The ideal form of generative AI allows its user to craft output according to their imagination. It necessarily has some defaults to fill in the gaps of unstate

The ideal form of generative AI allows its user to craft output according to their imagination. It necessarily has some defaults to fill in the gaps of unstated assumptions and these defaults may be shaped implicitly by initial training data or explicitly by later-stage product development but the tool loses a phenomenal amount of utility when crude approaches to "safety" get in the way of intuitive or everyday uses. Having its output default towards a contemporary, Western, cosmopolitan ideals about gender/race/etc diversity isn't unreasonable when contemporary, Western, cosmopolitan users are the most critical to this stage of the product growth. But most people around the globe don't actually live in that culture and so they quickly notice how the default is culturally myopic, and others are finding that in practice these are not being implemented as defaults so much as they're being implemented as output requirements that dramatically limit how useful the tool is. TLDR; In many cases, people can't get the product to make the specific, innocuous images they have in mind. The debate can sound political and certainly gets taken up by political activists, but it's actually rooted in quality decay. The tool is less useful when it does less, and people get annoyed by that.