The software we write and design can help or harm others. So it needs to be transparent, the code needs to be open and we try to design with those rights and freedoms from the beginning.
It's why I'm really concerned by the OSI diluting those ideas with the Open AI definition.
The training sets need to be accessible to everyone. You can't have an idea of why Open AI makes a decision, summarises a thing, recommends an action without those sets.