AI algorithms try to mimic some aspects of human biology or human behavior to solve problems that cannot be solved quickly using polynomial time algorithms. If the program's core is based on AI h, then you can loosely call 'em "AI programs".

You can think of AI as "imprecise computation". If you can improve precision over time, it becomes "machine learning" (this being the current buzzword of the field).

Note that I don't agree with simple rule-based systems being classified as AI, even with that naming being heavily used on games.