Fair Learning

AI training and copyright

Can AI be trained on protected works?

Freely allowed

Using copyrighted works to train machine learning should generally be fair use

What it says

Arguing that courts may wrongly treat non-transformative, whole-dataset copying as free riding, the authors propose 'fair learning': using works to train machine learning systems, which learn unprotectable patterns rather than expression, should generally be treated as fair use.

The influential US case for letting AI learn from works without permission.

Sources