idea that computer programs can exhibit broad intelligence and reason across domains like humans. Most researchers believe that true AGI may not be achievable in the near future.

Specialty AI involves the idea that computer programs can be trained to reason and solve specific dedicated tasks. Examples of these tasks include detecting certain images, classifying potential failures for equipment types, or detecting certain types of fraud. This field of specialty AI has been advancing rapidly in the last couple of decades.

There are different sub-fields within specialty AI, including ML, optimization, and logic. ML describes a class of algorithms which leverages powerful statistical learning techniques that operate on data. The power of machine learning is that the algorithms can be quite generic; just a few algorithms can address and solve many problems. Additionally, ML algorithms can learn from data, and can therefore, as described in the cat identification problem above, reduce the need for complex logic and code.

ML has proven its ability to unlock economic value by solving real-world problems, including enabling useful search results, providing personalized recommendations, filtering spam, and identifying fraud.

There are three main subcategories of ML techniques – supervised learning, unsupervised learning, and reinforcement learning. Within each category there are the “traditional” ML algorithms, and also newer, deep learning algorithms (that are described further later in this article). The following figure summarizes the most common machine learning categories and approaches.