I think we just haven't built complex enough architectures to allow this. I have a few ideas boiling that would help facilitate long term context understanding and recall.