I've been thinking about a semi-related problem to this idea....which is "how do you authenticate what the training sources used for an AI bot is?" Is there some way to publish that and know with cryptographic assurance what sources where used to train it?

Once thats possible....then it would make people more likely to trust these ideas. (ie. Imagine I had a bot trained on all writings from Founding Fathers + Supreme Court Case law -- you might be willing to trust that more than some newly minted commie from Commie U).