I'm not sure i'm understanding you, or i'm not sure I agree. Humans, for instance, can learn from substantially less data than (most) machine learning models, i

I'm not sure i'm understanding you, or i'm not sure I agree. Humans, for instance, can learn from substantially less data than (most) machine learning models, if the data is presented in a way that we're optimized for (e.g. human facial recognition). If we can devise ML algorithms capable of generalizing from substantially smaller datasets, while data will still be relevant, it will be a less important bottleneck in the process than it is now.