Try to perceive questions related to machine learning in this perspective such as Data, Task, Model, Loss Functions, Learning Algorithms, and Evaluation.
The roles of human: [Data, Task, Model, Loss Functions]
As we have seen from the first week of the course the different steps involved in doing data science are to Collect, Store, Process, Describe and Model data. All these are or can only be decided by a human and not by a machine.
The stages such as collect, store, process and describe are essential to even see whether the data is compatible for a task or to even formulate a suitable task. Once the task is formulated by a human then again it is human who decides what model (whether statistical or machine learning or deep learning model) can be chosen for a particular task based on the experience of handling data. Same goes for choosing Loss Function.
Role of a machine: [Learning Algorithm]
But in the case of the learning algorithm, this is entirely up to the machine to learn the parameters associated with the features.
Think an example of say 1000 rows and 1000 columns or features to which you have chosen yourself to learn the parameters associated with the columns instead of machine learning algorithm learning it. What will happen? How will you relate between variables/features or say how will you say that these features have more weightage than other features and also they are related or unrelated? It will be a tedious task for a human right (that is to learn without the experience of handling and learning the weightage of the features). So, this is where machine learning algorithms come in rescue so that the complexity of human tackling such a task is reduced drastically. In this sense, the machine will learn here and all other tasks are handled by humans.
Summary: Machines (learns) are directed entirely by humans (dictates) where actions of a machine are the consequence of human knowledge.