The problem with AI is AI is actually really stupid. AI does not actually know what a Viking is, what a Pope is or even who Abraham Lincoln is. It has images in it's training data and from them it roughly knows that a Pope has a funny hat, and that vikings are burly dudes with swords or axes. And it also know it's supposed to be a person. That's about it. This is also why it easily gets wither laughably racist with it not being able to understand a black person is a person because the training data only has white people, or the opposite, making it "woke" because it has been trained to identify people of color as people, but does not know Vikings are supposed to be white.
It doesn't even know that. If we are talking image processing, it knows that certain pixels tend to be groups in certain ways in certain associations with other groups of pixels.
For example, it knows that "these groups of pixel configurations" are associated with "these key words", so if you write pope hat, it has examples of pixel groupings that are associated with that key word (and honestly, it doesn't know what a word is, its actually something called tokens, which are often pieces of words). Then there are other groups of pixels it associates with key words like heads or faces. It knows that these groups of pixels are arranged in a certain order (the hat pixels tend to go above the head pixels, for instance). It knows this because in the image examples, a group of pixels with the "human head" tag tends to be under the "pope hat" tag (as a simplified example).
Then through a mish mash of statistics, vector calculations, etc, it puts the pixels in the right places. Basically paint by numbers + a statistical blender mishmashing up various image examples. The tokens point to image tags and are then brought in, blended together, and then painted on to the image.
You can see this sort of thing and its lack of concept in terms of what these pixels represent if you look at AI hands. An AI has no idea what a finger is, it just knows that you blend pixels together from whatever examples associated to whatever key words (again, its not actually key words, but a series of tokens), so it blends a few examples together and sometimes you get a few extra fingers. AI don't care because the pixels are blended together appropriately per the algorithm.
In terms of whether a black person or white person is George Washington, I would suspect that the data tagging strategy (which is still done by people in most cases) and prompt strategy would need to be looked at.
What people refer to as "concepts" when it comes to AI are actually vectors, which are statistical weights between various data artifacts that can be influenced at different levels of the system (as the prompt level, at the algorithm level, it can be run through another AI for influencing, at the human interaction level, etc). Its, in effect, a giant n-dimensional lattice of associations (and would probably make for some neat art when visualized in 2 or 3 dimensions).
This philosophy concept explains what an AI is pretty well I think -
https://plato.stanford.edu/entries/chinese-room/
