First steps towards what could be an ethnographic AI that has some cultural understanding of what it sees. (Use the menu at upper left to see the experiments.) Still some issues with bandwidth (currently I'm referencing remote machine learning models) and displaying video on mobile devices.
The environmental cost of AI concerns me. One way to minimise this is to use small, even personal datasets. For example, the open source community is working on models (e.g. MobileNets) that we can use individually, offline on a personal device. It was one of the original dreams of AI - that it would be a truly personal assistant, and not commercial exploitation masquerading as a personal service.