Machine learning
An approach to building software where, instead of a person writing out explicit rules for the computer to follow, the program is shown a large number of examples and adjusts itself until it gets good at a task. The patterns it picks up are stored in its model.
Consider a translation example. The traditional approach is to hand-write rules ("the past tense adds this suffix, this word order becomes that"). The machine learning approach is instead to show a program many English–target sentence pairs and let it discover the patterns itself. The rule-based approach involves a lot of manual work and is brittle (there are always exceptions to the rules) but transparent and deterministic. The machine learning approach works well with messy, open-ended input but is probabilistic and harder to inspect.
"Machine learning" is the more precise name for most of what gets loosely marketed as "AI." It is also a data-hungry approach: it needs many examples to learn from, which is exactly why it is hard to apply to low-resource languages, where little digital data exists. Much of the difficulty, and the ethical tension, in applying these methods to Indigenous languages comes down to how that data is gathered, who controls it, and how it is used.