At its core, AI is machine learning. All applications of AI that you hear about today use machine learning (ML) in some form.
Rules engines or expert systems are considered by some as AI, while others argue that they are not intelligent, as the outputs are clearly deterministic. It is simpler to think of all the AI systems in use today as the user-friendly manifestation of arcane ML models.
It will make life easier for everybody if we accept the base condition for any AI system to be the ability to learn without explicit external intervention. As long as there is an element of self-improvement baked into a system or application, it is appropriate to use the AI rubric. The learning itself can be supervised, unsupervised, semi-supervised, or reinforcement. But the model needs to explicitly allow for self-learning and improvement.
Learn the basics at the heart of the AI debate with Deconstructing AI