Since I’ve started learning and working in Data-Driven industry, I always tried to emphasis that ML is a subset of AI but doesn’t represent the whole picture/aim and capabilities of AI.
Although during lecture I didn’t feel that Andrew sees it that way, particularly the later case. Although he illustrates the ML under AI umbrella but doesn’t (yet) emphasis that ML can be “intelligent” and also cannot be " intelligent but just data extract machine". Although he connects the “extract knowledge” to DS.
To me, what’ve learned from my own working experiences and when I have to talk about it I always come with this definitions:
AI = it’s intelligent in a way to make decision (for control system), to see or to talk. doesn’t necessary need historical data (eg. RL) but uses logic.
ML = It doesn’t 't make any decision but uses the statistical relationship provided in data to suggest an accurate output.