Machine learning, deep-fat fryers, and community cultivation

Maciej Ceg?owski's (previously) speech at the Library of Congress, "Deep-Fried Data," describes the way that data begs to be analyzed and how machine learning is like a deep-fat fryer -- a fryer makes anything you put in it "kind of" delicious, and machine learning "kind of" finds insights in your data-set. But unless you know what your food is being fried in, you have no idea what's actually happening to it. And unless you know what data is used to train your machine-learning system, you can't know if it's finding real insight or just serving as a "money-laundry for bias." Ceg?owski does a great job on the structural limits and seductive appeal of machine learning, but then moves on to how archivists and librarian can use large data-sets, and the weird problems of providing data to strangers who use it in ways you may find disturbing or frivolous, or just inexplicable. From here, Ceg?owski talks about where data-sets to analyze can come from -- whether you can work with companies addicted to the surveillance business-model and keep your ethics intact -- and what good archiving practice should be in an era of dynamic documents served to rapidly obsoleted technologies.