They both need to learn from a variety of datasets, so “machine learning” being a buzzword ten or fifteen years ago is probably indistinguishable from AI nowadays.
One concern with the AI training data is whether anybody gave permission to have their comments/artwork/literature/computer code etc., etc., used by anyone with no compensation before the AI lifted it from the World Wide Web and incorporated it to generate its own version of product.
The only difference, perhaps, is that AI seems to be willing to make up nonsense and fake citations when “machine learning” was trying to make it easier to analyze massive amounts of data in a speedy fashion.
I have taken part in the https://www.zooniverse.org/ where they ask “citizen scientists” to help sift through a vast amount of data, often with the end goal of teaching a computer to do it, once it has learned enough from the mere humans willing to look at, categorize, and analyze a variety of images. In all cases, they say that humans are better than computers at distinguishing subtle features, but eventually, with enough data, the computer can take over the task.
Some of the projects have included screening breast cancer biopsies, or cancer cells, or galaxies.
News 
Market Data 
Discover 
