> ML model trained on this past data will likely learn that bias That's the opposite of what the author is saying, though - or rather, she's saying that data b

> ML model trained on this past data will likely learn that bias That's the opposite of what the author is saying, though - or rather, she's saying that data bias exists, but the algorithm itself introduces bias that would be there even if the data itself were somehow totally fair, for some unspecified definition of "fair".