Generative models can credit the sources from which the draw patterns to form their output.
We can create mechanisms that learn to trust certain sources of information more than others on specific topics based on specific criteria.
So instead of having everyone's voice be equally weighted in the output, someone with a history of predicting real world outcomes could be trusted and weighted higher when the model forms an output.
This can mean things like credit for art output by generative models. Obviously for example if you specify a certain style it is pulling from some artists more than others. This would mean that when an image is output it would say
This art draws primarily from Artist A (65%), Artist C (23%), Artist D (12%).
Likewise in a generative text model It can say: this output is sampling source A (13%), source B (27%), source C (70%)
And that distribution would represent a form of reputational trust in the model for who it values. It also would represent a type of credit.
If it was bound to a blockchain mechanism you could even distribute tokenized rewards based on these distributions.