As AI starts to become relevant in the free software space, how should we apply our existing ways of thinking about software freedom to this new class of programs?
Deep neural networks present a big shift to the traditional software model; not only we have to consider program code, but we also now have training data as a new component to fit into the picture.
With the help of an insider from the Debian Developer community, we’ll look at how existing distributions are approaching this problem, how more attention about training data can impact existing practice (sometimes in surprising ways), and the way the moral principle of freedom can be both helpful and puzzling for technical goals like redistribution, reproducibility, and security.
