(Honestly, this interviewer had made up his mind to do this for the heck of doing and reject this application, so he kept interrupting and reading out from Google). My answer: Machine Learning performs well on Smaller datasets, they are more prone to outliers and cannot be used in use cases where memory is needed to store data for training/validation. There is more human interference and we need to select features as DL works better on more enormous datasets, they can learn/ select features and also be used in LSTM/GRU/Transformers where memory is used.