A dataset when you're talking about machine learning, is the input that you put into the algorithm, it's a vast array of information, it could be anything from pictures, it can be numbers, it can even be sound – that essentially becomes the world that the algorithm knows. Through constructing your own dataset: [you] start to understand how they can have biases; gaps; what's there; what's not there; how they might be labelled; what those labels might mean to a machine, but also to a human. And the amount of time, labour, effort that goes into constructing these things. When you get them off the shelf, when you download them from the internet, you don't really understand the process of ... constructing these things.
Anna Ridler, Algorithms in the Arts, Front Row, BBC Radio 4, Aug 2020
