Why do we need vector embeddings in NLP?

Most ML algorithms can only take low-dimensional numerical data as inputs. Why? → Faster for computation. That means we must transform non-numeric variables (ex. items and users) into numbers and vectors. We could try to represent items by numerical...