In brief, AI uses existing knowledge and/or heuristics (to solve problems that lack a closed-form solution), while ML acquires knowledge and heuristics toward the same end, with the added goal of improving performance as it learns and adapting to changing conditions.

Traditionally, AI has been divided into distinct subfields (e.g. search, planning, natural language and speech processing, game playing, computer vision, robotics, knowledge representation, expert systems, logic, and ML). Today, ML is employed in all AI subfields, but until recently, most subject matter in each AI subfield had been unrelated to ML. In the past decade especially, that's changed as deep learning and probabilistic methods have gained mindshare and now are largely unavoidable when tackling AI-related problems.

In general, AI's subfields have focused on identifying fundamental obstacles and important features in their own problem domain and developing appropriate techniques that operate on those features when solving problems (like using object recognition and localization to solve vision problems like autonomous driving). I suspect AI's past emphasis on feature engineering has faded as NN-based ML has risen.

It would seem that many people have divergent definitions but I’ve always learned that AI was Artificial intelligence broadly construed (goal seeking, planning, rule based expertise, logic etc) while machine learning is the specific subset that is (roughly speaking) statistical (neural nets, regressions, svm and the like).

Don’t have strong feelings about these definitions.