At the moment we require people to train models themselves, this issue is about including trained models with Moodle.
Those trained models would come from Moodle HQ, and they would be trained with the anonymised data obtained from partners.
Each machine learning backend stores trained algorithms using a different format, for example, mlbackend_php stores a serialized php class that contains the weights and other stuff required by the ml algorithm to restore its previous state, mlbackend_python stores the tensorflow session.
Summary: We need to expand mlbackend API to load a trained model file and include those files in core.
We can not do this until we have more anonymised datasets to train the built-in models with.