Comment by pton_xd

If models had persistent memory or an evolving set of weights, you could easily construct an argument that they can experience "suffering" or other emotions which resultingly adjusts their personality.

The current implementation as stateless matrix multiplication... yeah there's nothing going on there.