From procedurally reactive ghosts in Pacman to IBM Watson’s Jeopardy champion, AI has had a broad set of definitions. John McCarthy coined "AI" in 1956 and has written a pretty comprehensive definition that's also a very approachable read, linked here. However, the fundamental definition is still broad from an applied AI perspective: "it is the science and engineering of making intelligent machines, especially
intelligent computer programs".

Today, with the sudden abundance of neural networks, it more or less has settled into meaning software that can learn. For now, we also lean on that definition because it paints a stark contrast to the static, difficult to change software and data science work of the past.