his points resonate with my experiences dealing with big data analytics and machine learning efforts at all the corporations i've worked with in the last 8 years or so. so much time is spent on figuring out how to transport the data (large, large pipes) and/or massage the data (cleaning/sanitizing, enriching, categorizing, etc.) before it ever gets to the actual analytics.
it also resonates with me on the issue of getting good training data sets for learning algorithms. just as school curriculum and textbooks are carefully selected and curated, you have to do the same with things like ANNs.