Hello student
ML and AI are complementary, but they target different goals.

AI has a very specific purpose of building an intelligent agent that can take rational decisions under various different circumstances. By its very nature, the intelligent agent needs to interact with environment that were built for humans and hence must be capable of human-like capabilities i.e reading natural language text, understanding speech, recognizing different objects etc. The application focus is something like, a robot capable of automatically identifying and fixing issues that can occur in a nuclear reactor or a robot capable of self navigation, experimentation and reporting "interesting" observations during extra planetary missions etc. From these application, we also get some higher order AI problems. For example, a robot self-navigating a planet, needs to plan and chart its course from point A to point B or a robot that is attempting to fix a leak in a nuclear reactor needs to identify the set of task to perform so that the leak is fixed.

The goal of ML is to assimilate known knowledge in an abstract form called models, and be able to predict certain unknowns for instances that will be seen in the future. For example, in an email spam classification task, the goal is to learn a model from a known set of emails that have been labeled as spam or not spam and be able to predict whether the new email that we just received is spam or not spam. Classification is just one concrete instantiation of ML goal. Clustering can be thought of as predicting cluster variables for future instances and Regression can be thought of as predicting function values for points that haven't been observed so far etc.

AI with its very specific application focus can use tools built in ML to achieve its goals. However, ML serves a much broader class of problems.

With these in mind, what you want to start from is entirely upto you. For example, if you want to build planet-navigating robots then you probably need to start from AI, learn about the various interesting problems involved in AI, see what kind of ML tools you will need to learn, come to ML learn those tools and go back to AI. However, if your interest lies in building the ML tools that serve a fairly broad class of problems many of which are relevant in the real world today [1], then you may want to start learning ML first.

Note that, both ML and AI are vast fields with many things to learn. Even though I suggest switching from ML to AI and AI to ML, it is not as easy as it might sound :-). Unless you know what you want to learn and how you plan to use it, you might be aimlessly going into and learning things, that may or may not be relevant to what you might want to do. Not that it is wrong, but, as humans, we have very limited time and we need to use it as efficiently as possible.

[1] From a research perspective, researching in ML can get you funds. From a job perspective, many people are hiring folks who have reasonable understanding in ML and can use various tools like weka, libSVM etc.