Short answers, pulled from the story.
Arthur Samuel, an IBM employee and pioneer in computer gaming and artificial intelligence, coined the term machine learning in 1959. The synonym "self-teaching computers" was also in use during the same period.
The earliest machine learning program was introduced in the 1950s by Arthur Samuel. It calculated the probability of winning at checkers for each side of the game.
The three broad categories are supervised learning, where a program learns from labelled input-output pairs; unsupervised learning, where the algorithm finds structure in unlabelled data on its own; and reinforcement learning, where a program learns by receiving rewards as feedback while navigating a dynamic environment.
By 1980, expert systems had come to dominate AI and statistical approaches fell out of favour, creating a rift between AI and machine learning. Machine learning reorganised as its own field in the 1990s, shifting its goal from achieving artificial intelligence to solving practical problems using methods from statistics, fuzzy logic, and probability theory.
The black box problem refers to situations where the process by which a machine learning algorithm produces an output is entirely opaque, meaning even the engineers who built the system cannot audit the pattern it extracted from data or explain specific decisions it makes. The UK House of Lords Select Committee stated that an intelligence system with a substantial impact on an individual's life would not be acceptable unless it could provide a full and satisfactory explanation for its decisions.
In 1988, the UK's Commission for Racial Equality found that St. George's Medical School used a program trained on admissions data that had denied nearly 60 candidates because they were women or had non-European-sounding names. A ProPublica investigation found that a recidivism prediction algorithm flagged Black defendants as high risk twice as often as white defendants.