Machine learning is a methodology which provides us with a set of tools and techniques for statistical exploration of data. Its predictability depends on the quality of data available and the effectiveness of the algorithms which extracts patterns from the data to make decisions or predict outcomes. Software packages which also deal with decision boundaries or regressions usually have some sort of ML kernels at their core. If the processes are partially manual then the decisions might be subjective to a certain extent (semi-supervised) and can vary with respect to their reliability over a wide range of applications.

I think it is depending on the scale of training data set and type of training method. For a large number of training images with corresponding detailed annotations, deep learning methods can achieve better performance than traditional methods.