The terms "artificial intelligence" and "machine learning" are often used interchangeably, but one is more specific than the other.
Artificial intelligence (AI) is the broader of the two terms. It originated in the 1950s and can be used to describe any application or machine that mimics human intelligence. This includes both simple programs, such as a virtual checkers player, and sophisticated machines, such as self-driving cars. Some in the field distinguish between AI tools that exist today and general artificial intelligence—thinking, autonomous agents—that do not yet exist.
Machine learning describes a subset of artificial intelligence. This term arose in the 1970s. Machine learning is distinguished by a machine or program that is fed and trained on existing data and then is able to find patterns, make predictions, or perform tasks when it encounters data it has never seen before.
Diagram explaining the difference between categories of artificial intelligenceCredit: Lance Hayashida for Caltech Science Exchange
Diagram explaining the difference between categories of artificial intelligence
Credit: Lance Hayashida for Caltech Science Exchange
Machine learning can be thought of as the process of converting data and experience into new knowledge, usually in the form of a mathematical model. Once it is created, this model can then be used to perform other tasks. This allows for the design of applications that would be too complex or time consuming to develop without computer assistance. For example, a machine learning system may be trained on millions of examples of labeled tumors in MRI images. On the basis of these examples, the system recognizes patterns of characteristics that constitute a tumor. This serves as a model that can then determine if tumors are present in new MRI images. These systems are often able to outperform experts.
Machine learning is a powerful tool that increasingly is incorporated into more computer applications. Its ubiquity makes it harder to spot AI applications that are not trained on data but that rely on human-written and readable rules and facts. Applications that use artificial intelligence but do not learn from or produce new results based on exposure to data are sometimes referred to as "good old-fashioned AI" or "GOFAI." And some are still in operation. For example, a simple chatbot may address questions solely by supplying pre-written answers that contain relevant keywords.
Finally, deep learning is a subset of machine learning. Deep learning uses machine learning algorithms but structures the algorithms in layers to create "artificial neural networks." These networks are modeled after the human brain and have been effective in many situations. Deep learning applications are most likely to provide an experience that feels like interacting with a real human.
Professor Yaser Abu-Mostafa's Machine Learning Course